WO2022124276A1 - Indoor air quality prediction method and indoor air quality detection system - Google Patents

Abstract

Provided are an indoor air quality prediction method and an indoor air quality detection system with which it is possible to accurately predict in advance a trend value for a future certain period of time by selecting an appropriate historical time-series change model without considering a specific single variable as the basis for correcting the trend of detection value changes. The indoor air quality prediction method according to the present invention comprises: selecting a historical time-series change model from among a large number of historical time-series change models obtained through learning in which a large number of historical time-series measurement results, including a detection place in a detection area or another detection place, are used as historical time-series learning values, the historical time-series change model selected satisfying a condition which is preset with respect to a set of a plurality of time-series detection values for an object of detection that are continuously detected in a predetermined period of time by a sensor unit provided in the detection area; and estimating a prediction value for a future certain period of time on the basis of the historical time-series change model selected.

Description

Indoor air quality prediction method and indoor air quality detection system

The present invention relates to the field of gas detection, and more specifically, an indoor air quality prediction method for predicting indoor air quality for a certain period from now on based on actually measured indoor air quality detection values and The present invention relates to an indoor air quality detection system using the prediction method.

Currently, detection (or quality evaluation) for indoor air quality detects the concentration of each component in the indoor air, and is based on the detected detection value (concentration value) and a preset threshold value. The operation of air purifiers (air conditioners, air purifiers, etc.) installed in the room is controlled so that the concentration of each component in the room air is controlled within an appropriate range.

However, it is known that it takes a certain amount of time to detect an accurate concentration due to the characteristics of the sensor and the diffusion characteristics of the gas, and there is a clear hysteresis. Taking carbon dioxide as an example, if the concentration of carbon dioxide in the room air exceeds a preset threshold and then the exhaust fan is turned on or the exhaust mode of the air conditioner is activated, the concentration of carbon dioxide When it is detected that a preset threshold has been exceeded, people in the room are not only already in an environment where carbon dioxide has exceeded the standard, but also carbon dioxide is appropriate from turning on the exhaust fan or activating the exhaust mode. It takes a certain amount of time to drop to a certain range, and during that period, people continue to feel uncomfortable in such an environment.

Therefore, many methods for predicting the concentration of indoor air in advance have been proposed.

For example, in Patent Document 1 (Japanese Unexamined Patent Publication No. 2003-090819), the time derivative value A (t) of the sensor output Y (t) in the gas sensor element 1 (that is, the amount of change of the sensor output value with time) and A. Using the time derivative B (t) of (t) (that is, the amount of change of the above change amount with time), the sensor output for a certain period from now on is predicted, and the specific gas concentration is detected. Quantitative methods, substance concentration detectors and recording media have been proposed.

Further, for example, Patent Document 2 (Japanese Patent No. 3830846) proposes a gas concentration prediction method and a gas detection method when the gas sensor is activated, and here, in a predetermined period before the output of the gas sensor stabilizes. The gas concentration of the detected gas is predicted based on the rate of change of the output value of the gas sensor (that is, the time derivative value of the sensor output) in a predetermined period after the gas sensor is activated.

In both Patent Document 1 and Patent Document 2, based on the change of the actual data one by one, the trend line of the data change is fitted using the time derivative value of the sensor output, and the sensor output for a certain period from now on is obtained. Predict (or calculate).

However, this inevitably leads to a large amount of calculation, and such a large amount of calculation is required for each detection, which puts a heavy burden on the arithmetic unit installed in the detection device.

Currently, with the continuous development of AI technology and image recognition technology using neural networks, the method of predicting the concentration of indoor air is constantly being improved and optimized.

For example, in Patent Document 3 (Chinese Patent Application Publication No. 1094426995), a step of collecting a current indoor image and current indoor carbon dioxide concentration data, and the current indoor image and / or the current present. The step of recognizing the current number of people in the room based on the concentration data of carbon dioxide in the room, and the amount of temperature change after the first predetermined period based on the concentration data of carbon dioxide in the current room and the current number of people in the room. An air conditioning and new wind system based on the number of people in the room, including a step of predicting the amount of change in carbon dioxide concentration and a step of adjusting the air conditioning and / or the new wind system based on the amount of change in temperature and the amount of change in carbon dioxide. Predictive control methods and systems have been proposed.

Further, for example, in Patent Document 4 (Chinese Patent Application Publication No. 109812938), a data real-time data collection step of collecting data on indoor air quality and number of people in real time via a data collection node was collected. Introduce indoor air quality data into the air quality pre-analysis model, provide further learning training to the model, and change the indoor air quality data together with the change status of the indoor number data and the current indoor air quality data. A neural network-based air purification method and system is proposed, including a data pre-analysis step to analyze trends and a pre-purification step to pre-purify indoor air quality based on the data analysis results of the air quality pre-analysis model. ing.

In both Patent Document 3 and Patent Document 4, the number of people in the room is collected from the indoor image, the number of people in the room is accurately recognized, and the number of people in the room is taken into consideration, and a single detection target in the room air is detected. It is necessary to predict or analyze the change tendency of the detected value.
However, while image sensors and recognition methods that accurately recognize the number of people in a room inevitably lead to a significant increase in the cost of detection equipment, changes in the concentration of indoor air and the number of people in the room are not the only trend-corresponding relationships. Environmental factors such as the difference in CO ₂ emissions per person, the size of the area of the indoor space, whether or not there is green plant in the room, whether or not the window is opened / whether or not the door is opened, etc. At this time, the detection value of a single detection target (for example, CO ₂ concentration) in the room air when the number of people in the room is large is conversely the single detection target in the room air when the number of people in the room is small. It may be smaller than the detected value (for example, CO ₂ concentration). In this case, if the number of people in the room (single factor) is used as the basis for correcting the change tendency of the detected value of the indoor air as in Patent Documents 3 and 4, conversely, the prediction and analysis of the change tendency of the detected value can be performed. It will be inaccurate.

Therefore, as a basis for correcting the change tendency of the detected value, it is not possible to predict the tendency value for a certain period in advance and accurately without considering a specific single factor (for example, the number of people in the room). , Is a technical issue that needs to be resolved immediately.

The present invention has been made to solve the above technical problems, and an appropriate historical time-series change model is selected without considering a specific single variable as a basis for correcting the change tendency of the detected value. It is an object of the present invention to provide an indoor air quality prediction method and an indoor air quality detection system using the prediction method, which can predict a tendency value in a certain period from now on in advance and accurately.

In order to solve the above technical problem, the first aspect of the present invention is obtained by learning a large number of historical time-series actual measurement results including the detection location of the detection area (S) or other detection locations as historical time-series learning values. It is preset between a set of a plurality of time-series detection values to be detected continuously detected by a sensor unit provided in the detection area in one predetermined period from a large number of historical time-series change models. Provided is an indoor air quality prediction method characterized in that a historical time-series change model satisfying the above conditions is selected, and a predicted value for a certain period from now on is estimated based on the selected historical time-series change model. do.

According to the prediction method according to the first aspect of the present invention, a large amount of historical data is collected in the first period, an appropriate learning algorithm is selected, and as much induction bias is set as possible, and these historical data are used as a training sample or a training sample. By learning as a training sample case, many historical time-series change models are generated, and a historical time-series change model that matches the above-mentioned multiple detection values based on multiple detection values detected in a specific detection area is generated. It is selected and the predicted value for the future fixed period of the specific detection area is estimated based on the selected historical time series change model. As a result, when the number of people in the detection area is not detected or is unknown, it is only necessary to acquire the detection value of a small part of the detection target in the indoor air, and the detection target in the specific detection area will be constant in the future. It is possible to predict the change tendency (change value) during the period of.

Furthermore, since it is not necessary to use an expensive image sensor or a complicated recognition method, the equipment cost of the indoor air quality detection system can be reduced.

The indoor air quality prediction method according to the second aspect of the present invention is the set when there are a plurality of historical time series change models satisfying the preset conditions in the indoor air quality prediction method according to the first aspect of the present invention. It is characterized by selecting a historical time-series change model that has the highest degree of matching with.

According to the prediction method according to the second aspect of the present invention, the historical time series change model in which the degree of matching between the plurality of historical time series change models satisfying the preset conditions and the current time series detection value is the highest. By selecting, the prediction accuracy can be improved as much as possible.

The indoor air quality prediction method according to the third aspect of the present invention is the indoor air quality prediction method according to the first aspect of the present invention. Until we find a historical time series change model that meets the set conditions, we reselect a new set of multiple time series detection values for the next predetermined period whose start time and / or length is different from the one predetermined period. It is characterized in that it continues to determine whether or not there is a historical time series change model that satisfies the preset condition with the new set.

According to the prediction method according to the third aspect of the present invention, by reselecting a new set of time series detection values, an appropriate historical time series change model is found, and it is unpredictable due to the scarcity of local data. Can be avoided.

The indoor air quality prediction method according to the fourth aspect of the present invention is the time until the time when the historical time series change model satisfying the preset condition is found in the indoor air quality prediction method according to the third aspect of the present invention. It is characterized in that it is generated as a new historical time-series change model by learning all the time-series detection values before the time.

According to the prediction method according to the fourth aspect of the present invention, it is possible to effectively utilize the detected value of the detection target that has already been detected and increase the number of models in the database. It also reduces the amount of calculation at the next system startup.

The indoor air quality prediction method according to the fifth aspect of the present invention is the indoor air quality prediction method according to the fourth aspect of the present invention. It is characterized in that the new historical time-series change model is updated by continuously inputting to the new historical time-series change model.

According to the prediction method according to the fifth aspect of the present invention, the adaptability (generalization ability) of the new historical time series change model to a new data sample can be constantly enhanced.

The indoor air quality prediction method according to the sixth aspect of the present invention is predetermined in the indoor air quality prediction method according to the first aspect of the present invention when the predicted value exceeds a preset threshold value. It is characterized in that it notifies the possibility of occurrence of the above case by a method, notifies that the corresponding measures are taken, or automatically controls to take the corresponding measures.

According to the prediction method according to the sixth aspect of the present invention, the actual detection value in the detection area is set to the preset threshold value (that is, the person in the detection area) by comparing the predicted value with the preset threshold value. Before exceeding (being in an unsafe air environment), be notified in advance (whether to notify the possibility of occurrence in the above case or to take corresponding measures), or to take corresponding measures. Since it can be controlled automatically in advance, a safe environment can be ensured. In addition, since it is possible to notify in advance whether or not there is a possibility of exceeding a preset threshold value, time for first aid is secured.

The indoor air quality prediction method according to the seventh aspect of the present invention is the indoor air quality prediction method according to the sixth aspect of the present invention, and the predetermined method is a buzzer, an alarm, an image display facility, and an indicator lamp. It is characterized by including issuing a notification and / or receiving a signal and issuing a notification via a remote controller, a mobile phone, or a PC side.

The indoor air quality prediction method according to the eighth aspect of the present invention is the indoor air quality prediction method according to the sixth aspect of the present invention. Or urge to control the number of people in the detection area and / or urge them to leave the detection area and / or manually turn on the air treatment device or force the air treatment device to be automatic. It is characterized by including encouraging people to turn it on.

According to the seventh aspect and the eighth aspect of the present invention, the person concerned (person in the detection area, person in the monitoring center, etc.) can be notified by one or more predetermined methods. It is possible to call attention to more efficiently. In addition to simply notifying the possibility of occurrence of the above case (the person in the detection area is in an unsafe air environment), by specifically instructing the persons concerned to take the corresponding action, the above case Can be avoided more effectively. In addition, by forcibly and automatically turning on the air treatment device, it is possible to prevent the actual occurrence in the above case.

The indoor air quality prediction method according to the ninth aspect of the present invention predicts that the predicted value after a certain period will fall below a preset threshold value in the indoor air quality prediction method according to the sixth aspect of the present invention. Notifies the person concerned or related equipment of the possibility of the above case, or notifies the person concerned or related equipment to reduce the operating load of the air treatment device or stop the operation of the air treatment device. It is characterized by that.

According to the prediction method according to the ninth aspect of the present invention, when the predicted value is predicted to be lower than the preset threshold value, the operating load of the air treatment device is reduced in a timely manner or the operation of the air treatment device is performed. By stopping, it is possible to avoid wasting unnecessary power.
In the indoor air quality prediction method according to the tenth aspect of the present invention, in the indoor air quality prediction method according to the sixth aspect of the present invention, the predicted value is still preset after a predetermined time from the start of the initial notification. When the threshold value is exceeded, the air treatment device is forcibly started or the performance of the air treatment device is forcibly adjusted.

According to the prediction method according to the tenth aspect of the present invention, if the predicted value still exceeds the preset threshold value after a predetermined time from the start of the initial notification, is the air treatment device forcibly started? Alternatively, in order to force the performance of the air treatment device, if the corresponding measures to be notified do not work, it can be forced to intervene automatically, thereby the actual in the above case. The occurrence can be prevented more reliably.

The indoor air quality prediction method according to the eleventh aspect of the present invention is the time series change model selected in the indoor air quality prediction method according to the first aspect of the present invention at a plurality of times in the one predetermined period. Using one historical time-series learning value immediately after a plurality of historical time-series learning values corresponding to the series detection values as the predicted value of the next unit time, the predicted value of the next unit time and the sensor unit detect it. The error is calculated from the detected value of the next unit time, and when the error becomes equal to or less than the allowable value, the subsequent prediction is continued using the selected historical time series change model.

The indoor air quality prediction method according to the twelfth aspect of the present invention reselects the historical time-series change model when the error becomes larger than the permissible value in the indoor air quality prediction method according to the tenth aspect of the present invention. , Characterized by that.

According to the prediction method according to the eleventh aspect and the twelfth aspect of the present invention, the prediction accuracy of the model can be evaluated by comparing the predicted value with the detected value of the next unit time. Also, by calculating the error between the predicted value and the detected value for the next unit time, the current model can be continued to be used or a more accurate model can be reselected based on the magnitude of the error. You can decide whether to do it. This makes it possible to improve the prediction accuracy. Compared to the method of reselecting the matching model in real time, the amount of calculation is simplified.

The indoor air quality prediction method according to the thirteenth aspect of the present invention is the one predetermined in the historical time series change model selected in the indoor air quality prediction method according to the first aspect or the eleventh aspect of the present invention. Using the multiple historical time-series learning values immediately after the multiple historical time-series learning values corresponding to the multiple time-series detection values of the period as the predicted values for the next predetermined period, the sensor unit is continuous in the next predetermined period. It is determined whether the data set error between the set of the plurality of time-series detected values detected in the process and the set of the predicted values for the next predetermined period is within the allowable error range, and the data set error is the allowable value. When it is within the range, it is characterized in that it continues to make subsequent predictions using the selected historical time series change model.

The indoor air quality prediction method according to the fourteenth aspect of the present invention is based on a set consisting of a plurality of time-series detection values in the following predetermined period in the indoor air quality prediction method according to the thirteenth aspect of the present invention. Or, based on a set consisting of a plurality of time-series detection values of the one predetermined period and a plurality of time-series detection values of the next predetermined period, the selected historical time-series change model is modified and modified. Is independently generated, and subsequent predictions are made using the modified model.

The indoor air quality prediction method according to the fifteenth aspect of the present invention is a historical time series change model when the data set error is out of the permissible range in the indoor air quality prediction method according to the thirteenth aspect of the present invention. It is characterized by reselecting.

According to the prediction method according to the thirteenth to fifteenth aspects of the present invention, the prediction of the selected historical time series change model is performed rather than using only the method of comparing one detected value with one predicted value. Accuracy can be evaluated in terms of overall trend of change, avoiding unnecessary model exchanges due to individual singularities.

The indoor air quality prediction method according to the sixteenth aspect of the present invention is the indoor air quality prediction method according to any one of the first to the twelfth aspects of the present invention, and the detection target is carbon dioxide and carbon dioxide. It is characterized by having one or more of gas components other than carbon, particulate matter, temperature, and humidity.

According to the prediction method according to the sixteenth aspect of the present invention, it can be appropriately applied to the prediction of carbon dioxide, but it can also be applied to the prediction of other detection targets.

On the other hand, the present invention has at least one detection terminal having one or more sensor units provided in the detection area, and a plurality of time series of detection targets continuously detected by the sensor units in one predetermined period. By learning a data receiving module that receives the detected value and a large number of historical time-series actual measurement results including the detection location of the detection area or other detection locations as the historical time-series learning value, many historical time-series change models can be obtained. By selecting a historical time-series change model that satisfies a preset condition between the generated data analysis module and a set consisting of a plurality of the historical time-series detection values from many of the historical time-series change models. Provided is an indoor air quality detection system, including a selective prediction module that estimates predicted values over a period of time.

The indoor air quality detection system according to the present invention utilizes a plurality of time-series detection values (detection values of a plurality of times within one unit time) of the current time detected by the sensor unit of the detection terminal. By being able to predict the predicted value of the detection target in the indoor air of the detection area for a certain period from now on, it is possible to predict in advance the air quality in the room for a certain period from now on.

The indoor air quality detection system according to the present invention includes a notification module for issuing a notification command to a notification unit when the predicted value exceeds a preset threshold value or falls below the preset threshold value. It is characterized by that.

According to the indoor air quality detection system according to the present invention, the notification unit can call the attention of the persons concerned (people in the detection area, people in the monitoring center, etc.). In addition to the possibility of occurrence in the above case (that is, the person in the detection area is in an unsafe air environment), the content of the notification may be specifically instructed to take an action corresponding to the persons concerned. , Thereby, the actual occurrence in the above case can be avoided more effectively.

In the indoor air quality detection system according to the present invention, when the predicted value exceeds a preset threshold value, the air treatment device is started, the performance of the air treatment device is adjusted, or the predicted value is preset. It is characterized by including a control unit forcibly starting the air treatment device or forcibly adjusting the performance of the air treatment device after a certain period of time has passed since the threshold value was exceeded. Further, when the predicted value after a certain period is predicted to be lower than the preset threshold value, the control unit reduces the operating load of the air treatment device or stops the operation of the air treatment device.

According to the indoor air quality detection system of the present invention, the actual occurrence in the above case can be prevented by automatically turning on the air treatment device in the control unit. Further, if it is predicted that the predicted value after a certain period will be lower than the preset threshold value after the above processing, it is unnecessary by reducing the operating load of the air treatment device or stopping the operation of the air treatment device. You can avoid wasting power.

In the indoor air quality detection system according to the present invention, the control unit is a smart gateway.

The indoor air quality detection system according to the present invention is characterized in that the control unit is attached to a monitoring center provided outside the detection area.

It is a schematic diagram which shows the 1st example of the system architecture diagram of the indoor air quality detection system which concerns on this invention. It is a schematic diagram which shows the 2nd example of the system architecture diagram of the indoor air quality detection system which concerns on this invention. It is a schematic diagram which shows the 3rd example of the system architecture diagram of the indoor air quality detection system which concerns on this invention. It is a schematic diagram which shows the 4th example of the system architecture diagram of the indoor air quality detection system which concerns on this invention. It is a flowchart explaining the 1st Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the modification 1 of the 1st Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the modification 2 of the 1st Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the modification 3 of the 1st Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the 2nd Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the modification 1 of the 2nd Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the modification 2 of the 2nd Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the 3rd Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the modification 1 of the 3rd Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the modification 2 of the 3rd Embodiment of the indoor air quality prediction method which concerns on this invention. It is a flowchart explaining the modification 3 of the 3rd Embodiment of the indoor air quality prediction method which concerns on this invention. It is a graph which shows the time-series change of the CO ₂ concentration in the case of a different number of people in the same detection area.

Hereinafter, the indoor air quality detection systems 100, 100A, 100B and 100C according to the present invention will be described with reference to FIGS. 1 to 4. In the following description of these indoor air quality detection systems, the same or similar components are designated by the same or similar reference numerals, and the description thereof will be omitted as appropriate. Further, the indoor air quality detection system according to the present invention is not limited to these cases, and can be appropriately modified according to actual system needs.

(Indoor air quality detection system 100)
First, with reference to FIG. 1, the entire system of the indoor air quality detection system 100 according to the present invention will be described.

As shown in FIG. 1, the indoor air quality detection system 100 according to the present invention includes a data reception module 120, a data analysis module 130, a selection prediction module 140, a routing device 150, a smart gateway 160, and a notification module 170. And at least one detection terminal 110. Next, each component of the indoor air quality detection system 100 will be described in detail.

The detection terminal 110 is, for example, a module provided in a detection area S such as an office area or a conference room, and is one or more sensor units for detecting indoor air quality in one or more detection areas S. Includes 111. More specifically, each of the sensor units 111 is provided with one or more sensor elements (not shown) for detecting one or more detection targets in the room air. As a result, each detection terminal 110 is a target for (one or more) detection of indoor air in one or more detection areas S by the one or more sensor units 111 (and one or more sensor elements included therein). Get the detected value.

More specifically, the detection target that can be detected by each sensor unit 111 (each sensor element) is, for example, a gas such as carbon dioxide, volatile organic substances, nitrogen oxides, sulfur oxides, ozone, carbon monoxide, and formaldehyde. It may be one or more of the components, or it may be one or more of the particulate matter contents such as PM2.5, PM10, or one or more of the physical parameters such as temperature and humidity. But it may be.

Regarding the transmission of the detected value, the detection terminal 110 has a plurality of processing methods. For example, the detection terminal 110 indicates one detection value of one or more detection targets (that is, when there is one detection target, the detection target indicates a detection value at a certain time, and when there are a plurality of detection targets, the detection terminal 110 indicates a detection value. When each detection value of a plurality of detection targets at the same time is detected), the detection value (and equipment ID) is directly transmitted (specifically, via the Internet by the routing device 150 described later). It is transmitted to the data receiving module 120 described later), but the present invention is not limited to this. For example, the detection terminal 110 may have a temporary cache function, and the detection terminal 110 accumulates a specific number of detection values (for example, the specific number is 10) of one or more detection targets. After the detection, the detection terminal 110 simultaneously transmits the above-mentioned specific number of detection values (and equipment ID).

In this example, one or more detection values of each detection target detected by each detection terminal 110 (sensor unit 111) are transmitted to the data reception module 120 described later via the routing device 150. However, the transmission method of the detected value is not limited to this. Depending on the actual architectural situation and performance of the system, it is not necessary to provide the routing device 150, that is, for example, when the smart gateway 160 described later has a communication module such as 2G or NB-IoT, the smart gateway 160 detects these. The values can be received directly, and then the smart gateway 160 transmits these detected values to the data receiving module 120 described later. However, in this example, it will be further described based on the fact that the routing device 150 is provided.

The data receiving module 120 is connected to the routing device 150 via the Internet, receives one or more detection values of each of the detection targets transmitted from the routing device 150 via the Internet, and then receives the detection values. One or more detected values are transmitted to the data analysis module 130 and the selection prediction module 140, which will be described later, respectively, to perform subsequent model matching, selection, and prediction work. In this indoor air quality detection system 100, the data receiving module 120 is, for example, an independent cloud module. Further, in the indoor air quality detection system 100, the data receiving module 120 has a plurality of predicted values (prediction) of one or more detection targets estimated by the selection prediction module 140 described later at a specific time or a specific period in the future. After receiving the data), it is determined whether the predicted value of the specific time or the predicted value of the plurality of predicted values of the specific period exceeds the preset threshold value or falls below the preset threshold value. At the same time, when the predicted value exceeds the preset threshold value or falls below the preset threshold value, the case is transmitted to the notification module 170 described later. Predicted values, preset thresholds and notification module 170 will be described in detail below. The data receiving module 120 may further have a historical data storage function, and by storing all the input detected values, the data receiving module 120 is used for reselection of the historical time series change model, generation of a new model, and the like. The "history data" here includes not only the historical time-series actual measurement results including the detection location of the detection area S or other detection locations, but also each of the continuously detected by the sensor unit 111 in one predetermined period. It may include one or more time-series detection values to be detected. As a result, modules that realize predictive functions can be integrated into a single cloud, communication methods are simple, communication is fast, and problems such as data packet loss and loss due to data transmission can be reduced. can. Reselection of the historical time series change model, generation of a new model, etc. will be described in detail later.

The data analysis module 130 receives one or more detected values from the data receiving module 120 by being connected to the data receiving module 120 via the Internet. In the indoor air quality detection system 100, the data analysis module 130 is also, for example, an independent cloud module. Further, the data analysis module 130 is an analysis module having a machine learning function, and has a large number of historical time-series measured values of each detection target detected in the detection area S and / or other detection areas other than the detection area S in the past. By making the set of training sample datasets or training sample case (ie, labeled training samples) datasets, and training the above datasets with appropriate learning algorithms, depending on the actual situation, many Generate (build) a trained model. The above learning algorithms include neural network algorithms (including convolutional neural network algorithms, circular neural network algorithms, etc.), regression learning algorithms (including linear regression analysis algorithms, logarithmic probability regression algorithms, multi-category learning algorithms, etc.), support vectors. Includes machine algorithms, decision tree algorithms, Bayesian classifiers, cluster analysis algorithms, etc. Based on the characteristics of each algorithm, the presence or absence of labels, actual needs, etc., the above learning algorithms can be used for supervised learning, semi-supervised learning, and unsupervised learning. Further, in the present invention, considering that the data is time-series, the trained model is also called a historical time-series change model, and has a fitting curve, a neural network structure (the structure itself, a weighting coefficient between each node, and a weight coefficient). It must be specifically explained that it includes (including activation function) and so on. Regarding the creation of the historical time series change model, taking the training sample case as an example, when the historical time series actual measurement value with the equipment ID label is input to the data analysis module 130, the data analysis module 130 first becomes the equipment ID. Based on this, the historical time-series measured values are classified and learned. Specifically, for example, when there are a plurality of detection areas such as a detection area S, a detection area P, and a detection area Q, by recognizing the equipment ID, the data analysis module 130 obtains each historical time-series measured value. It is classified into a historical time-series actual measurement value of the detection area S, a historical time-series actual measurement value of the detection area P, a historical time-series actual measurement value of the detection area Q, and a historical time-series detection value of another detection area. Then, after the classification learning is completed, the data analysis module 130 trains a set of historical time-series measured values (labeled) of the detection area for the historical time-series measured values (labeled) for each detection area. The data set is trained as, for example, {DT0, _{DT0 + 1} , _{DT0 + 2} , ..., D _TE , ..., y _s } (where _T0 indicates the start time of a certain detection period, where TE is the said. Indicates the end time of the detection period, where y _s indicates the equipment ID in the detection area S). The above data set { _DT0 , _{DT0 + 1} , _{DT0 + 2} , ..., _DTE , ..., y _s } may be a set of all historical time-series measured values detected in the detection area S, and may be a detection area. It may be a set of historical time-series measured values in one past detection period detected in S, or may be a set of historical time-series measured values in a plurality of past detection periods detected in the detection area S. The set { _DT0 , _{DT0 + 1} , _{DT0 + 2} , ..., D _TE , ..., y _s } indicates a set of historical time-series measured values within one past detection period detected in the detection area S. Or, when showing a set of historical time-series measured values detected in the detection area S within a plurality of past detection periods, there are a plurality of the above sets, and the same learning algorithm and regression bias are applied to each set. Based on (inductive basis), one corresponding historical time series change model can be generated. On the other hand, the set { _DT0 , _{DT0 + 1} , _{DT0 + 2} , ..., D _TE , ..., y _s } indicates a set of all historical time-series measured values detected in the detection area S, and is shown in the learning process. , Even if the same learning algorithm is used, a plurality of different historical time series change models (trained models) will be generated based on different induction biases. For example, if the learning algorithm is a regression learning algorithm, the "Occam's razor" principle will generate the smoothest historical time series change model, while other principles will generate other historical time series. As the change models are generated, the smoothness of these historical time series change models is not as good as the historical time series change models generated based on the "Occam's razor" principle, but the generalization ability is better. It may be. More specifically, when the detection area S is a conference room, different induction biases can be set, for example, "the overall change tendency of the detection target is stable (for example, the number of people in the conference room is stable). The bias that "the overall tendency of CO ₂ to change is stable and there is no sudden change because it does not change constantly or hardly", "the detection target suddenly changes suddenly (for example, the meeting was originally held in the conference room) The number of people is 5, and 10 people suddenly came in during the meeting.) ”Can be set.

As an example of the historical time-series change model, FIG. 16 shows a time-series change graph of CO ₂ concentration fitting in the case of different numbers of people in the same detection area. Here, each measured curve can be regarded as one historical time-series change model. More specifically, three meetings were held in the same detection area, for example, a meeting room, but the number of people varies from meeting to meeting. Further, for each conference, the sensor unit 111 of the detection terminal 110 detects a time-series change value of carbon dioxide ppm in the conference room, and thereby obtains a plurality of actually measured values of carbon dioxide ppm that change with time. Obtained. Based on the above-mentioned plurality of time-series measured values of carbon dioxide ppm, one fitting curve is generated as one historical time-series change model by using regression analysis.

The selection prediction module 140 receives one or more detection values of each detection target from the data reception module 120, and the selection prediction module 140 is based on a set of one or more detection values of each detection target. , From many historical time series change models generated by the data analysis module 130, select a historical time series change model that satisfies the conditions preset with the above set, and based on the selected historical time series change model. Estimate the predicted value of each detection target for a certain period in the future or at a specific time in the future. Further, after the predicted value of each detection target for a certain period from now on or at a specific time from now on is estimated, the selection prediction module 140 transmits the predicted value to the data receiving module 120.

The notification module 170 is a module that notifies when the predicted value exceeds or falls below a preset threshold value, and in the indoor air quality detection system 100, the notification module 170 is also one. It is an independent cloud module. Specifically, after receiving one or more predicted values from the selection prediction module 140, the data receiving module 120 compares these predicted values with a preset threshold value, and any of these predicted values. When the predicted value (which is the predicted value when there is one predicted value) exceeds a preset threshold value or falls below a preset threshold value, the data receiving module 120 "is" The information that the predicted value exceeds the preset threshold value or falls below the preset threshold value is transmitted to the notification module 170. Upon receiving the above information, the notification module 170 informs the corresponding equipment (for example, the smart gateway 160 described later and the monitoring center 180 or the mobile phone APP 190 described later as a preferred example) "at a certain time or a specific time in the future". Send a notification that the predicted value exceeds or falls below a preset threshold, and take the corresponding steps (or instruct the corresponding device to perform automatic control). Encourage or urge stakeholders to do so.

The smart gateway 160 receives the notification from the notification module 170 by being connected to the notification module 170 via the Internet. Upon receiving the notification from the notification module 170 that the predicted value exceeds the preset threshold value or falls below the preset threshold value, the smart gateway 160 automatically controls the air treatment device 200 described later. conduct. Specifically, upon receiving the above notification, the smart gateway 160 automatically starts the air treatment device 200 or automatically adjusts the performance of the air treatment device 200 (for example, the detection target is CO ₂ ). In the case of ppm, when the predicted value of ppm becomes larger than the preset threshold value, the smart gateway 160 automatically starts the air treatment device 200, increases the air volume of the air treatment device 200, or increases the air volume of the air treatment device 200. When the predicted value of ppm drops from the dangerous concentration value to a preset threshold value (safe concentration value), the operation of the air treatment device 200 is stopped or the operating load of the air treatment device 200 is lowered).

Further, as a preferred example, the smart gateway 160 further has a notification unit (not shown). Specifically, when the smart gateway 160 has a notification unit, the smart gateway 160 performs automatic control and at the same time emits a notification signal to a person in the detection area S via the notification unit. The notification unit may be a notification member provided in the detection area S such as a buzzer, an alarm, an image display facility (eg, an LED panel, a television monitor, etc.), a remote controller, an indicator lamp, or the like.

Further, as a preferred example, the indoor air quality detection system 100 may further include a mobile phone APP 190 installed in a monitoring center 180 and / or a smart mobile phone provided outside the detection area S. In this case, if it is determined that the predicted value exceeds the preset threshold value or falls below the preset threshold value, the notification module 170 selects the information for the smart gateway 160, the monitoring center 180, and the mobile phone APP 190. It can be sent uniformly. After any one of the smart gateway 160, the monitoring center 180 and the mobile phone APP 190 receives the above information, the air treatment device 200 is automatically started / stopped or the performance of the air treatment device 200 is adjusted automatically. Take control.

On the other hand, when the system 100 includes the monitoring center 180 and / or the mobile phone APP 190, the information that "the predicted value exceeds the preset threshold value or falls below the preset threshold value" is transmitted to the monitoring center 180. Then, the monitoring center 180 automatically controls and at the same time issues a notification to the related stap via the notification unit provided. The notification unit of the monitoring center 180 may be a notification member such as a buzzer, an alarm, an image display device (for example, an LED panel, a television monitor, etc.), a remote controller, an indicator lamp, or a PC. When the above information is transmitted to the mobile phone APP 190, the mobile phone APP 190 gives an automatic control notification and issues a notification to the mobile phone owner via the notification module provided in the mobile phone.

As the notification method of the above notification unit, for example, a buzzer sound, an alarm of an alarm, blinking of the liquid crystal screen of the remote control, blinking of a light, vibration, etc. may be used, or an APP notification of a mobile phone, a message of a mobile phone, or a mobile phone. It may be a way chat notification of a telephone, a vibration of a mobile phone, or an email on the PC side, a notification of monitoring software installed on the PC side, or the like.

In the above description, the smart gateway 160, the monitoring center 180, or the mobile phone APP 190 automatically receives the information that "the predicted value exceeds the preset threshold value or falls below the preset threshold value". Control and send notifications to related parties. However, instead, after the smart gateway 160, the monitoring center 180 or the mobile phone APP 190 receives the above information, the relevant parties are notified and manual measures are taken without automatic control. For example, it is recommended that the parties concerned open the doors and windows by themselves, or that the number of people in the detection area S is controlled, or that the air treatment device 200 is manually operated. It is recommended to turn it on or to recommend the parties concerned to move away from the detection area S.

Furthermore, it must be explained that in the indoor air quality detection system 100, the data reception module 120, the data analysis module 130, the selection prediction module 140, and the notification module 170 are commonly arranged in one cloud. However, the arrangement form of the above module is not limited to this, and may belong to a plurality of different clouds, which will be described in detail later.

The "air treatment device 200" referred to in the present invention includes an air conditioner, a ventilation device, a dehumidifier, a humidifier, an air purifier, and the like.

(Indoor air quality detection system 100A)
Next, the indoor air quality detection system 100A according to the present invention will be described with reference to FIG.

Compared to the indoor air quality detection system 100, the difference of the indoor air quality detection system 100A is that the notification module 170 has a function of determining whether the predicted value exceeds the preset threshold value or falls below the preset threshold value. To be executed by.

Specifically, in the indoor air quality detection system 100A, the notification module 170 is directly connected to the selection prediction module 140 via the Internet. In this case, the selection prediction module 140 directly transmits the prediction value to the notification module 170 after estimating the prediction value of the detection target for a certain period from now on or at a specific time from now on. After receiving the predicted value, the notification module 170 compares these predicted values with a preset threshold value, and any one of these predicted values (when the predicted value is one, it is the predicted value) is predetermined. When the set threshold is exceeded or falls below the preset threshold, the notification module 170 provides information that "the predicted value exceeds the preset threshold or falls below the preset threshold" to the smart gateway 160. Alternatively, it is transmitted to the monitoring center 180 or the mobile phone APP 190. In this way, compared to the indoor air quality detection system 100, the data transmission step can be simplified, that is, the predicted value is first transmitted to the data receiving module 120 and then the data receiving module 120. It is not necessary to send the above information to the notification module 170 via.

(Indoor air quality detection system 100B)
Next, the indoor air quality detection system 100B according to the present invention will be described with reference to FIG.

Compared with the indoor air quality detection system 100 and 100A, the difference between the indoor air quality detection system 100B is that the data receiving module 120, the data analysis module 130, the selection prediction module 140 and the notification module 170 are two independent. The points arranged in the cloud and the plurality of detected values detected by the detection terminal 110 are first transmitted to the smart gateway 160, and then the plurality of detected values including the system ID and the equipment ID via the smart gateway 160. When the monitoring center 180 receives the information that "the predicted value exceeds the preset threshold or falls below the preset threshold", the staff of the monitoring center 180 is smart. When the air treatment device 200 is remotely controlled via a building system or the like, or when the mobile phone APP receives the above information, the owner of the mobile phone manually operates the mobile phone APP to manually control the air treatment device. There are three points: the point where the 200 can be controlled and the point where the 200 can be controlled.

Specifically, the indoor air quality detection system 100B has a prediction cloud A and a monitoring cloud B, and a data reception module 120, a data analysis module 130, and a selection prediction module 140 are arranged in the prediction cloud A. The notification module 170 was placed in the monitoring cloud B. With the above arrangement, each cloud can have a single function, the structure of the cloud can be simplified, and the architecture of the entire system can be made clear and clear.

Further, in the indoor air quality detection system 100B, the detection terminal 110 does not have to have a temporary cache function, and immediately after detecting one detection value of one or more detection targets, the detection value (and) , Equipment ID) is transmitted to the smart gateway 160. When a specific number of detected values are transmitted to the smart gateway 160, the smart gateway 160 assigns a system ID to these detected values, and together the equipment ID and the detected value to which the system ID is assigned are sent to the prediction cloud A. It is transmitted to the arranged data receiving module 120.

Regarding the third difference above, specifically, when the monitoring center 180 receives the information that "the predicted value exceeds the preset threshold value or falls below the preset threshold value", the monitoring center 180 receives the information. The stub only issues a notification to the stub without automatic control, and after noticing the notification, the stub manually sets the air treatment device 200 located in the detection area S by an intelligent system such as a smart building system. Can be controlled remotely with. On the other hand, when the mobile phone APP190 receives the above information, the smart mobile phone in which the mobile phone APP190 is installed only issues a notification to the owner of the mobile phone via the notification module, and the owner of the mobile phone. After noticing the notification, the owner of the mobile phone manually controls the air treatment device 200 by utilizing communication functions such as Bluetooth (registered trademark), NFC, and Wi-Fi of the mobile phone. Of course, when the monitoring center 180 or the mobile phone APP 190 receives the above information, it can be notified and automatically controlled directly, or the air treatment device can be controlled only after a predetermined time has passed since the notification was issued. If the 200 is not manually controlled, it can also be forced and automatically controlled again (eg, issuing a corresponding operation control command to the air treatment device 200 via the smart gateway 160).

(Indoor air quality detection system 100C)
Finally, the indoor air quality detection system 100C will be described with reference to FIG.

Compared to the indoor air quality detection system 100B, the difference of the indoor air quality detection system 100C is that the monitoring cloud B further has a data storage module 300.

As shown in FIG. 4, the smart gateway 160 predicts a specific number of detected values via the Internet by the routing device 150 in the data receiving module 120 arranged in the cloud A and the data storage module 300 arranged in the monitoring cloud B. Send at the same time. Here, the data storage module 300 stores all the input detected values for reselection of the historical time-series change model, generation of a new model, and the like. For example, when the data receiving module 120 needs to call the historical data stored in the data storage module 300, the data receiving module 120 sends a request to the smart gateway 160 and acquires the corresponding historical data via the smart gateway 160. .. At this time, the smart gateway 160 plays a role of connecting all the clouds, and when there are a plurality of clouds, the communication framework is simple and the structure of the clouds can be simplified. Reselection of the historical time series change model, generation of a new model, etc. will be described in detail later.

Further, in the above systems 100, 100A, 100B and 100C of the present invention, the routing device 150 is provided, but depending on the actual architectural situation of the system, the routing device 150 is not provided and 2G or NB-. It must be emphasized again that data reception and transmission may be performed via the smart gateway 160 having a communication module such as IoT.

(Indoor air quality prediction method)
Hereinafter, the indoor air quality prediction method according to the present invention will be described with reference to FIGS. 5 to 16 based on FIG. 1 by taking the indoor air quality detection system 100 according to the present invention as an example.

<First Embodiment>
Next, with reference to FIG. 5, a specific step of the indoor air quality prediction method according to the first embodiment will be described.

As shown in FIG. 5, first, the indoor air quality detection system 100 is activated.

Next, in step S110, the detection terminal 110 continuously detects one or more detection targets in the detection area S. Specifically, the detection terminal 110 acquires the detection value d of each detection target per unit time (for example, per minute) by one or more sensor units 111, and the detection target has a plurality of detection targets (for example, per minute). , The detection value d is a vector having three components when there are three detection targets and includes the concentration, temperature, and humidity of CO ₂ , and the detection is performed when there is only one detection target. The value d is a scalar.

In step S120, the data receiving module 120 receives the detected value transmitted from the detection terminal 110 and sets a predetermined period. Specifically, the start time of the predetermined period is T0 (for example, the first minute after activation), the end time is TE = T0 + a-1, and here, a is a unit time (for example, a). Minutes). As a result, the real-time time-series sample data set {d _T0 , d _{T0 + 1} , d consisting of a time-series detection values d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE for each detection target in the data receiving module 120. _{T0 + 2} , ..., d _TE } is formed. Further, as described above, the detection terminal 110 transmits the time-series detection value, and at the same time, further transmits the equipment ID, so that the real-time time-series sample data set {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ... ..., d _TE } can be converted into a real-time time series sample case dataset {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE , y _s }, where y _s is the detection area. The equipment ID in S is shown.

In step S130, the data receiving module 120 sets the data analysis module 130 and the selection prediction module 140 to the real-time time-series sample data set {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE } or the real-time time-series sample case, respectively. The data set {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE , y _s } is transmitted.

が所定値よりも小さければ、予め設定された条件を満たすと判定される。さらに、履歴時系列変化モデルを選択する方法は、様々あり、上記方法に限らず、例えば、さらにホールドアウト法（ｈｏｌｄ－ｏｕｔ）を用いてデータセット｛ｄ_Ｔ０、ｄ_Ｔ０＋１、ｄ_Ｔ０＋２、……、ｄ_ＴＥ｝を二つの相互排他的な第一サブ集合｛ｄ_Ｔ０、ｄ_Ｔ０＋１、ｄ_Ｔ０＋２、……、ｄ_Ｔ０’｝と第二サブ集合｛ｄ_{Ｔ０’＋１}、ｄ_{Ｔ０’＋２}、ｄ_{Ｔ０’＋３}、……、ｄ_ＴＥ｝に分けるとともに、第一サブ集合を各履歴時系列変化モデルに入力して上記第二サブ集合に対応する一つの予測値サブ集合｛Ｃ_{Ｔ０’＋１}、Ｃ_{Ｔ０’＋２}、Ｃ_{Ｔ０’＋３}、……、Ｃ_ＴＥ｝を取得し、第二サブ集合を該予測値サブ集合と比較することによって、

が所定値よりも小さい履歴時系列変化モデルを選択する。予め設定された条件を満たす履歴時系列変化モデルが存在しないと、データ分析モジュール１３０に予め記憶された初期履歴時系列変化モデルを選択するとともに、ステップＳ１５０に進む。 Next, in step S140, the selection prediction module 140 sends a model selection command to the data analysis module 130, and conditions preset between many historical time series change models and the real-time time series sample case data set. Request to select a historical time series change model that satisfies. Specifically, taking the case of a sample case as an example, after the real-time time series sample case data set is input to the data analysis module 130, first, based on the label, the data analysis module 130 uses the real-time time series. The detection values in the sample case data set are classified, that is, it is determined in which detection area the detection values in the real-time time series sample case data set belong to. Next, when it is determined that the detected value in the real-time time-series sample case data set belongs to the detection area S, the real-time time-series sample case data set is used as an all-historical time-series change model in the detection area S (these histories). It may be a set of historical time-series measured values that generate a time-series change model), and a historical time-series change model that satisfies preset conditions with the data set is selected. For preset conditions, for example, the set { _DT0 , _{DT0 + 1} , _{DT0 + 2} , ..., D _TE , ..., y _s } is within one past detection period detected in the detection area S. When showing a set of historical time-series measured values, or when showing a set of historical time-series measured values within the past multiple detection periods detected in the detection area S, that is, a set consisting of historical time-series measured values is If there are multiple data sets {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE } in each set { _DT0 , _{DT0 + 1} , _{DT0 + 2} , ..., D _TE , ...} Compare with the subset { _DT0 ', _{DT0 + 1} ', _{DT0 + 2} ', ..., D _TE '} consisting of a consecutive historical time-series measured values.

If is smaller than a predetermined value, it is determined that the condition set in advance is satisfied. Further, there are various methods for selecting the historical time-series change model, and the method is not limited to the above method. For example, a data set {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ... , D _TE } with two mutually exclusive first subsets {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _T0' } and a second subset {d _{T0'+ 1} , d _{T0' + 2} , d _{T0 '+3} , ..., d _TE }, and input the first subset into each historical time series change model, and one predicted value subset corresponding to the above second subset { _{CT0'+ 1} , _CT0 . By acquiring'+ ₂ , _CT0'+3 , ..., C _TE } and comparing the second subset with the predicted value subset.

Selects a historical time series change model in which is less than a predetermined value. If there is no historical time-series change model that satisfies the preset conditions, the initial historical time-series change model stored in advance in the data analysis module 130 is selected, and the process proceeds to step S150.

In step S150, the selection prediction module 140 sends a command to the data receiving module 120 requesting the reselection of a real-time time series sample or sample case data set of new time series detection values, and based on the command, The data receiving module 120 reselects a data set of time series detection values. Specifically, in the present embodiment, the previous N time-series detection values are deleted while the length of the set of time-series detection values does not change, and the subsequent N new times are detected. By increasing the series detection values, a new set of time series detection values {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE } is formed. After that, the processes of steps S130 to S150 are repeated until a historical time-series change model that satisfies the preset conditions is found. If there is a historical time series change model that satisfies the preset conditions, the process proceeds to step S160.

が最小であり、

が最小であることを指す。予め設定された条件を満たす、および／またはマッチング程度が最高となる履歴時系列変化モデルを選択したと、ステップＳ１７０に進む。 In step S160, it is determined whether or not there are a plurality of historical time-series change models that satisfy the preset conditions. If there is only one historical time-series change model that satisfies the preset conditions, the historical time-series change model is used (step S161), and if there are a plurality of historical time-series change models that satisfy the preset conditions, matching is performed. A historical time-series change model with the highest degree is used (step S162). For example, the highest degree of matching is

Is the smallest

Indicates that is the minimum. When the historical time-series change model that satisfies the preset conditions and / or has the highest degree of matching is selected, the process proceeds to step S170.

In step S170, the selection prediction module 140 determines the predicted values D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D for a certain period from now on based on the latest set of time series detection values and the corresponding historical time series change model. Estimate _{TE + t} , where t is a natural number greater than 1. Next, the process proceeds to step S180.

After updating the start time of the predetermined period in step S180, the process returns to step S120 to make the next prediction. Specifically, in the present embodiment, the original start time is updated to the unit time next to the original start time to obtain a new start time. For example, when the predicted value of the 11th to 20th minutes is estimated using the 10 time-series detection values of the 1st to 10th minutes, the start time of the predetermined period is updated to the 2nd minute. The predicted value for the 12th to 21st minutes is estimated using the time-series detection value for the 2nd to 11th minutes.

(Effect of the first embodiment)
According to the prediction method according to the first embodiment of the present invention, a large amount of historical data is collected in the previous period, an appropriate learning algorithm is selected, and as much induction bias is set as possible, and these historical data are used as a training sample. Alternatively, by training as a training sample case, many historical time-series change models are generated, and a historical time-series change model that matches the above-mentioned multiple detection values based on the multiple detection values detected in a specific detection area. Is selected, and the predicted value for the future fixed period of the specific detection area is estimated based on the selected historical time series change model. As a result, when the number of people in the detection area is not detected or is unknown, it is only necessary to acquire the detection value of a small part of the detection target in the indoor air, and the detection target in the specific detection area will be constant in the future. It is possible to predict the change tendency (change value) during the period of. Further, since it is not necessary to use an expensive image sensor or a complicated recognition method, the equipment cost of the indoor air quality detection system 100 can be reduced.

(Modification 1 of the first embodiment)
Next, a modification 1 of the first embodiment of the indoor air quality prediction method will be described with reference to FIG.

In step S180 of the first embodiment, as a method of updating the start time, the original start time is updated to the unit time next to the original start time to obtain a new start time. However, the method of updating the start time is not limited to this, and for example, as shown in step S180', the original start time is updated to the next unit time of the original end time TE to obtain a new start time. For example, when the predicted value of the 11th to 20th minutes is estimated using the 10 time-series detection values of the 1st to 10th minutes, the start time of the predetermined period is updated to the 11th minute. The predicted value of the 21st to 30th minutes is estimated by using the time series detection value of the 11th to 20th minutes.

(Effect of the modified example of the first embodiment)
According to the modification, as compared with the first embodiment, it is possible to estimate the predicted value of the detection target for each fixed period from now on, and at the same time, reduce the amount of calculation of the system.

(Modification 2 of the first embodiment)
Next, a modification 2 of the first embodiment of the indoor air quality prediction method will be described with reference to FIG. 7.

In step S150 of the first embodiment, as a method of reselecting a data set of time-series detection values, the previous N time-series detection values are deleted while the length of the set of time-series detection values does not change. At the same time, by increasing the N new time-series detection values detected subsequently, a set of new time-series detection values {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., D _TE } is formed. .. However, the method of reselecting the above data set is not limited to this, and for example, as shown in step S150', a set of original time series detection values {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE }. In, by increasing N new time-series detection values d _{T0 + 1} , d _{T0 + 2} , ..., d _{T0 + N} , a new set {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE , d _{T0 + 1} , d _{T0 + 2} , ..., d _{T0 + N} } is formed. Specifically, as shown in FIG. 7, the above-mentioned new set is obtained by simply delaying N unit times corresponding to the end times while the start time T0 of the original predetermined period does not change. Can be formed. However, as shown in FIG. 7, this modification further includes step S190. Specifically, in step S180, after updating the start time, the process proceeds to step S190, and in step S190, a is set so that the number of actually measured values in the time-series measured value set for each prediction start is the same. Reset.

(Modification 3 of the first embodiment)
As shown in FIG. 8, the difference from the first embodiment is that the modification 3 includes step S180'and step S150'. That is, as a method of updating the start time, there is a historical time-series change model that updates the original start time to the next unit time of the original start time to make it a new start time and satisfies a preset condition. If not, the number of detected values in the set of time-series detected values is increased so as to be the basis for continuously selecting the historical time-series change model.

(Effects of Modifications 2 and 3 of the First Embodiment)
According to the above modifications 2 and 3, when a historical time-series change model that satisfies a preset condition cannot be selected based on the original set of time-series detection values, it is subsequently detected by the set. By increasing the time-series detection values and forming more new sets of set elements, it is easier to find a historical time-series change model that is appropriate and has better generalization ability.

<Second embodiment>
Next, a second embodiment of the indoor air quality prediction method will be described with reference to FIG. 9 based on FIGS. 5 to 8.

In the second embodiment, the prediction method not only includes the prediction step in the first embodiment and its variants, but also further includes a notification step.

After estimating the predicted values D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + t} for a certain period from now on, the process proceeds to the notification step. In step S210, the data receiving module 120 acquires the predicted value from the selection prediction module 140. Next, in step S220, any of the predicted values D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + t} (indicating that there is only one predicted value when t = 1) is predetermined. Determine if the set threshold has been exceeded. When all the predicted values do not exceed the preset threshold value, the data receiving module 120 acquires a new predicted value succeeding from the selection prediction module 140 and returns to step S220. When any of the predicted values exceeds a preset threshold value, the data receiving module 120 transmits information that "the predicted value has exceeded the preset threshold value" to the notification module 170, and proceeds to step S230. .. In step S230, the notification module 170 sends a notification command to the notification unit by a predetermined method, so that the notification unit issues a notification to a related party or a related device in which the above case may occur in the future. , Or notify the parties concerned to take the corresponding measures, or instruct the related equipment to control automatically, and then proceed to step S240. In step S240, when the person concerned receives the above notification, the air treatment device 200 is immediately activated, the performance of the air treatment device 200 is adjusted, or the smart gateway 160 or the monitoring center 180 (or the mobile phone APP 190). Upon receiving the notification, the notification unit of the above immediately activates the air treatment device 200 or automatically adjusts the performance of the air treatment device 200. After that, the process proceeds from step S240 to step S250, the subsequent new detection value is acquired, and the process returns to step S220.

(Effect of the second embodiment)
According to the second embodiment, by comparing the predicted value with the preset threshold value, it is possible to take appropriate measures in a timely manner to improve the indoor air quality.

(Modification 1 of the second embodiment)
Next, a modified example 1 of the second embodiment of the indoor air quality prediction method will be described with reference to FIGS. 10 based on FIGS. 5 to 8.

In the second embodiment, when the predicted value becomes larger than the preset threshold value, the air treatment device 200 is immediately started or the performance of the air treatment device 200 is adjusted. However, the processing method is not limited to this when the predicted value becomes larger than the preset threshold value. For example, as shown in FIG. 7, when the predicted value becomes larger than the preset threshold value, the air treatment device 200 is not directly started or the performance of the air treatment device 200 is not adjusted, but the notification is given first. It may be determined whether the time lasts for a predetermined time (whether a predetermined time has elapsed) (step S231). When the duration of the notification exceeds a predetermined time, the control module of the smart gateway 160 or the monitoring center 180 or the mobile phone APP forcibly activates the air treatment device 200 or forces the performance of the air treatment device 200. Adjust (step S240').

(Effect of Modification 1 of the Second Embodiment)
According to the modification, if the predicted value still exceeds the preset threshold value after a predetermined time has elapsed from the initial notification, the air treatment device 200 is forcibly started or the air is started. Since the performance of the processing device 200 is forcibly adjusted, it is possible to avoid continuous deterioration of the air quality in the room in the detection area S due to not paying attention to the notification. Further, as compared with the second embodiment, the modification gives priority to the manual control operation of the user, whereby it is possible to avoid frequent adjustment of the air treatment device and avoid unnecessary adjustment. be able to.

(Effect of Modification 2 of the Second Embodiment)
In the second embodiment and the first modification thereof, it is notified that the predicted value becomes larger than the preset threshold value, but the present invention is not limited to this. For example, in the second modification, step S221 may be included as shown in FIG. Specifically, when the predicted value becomes larger than the preset threshold value, the air treatment device 200 is started or the performance of the air treatment device 200 is adjusted, and in this case, the value to be detected always changes. do. Taking the ppm of CO ₂ as an example, after starting the above activation or adjustment, the subsequent new predicted value is acquired, the predicted value is compared with the preset threshold value, and if the new predicted value is still preset. When it becomes larger than the threshold value, the air treatment device 200 is continuously turned on or the performance of the air treatment device 200 is continuously adjusted, and when the new predicted value becomes less than or equal to the preset threshold value, the air treatment device 200 is continuously turned on. It has been found that the operation of the device 200 plays a more useful role, and at this time, information that the new predicted value is equal to or less than a preset threshold value is transmitted to the smart gateway 160, the monitoring center 180, or the mobile phone APP. At this time, the smart gateway 160, the monitoring center 180, or the mobile phone APP 190 can stop the operation of the air treatment device 200 or reduce the operation load of the air treatment device 200 depending on the actual situation.

(Effect of Modification 2 of the Second Embodiment)
According to the modification 2, the information that "the value of the detection target is lowered to the safe concentration at a certain time from now on" can be notified in a timely manner. As a result, the operation of the air treatment device 200 can be stopped or the operating load of the air treatment device 200 can be reduced depending on the actual case, and unnecessary waste of electric power can be avoided.

<Third embodiment>
Next, a third embodiment of the indoor air quality prediction method will be described with reference to FIGS. 12 based on FIGS. 5 to 8.

The prediction method of the third embodiment includes not only the prediction step in the first embodiment but also the model evaluation step.

Specifically, after estimating the predicted values D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + t} for a certain period from now on, the selection prediction module 140 receives the data receiving module 120 after the predetermined period. The detection value d _{TE + 1} of the next unit time of is acquired (step S310). Next, in step S320, the selection prediction module 140 determines whether the error between the detected values d _{TE + 1} and D _{TE + 1} is larger than the permissible value. When the error becomes less than or equal to the allowable value, the prediction is continued using the selected historical time series change model (step S330). When the error becomes larger than the allowable value, d _{TE + 1} is added to the original set of time series detection values {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE }, and a new set {d _T0 , d _{T0 + 1} is added. , D _{T0 + 2} , ..., d _TE , d _{TE + 1} }, and using the new set, preset conditions with the new model according to steps S120 to S162 of the first embodiment. Reselect the historical time series change model to be satisfied (step S340). After selecting a new historical time series change model, the predicted values {D _{TE + 2} , D _{TE + 3} , ..., D _{TE + t} } are updated based on the model (step S350).

(Effect of the third embodiment)
According to the third embodiment, the prediction accuracy of the model can be evaluated by comparing the predicted value with the detected value of the next unit time. Also, by evaluating the error between the predicted value and the detected value of the next unit time, whether to continue using the current model based on the magnitude of the error or to reselect a model with higher accuracy. Can be determined. This makes it possible to improve the prediction accuracy.

(Modification 1 of the third embodiment)
Next, a modified example 1 of the third embodiment of the indoor air quality prediction method will be described with reference to FIGS. 13 based on FIGS. 5 to 8.

In the third embodiment, the detected value d _{TE + 1} of the next unit time after a predetermined period is acquired, the error between the detected value d TE _{+ 1} and the predicted value D _{TE + 1} is calculated, and the history is calculated based on the error. Determine if the time series change model needs to be replaced. However, the method of model evaluation is not limited to that.

を両者のマッチング程度の尺度として用いることができる）（ステップＳ３２０’）。誤差が許容値よりも小さくなると、選択された元の履歴時系列変化モデルを続けて用いる。誤差が許容値よりも大きくなると、ｄ_ＴＥ＋１、ｄ_ＴＥ＋２、ｄ_ＴＥ＋３、……、ｄ_ＴＥ＋ｍを元の時系列検出値の集合｛ｄ_Ｔ０、ｄ_Ｔ０＋１、ｄ_Ｔ０＋２、……、ｄ_ＴＥ｝に追加して、新たな集合｛ｄ_Ｔ０、ｄ_Ｔ０＋１、ｄ_Ｔ０＋２、……、ｄ_ＴＥ、ｄ_ＴＥ＋１、ｄ_ＴＥ＋２、ｄ_ＴＥ＋３、……、ｄ_ＴＥ＋ｍ｝を形成するとともに、該新たな集合を利用して第一実施形態のステップＳ１２０～ステップＳ１６２に従って、該新たなモデルとの間で予め設定された条件を満たす履歴時系列変化モデルを再選択する（ステップＳ３４０’）。新たな履歴時系列変化モデルを選択した後で、該モデルに基づいて、予測値｛Ｄ_{ＴＥ＋ｍ＋１}、Ｄ_{ＴＥ＋ｍ＋２}、……、Ｄ_ＴＥ＋ｔ｝を更新する（ステップＳ３５０’）。 For example, as shown in FIG. 13, not only one detection value d _{a + 1} is selected, but a set of detection values in the next period after a predetermined period {d _{TE + 1} , d _{TE + 2} , d _{TE + 3} , ..., D. _{TE + m} } is selected, where m is a natural number smaller than t (step S310'). Next, a subset {D _TE + 1] in the set of detected values {d _{TE + 1} , d _{TE + 2} , d _{TE + 3} , ..., d _{TE + m} } and the set of predicted values {D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + t} }. , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + m} } (for example,

Can be used as a measure of the degree of matching between the two) (step S320'). If the error is less than the tolerance, the original historical time series change model selected will continue to be used. When the error becomes larger than the allowable value, d _{TE + 1} , d _{TE + 2} , d _{TE + 3} , ..., d _{TE + m} are added to the original set of time series detection values {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE }. Then, a new set {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE , d _{TE + 1} , d _{TE + 2} , d _{TE + 3} , ..., d _{TE + m} } is formed, and the new set is used. According to steps S120 to S162 of the first embodiment, the historical time series change model satisfying the preset conditions with the new model is reselected (step S340'). After selecting a new historical time series change model, the predicted values {D _{TE + m + 1} , D _{TE + m + 2} , ..., D _{TE + t} } are updated based on the model (step S350').

(Effect of Modification 1 of the Third Embodiment)
Compared to the third embodiment, the variant 1 compares a plurality of predicted values with a plurality of detected values within one period. In this way, compared to using only a method of comparing one detected value with one predicted value, the prediction accuracy of the selected historical time series change model can be evaluated from the viewpoint of the overall change tendency, and is individual. Avoid unnecessary model exchanges due to singularities in.

(Modification 2 of the third embodiment)
Next, a modified example 2 of the third embodiment of the indoor air quality prediction method will be described with reference to FIGS. 14 based on FIGS. 5 to 8.

In the first modification, the set of detected values {d _{TE + 1} , d _{TE + 2} , d _{TE + 3} , ..., d _{TE + m} } and the set of predicted values {D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + t} } are subsets. When the error between {D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + m} } is less than the allowable value, the selected historical time series change model is continuously diverted (step S330). However, the processing in this case is not limited to that.

For example, when the above error is equal to or less than the allowable value, it can be determined whether to modify the existing model (step S321). For example, if a correction threshold value smaller than the correction threshold value is set in advance and the error becomes smaller than the correction threshold value, the selected original historical time-series change model is continuously used, and the error is equal to or larger than the correction threshold value and the allowable value. Then, further learn the original historical time series change model selected by {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE , d _{TE + 1} , d _{TE + 2} , d _{TE + 3} , ..., d _{TE + m} }. A modified historical time-series change model can be generated separately, and a subsequent prediction can be made using the modified historical time-series change model. Taking the neural network structure as an example, deep learning allows the original historical time-series change model (teacher model) to generate a separately modified historical time-series change model (student model), as well as the actual needs and system. Based on the performance of itself, the modified historical time series change model may be a derived model (same structure but different weighting coefficients) or a distillation model (same structure and different weighting coefficients).

(Effect of Modification 2 of the Third Embodiment)
Compared to the third embodiment and the first modification, in the second modification, a set of detected values {d _{TE + 1} , d _{TE + 2} , d _{TE + 3} , ..., d _{TE + m} } and a set of predicted values {D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + t} }, when the error between the subset {D _{TE + 1} , D _{TE + 2} , D _{TE + 3} , ..., D _{TE + m} } is less than the allowable value, the selected historical time series change model is always selected. It is not a diversion, and it is further determined whether to perform deep learning based on the magnitude of the above error. When it is determined that deep learning is unnecessary, the original historical time-series change model is used, while when it is determined that deep learning is necessary, a set of detected values {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d Using _TE , d _{TE + 1} , d _{TE + 2} , d _{TE + 3} , ..., d _{TE + m} }, the original historical time series change model is further learned to generate a modified historical time series change model with higher accuracy. In this way, the prediction accuracy can be further improved by using the corrected historical time series change model with higher accuracy.

(Modification 3 of the third embodiment)
Next, a modified example 3 of the third embodiment of the indoor air quality prediction method will be described with reference to FIGS. 15 based on FIGS. 5 to 8.

In the first modification of the third embodiment, when the error between the detected value d _{TE + 1} and the predicted value D _{TE + 1} in the next unit time after a predetermined period is equal to or less than the allowable value, the selected original historical time series change Continue to use the model. However, the processing method when the above error is equal to or less than the allowable value is not limited to that. For example, by combining the modified example 1 and the modified example 2, the plan shown in FIG. 15 can be obtained. That is, when the error between the detected value d _{TE + 1 and} the predicted value D _{TE + 1} is equal to or less than the allowable value, the detected values d _{TE + 2} , d _{TE + 3} , ... , D _{TE + m} is obtained, and further processing is performed according to the step described in Modification 2.

(Effect of Modification 3 of the Third Embodiment)
Compared with the modified example 1 and the modified example 2 of the third embodiment, the modified example 3 can simultaneously evaluate the prediction accuracy of the selected historical time-series change model from the two viewpoints of local and overall. ..

<Other embodiments>
The first embodiment, the second embodiment, the third embodiment and the modified examples thereof of the indoor air quality prediction system 100 and the prediction method according to the present invention have been described above. However, the indoor air quality prediction system 100 and the prediction method according to the present invention are not limited thereto.

For example, in the first embodiment, it is constant when there is no historical time series change model satisfying a predetermined condition between the set of time series detection values {d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE }. By selecting a new set of time series detection values that have a different start time and end time, or by selecting a new set of time series detection values that have the same start time but different lengths. By selecting, the historical time-series change model satisfying the preset conditions with the new set is continuously selected until such a historical time-series change model is selected. In this case, preferably until the time of selecting a historical time series change model that satisfies a preset condition, the data analysis module 130 trains the time series and all the time series detection values before the time. By learning as a sample case, one new historical time series change model is generated. Further, more preferably, in the detection process after the above time, the detected new time series detection value is continuously input to the new historical time series change model to update the historical time series change model. In this way, the amount of calculation of the data analysis module 130 can be simplified.

Furthermore, if data loss occurs during transmission of multiple time-series detection values d _T0 , d _{T0 + 1} , d _{T0 + 2} , ..., d _TE , the new set is based on a new set of the remaining time-series detection values. You may reselect a historical time series change model that meets the preset conditions with the set. In this way, even if data loss occurs, it is possible to acquire an excellent historical time series change model and make subsequent predictions.

Those skilled in the art can easily conceive of other advantages and modifications. Accordingly, the invention is, in a broader sense, not limited to the specific details and representative examples set forth and described herein. Accordingly, it can be modified without departing from the spirit or scope of the overall invention concept limited by the appended claims and their equivalents.

Japanese Patent Application Laid-Open No. 2003-09089 Japanese Patent No. 3830846 Chinese Patent Application Publication No. 1094426995 Chinese Patent Application Publication No. 109812938

Claims (32)

From a large number of historical time-series change models obtained by learning a large number of historical time-series actual measurement results including the detection location of the detection area (S) or other detection locations as historical time-series learning values.
A history that satisfies a preset condition with a set of a plurality of time-series detection values to be detected continuously detected by a sensor unit (111) provided in the detection area (S) in one predetermined period. Select a time series change model and
Based on the selected historical time-series change model, the predicted value for a certain period from now on is estimated.
A method for predicting indoor air quality, which is characterized by the fact that. When there are a plurality of historical time-series change models satisfying the preset conditions, the historical time-series change model having the highest degree of matching with the set is selected.
The indoor air quality prediction method according to claim 1, wherein the indoor air quality is predicted. If the historical time-series change model satisfying the preset condition does not exist, the start time and / or the length differs from the one predetermined period until the historical time-series change model satisfying the preset condition is found. A new set of a plurality of time-series detection values for the next predetermined period is reselected, and it is determined whether or not there is a historical time-series change model that satisfies the preset conditions with the new set. continue,
The indoor air quality prediction method according to claim 1, wherein the indoor air quality is predicted. It is generated as a new historical time-series change model by learning the time and all the time-series detection values before the time until the time when the historical time-series change model satisfying the preset condition is found.
The indoor air quality prediction method according to claim 3, wherein the indoor air quality is predicted. In the detection process after the time, the new historical time-series change model is updated by continuously inputting the detected new time-series detection value into the new historical time-series change model.
The indoor air quality prediction method according to claim 4, wherein the indoor air quality is predicted. When the predicted value exceeds a preset threshold value, a predetermined method is used to notify the possibility of occurrence in the above case, notify that a corresponding measure is taken, or take a corresponding measure. Automatically control,
The indoor air quality prediction method according to claim 1, wherein the indoor air quality is predicted. The predetermined method is
Sending notifications with buzzers, alarms, image display equipment, indicator lamps, and / or receiving signals and issuing notifications via remote controls, mobile phones, and PCs.
including,
The indoor air quality prediction method according to claim 6, wherein the indoor air quality is predicted. The corresponding means
Encourage them to open doors and windows themselves, and / or control the number of people in the detection area (S), and / or move away from the detection area (S), and / Or prompting the air treatment device (200) to be turned on manually, or forcing the air treatment device (200) to be turned on automatically.
including,
The indoor air quality prediction method according to claim 6, wherein the indoor air quality is predicted. When it is predicted that the predicted value after a certain period will fall below the preset threshold value, the possibility of occurrence in the above case is notified to the persons concerned or related equipment, or the operating load of the air treatment device (200) is increased. Notify the persons concerned or related equipment to lower or stop the operation of the air treatment apparatus (200).
The indoor air quality prediction method according to claim 6, wherein the indoor air quality is predicted. If the predicted value still exceeds the preset threshold after a predetermined time from the start of the initial notification, the air treatment device (200) is forcibly started or the performance of the air treatment device (200) is forcibly started. To adjust
The indoor air quality prediction method according to claim 6, wherein the indoor air quality is predicted. In the selected historical time-series change model, one historical time-series learning value immediately after the plurality of historical time-series learning values corresponding to the plurality of time-series detection values in the one predetermined period is the predicted value of the next unit time. Used as
The error between the predicted value of the next unit time and the detected value of the next unit time detected by the sensor unit (111) is calculated.
When the error becomes less than or equal to the allowable value, the subsequent prediction is continued using the selected historical time series change model.
The indoor air quality prediction method according to claim 1, wherein the indoor air quality is predicted. When the error becomes larger than the allowable value, the historical time series change model is reselected.
The indoor air quality prediction method according to claim 11, wherein the indoor air quality is predicted. In the selected historical time-series change model, a plurality of historical time-series learning values immediately after a plurality of historical time-series learning values corresponding to a plurality of time-series detection values in the one predetermined period are predicted values in the next predetermined period. Used as
The data set error between the set of multiple time-series detection values continuously detected by the sensor unit (111) in the next predetermined period and the set of predicted values in the next predetermined period is within the margin of error. Judge if there is,
If the data set error is within the permissible range, subsequent predictions will continue to be made using the selected historical time series change model.
The indoor air quality prediction method according to claim 1 or 11, wherein the indoor air quality is predicted. Based on the set consisting of a plurality of time-series detection values of the next predetermined period, or to a set consisting of a plurality of time-series detection values of the one predetermined period and a plurality of time-series detection values of the next predetermined period. Based on this, the selected historical time-series change model is modified to generate a modified model independently, and subsequent predictions are made using the modified model.
The indoor air quality prediction method according to claim 13, wherein the indoor air quality is predicted. If the data set error is outside the permissible range, the historical time series change model is reselected.
The indoor air quality prediction method according to claim 13, wherein the indoor air quality is predicted. The detection target is one or more of carbon dioxide, gas components other than carbon dioxide, particulate matter, temperature, and humidity.
The indoor air quality prediction method according to any one of claims 1 to 12, wherein the indoor air quality is predicted. At least one detection terminal (110) having one or more sensor units (111) provided in the detection area (S), and
A data receiving module (120) that receives a plurality of time-series detection values of detection targets continuously detected by the sensor unit (111) in one predetermined period.
A data analysis module (130) that generates many historical time-series change models by learning a large number of historical time-series actual measurement results including the detection location of the detection area (S) or other detection locations as historical time-series learning values. )When,
By selecting a historical time-series change model that satisfies a preset condition between many of the historical time-series change models and a set consisting of a plurality of the time-series detection values, the predicted value for a certain period from now on can be obtained. The selection prediction module (140) to estimate and
including,
It features an indoor air quality detection system. When there are a plurality of historical time-series change models satisfying the preset conditions, the selection prediction module (140) selects the historical time-series change model having the highest degree of matching with the set.
17. The indoor air quality detection system according to claim 17. In the absence of a historical time series change model that satisfies the preset conditions, the data receiving module (120) may perform a plurality of times in the next predetermined period in which the start time and / or the length is different from the one predetermined period. Reselect a new set of series detection values and
The selection prediction module (140) has a historical time-series change model that satisfies the preset condition with the new set until it finds a historical time-series change model that satisfies the preset condition. Continue to judge whether or not,
17. The indoor air quality detection system according to claim 17. Until the time when the historical time-series change model that satisfies the preset condition is found, the data analysis module (130) is newly developed by learning the time and all the time-series detection values before the time. Generate a historical time series change model,
19. The indoor air quality detection system according to claim 19. In the detection process after the time, the data analysis module (130) updates the new historical time series change model based on the new detected time series detection values.
20. The indoor air quality detection system according to claim 20. The selection prediction module (140) further
In the selected historical time-series change model, one historical time-series learning value immediately after the plurality of historical time-series learning values corresponding to the plurality of time-series detection values in the one predetermined period is the predicted value of the next unit time. Used as
The error between the predicted value of the next unit time and the detected value of the next unit time detected by the sensor unit (111) is calculated, and the error is calculated.
When the error is less than or equal to the permissible value, it is configured to continue making subsequent predictions using the selected historical time series change model.
17. The indoor air quality detection system according to claim 17. The selection prediction module (140) is further configured to reselect the historical time series change model when the error is greater than the permissible value.
22. The indoor air quality detection system according to claim 22. The selection prediction module (140) further
In the selected historical time-series change model, a plurality of historical time-series learning values immediately after a plurality of historical time-series learning values corresponding to a plurality of time-series detection values in the one predetermined period are predicted values in the next predetermined period. Used as
The data set error between the set of multiple time-series detection values continuously detected by the sensor unit (111) in the next predetermined period and the set of predicted values in the next predetermined period is within the margin of error. Judging whether or not there is,
If the dataset error is within the margin of error, subsequent predictions will continue to be made using the selected historical time series change model.
The indoor air quality detection system according to claim 17 or 22, wherein the indoor air quality is detected. Based on the set consisting of a plurality of time-series detection values of the next predetermined period, or to a set consisting of a plurality of time-series detection values of the one predetermined period and a plurality of time-series detection values of the next predetermined period. Based on this, the selection prediction module (140) modifies the selected historical time series change model to independently generate a modified model, and makes subsequent predictions using the modified model.
24. The indoor air quality detection system according to claim 24. The selection prediction module (140) is further configured to reselect the historical time series change model if the dataset error is outside the margin of error.
24. The indoor air quality detection system according to claim 24. A notification module (170) for issuing a notification command to a notification unit when the predicted value exceeds a preset threshold value or falls below the preset threshold value.
17. The indoor air quality detection system according to claim 17. The notification unit includes a buzzer, an alarm, an image display facility, an indicator lamp, a remote controller, a mobile phone, and a personal computer.
27. The indoor air quality detection system according to claim 27. When the predicted value exceeds a preset threshold value, the air treatment device (200) is started, the performance of the air treatment device (200) is adjusted, or the predicted value exceeds the preset threshold value. Including a control unit that forcibly activates the air treatment device (200) or forcibly adjusts the performance of the air treatment device (200) after a certain period of time has passed.
The indoor air quality detection system according to any one of claims 17, 27 and 28. When it is predicted that the predicted value after a certain period will fall below a preset threshold value, the control unit reduces the operating load of the air treatment device (200) or operates the air treatment device (200). Stop,
29. The indoor air quality detection system according to claim 29. The control unit is a smart gateway (160).
The indoor air quality detection system according to claim 29 or 30, characterized in that. The control unit is attached to a monitoring center (180) provided in a region other than the detection area (S).
The indoor air quality detection system according to claim 29 or 30, characterized in that.

Images