KR20230050140A - Server, method and computer program for determining control policy that automatically controls equipments in data center - Google Patents

Abstract

A server determining a control policy for automatically controlling equipment in a data center comprises: a data collecting unit which collects sensor data sensed by a plurality of sensors from a building management system in a data center, and operating data for the operation of a plurality of pieces of equipment in the data center; a clustering unit which clusters the plurality of sensors based on the sensor data to generate a plurality of first groups for an indoor environment, and clusters the plurality of pieces of equipment based on the operating data to generate a plurality of second groups for a system environment; an influence degree analysis unit which, based on the fractal analysis between the plurality of first groups and the plurality of second groups, analyzes an influence degree between the plurality of first groups and the plurality of second groups; and a control policy determining unit which, based on the analysis result of the influence degree, determines a control policy for automatically controlling equipment.

Description

Server, method and computer program for determining a control policy for automatically controlling equipment in a data center

The present invention relates to a server, method and computer program for determining control policies for automatically controlling equipment in a data center.

A data center is a facility that integrates and manages equipment necessary to provide IT services, such as servers, networks, and storage, in one building. A data center consists of servers, storage, and network devices for providing computing services, and generators, uninterruptible power supplies (UPS), thermo-hygrostats, backup systems, and security systems necessary to maintain these devices.

The size of these data centers is gradually increasing due to the growth of the ICT (Information Communications Technology) industry, and as energy consumption increases accordingly, energy efficiency improvement projects are being promoted. Ho discloses an energy-saving Internet data center air conditioning system.

Methods to improve data center energy efficiency include passive methods such as sensor location change, high-efficiency system replacement, panel and containment installation, free cooling, and raised floor take-out tile location change. There are active methods such as on/off control, set temperature control, and mode control for heat sources/air conditioning/transport equipment using optimal control operation solutions for

Currently, energy efficiency applied to data centers relies on passive methods, which are operated according to the facility operator's know-how or subjective judgment, so excessive cooling operation is performed to maintain customer contract operating conditions rather than energy reduction compared to other buildings. It has the downside of being A process of performing an excessive cooling operation for the data center will be described with reference to FIG. 1 .

1 is an exemplary diagram for explaining a process of performing an excessive cooling operation in a conventional data center. Referring to FIG. 1 , a data center is divided into an indoor environment 100 where a plurality of servers are located and a system environment 110 where a plurality of thermostats are located.

In the case of a building requiring a lot of wiring, such as a data center, it is common to install a raised floor 120. However, in the case of a data center, airflow uncertainty is generated on the floor or ceiling of the raised floor 120 in an indoor environment where the server is located due to the raised floor 120 .

As a result, as uncertainty arises in the air supply pattern and ventilation pattern due to the mixing of airflow between the system environment and the indoor environment of the data center, the manager of the data center decides between hot aisle or cold aisle for each thermostat. It has a disadvantage that it is difficult to accurately grasp which path or rack it affects.

Excessive cooling operation is caused by the airflow uncertainty generated in the data center, and as the excessive cooling operation is set to a fixed set value for each thermo-hygrostat, precise control of the system in the facility due to rack load fluctuations in the data center is performed. I have a problem that makes it difficult.

It collects sensor data sensed by a plurality of sensors from the building management system of the data center and operation data on the operation of a plurality of facilities in the data center, and clusters the plurality of sensors based on the sensor data to provide a plurality of indoor environment information. It is intended to provide a server, method, and computer program for generating a first group and clustering a plurality of facilities based on operation data to create a plurality of second groups for a system environment.

Control for analyzing the degree of influence between the plurality of first groups and the plurality of second groups based on the fractal analysis between the plurality of first groups and the plurality of second groups, and automatically controlling the facility based on the analysis result of the influence degree It is intended to provide a server, method and computer program for determining policy.

However, the technical problem to be achieved by the present embodiment is not limited to the technical problems described above, and other technical problems may exist.

As a means for achieving the above-described technical problem, an embodiment of the present invention provides sensor data sensed by a plurality of sensors from a building management system of a data center and driving data for the operation of a plurality of facilities in the data center. A data collection unit to generate a plurality of first groups for indoor environments by clustering the plurality of sensors based on the sensor data, and clustering the plurality of facilities based on the driving data to generate a plurality of first groups for system environments. A clustering unit for generating a second group of , the degree of influence between the plurality of first groups and the plurality of second groups based on a fractal analysis between the plurality of first groups and the plurality of second groups It is possible to provide a server including an influence analysis unit for analysis and a control policy determination unit for determining a control policy for automatically controlling facilities based on the analysis result of the influence.

Another embodiment of the present invention is a step of collecting sensor data sensed by a plurality of sensors from the building management system of the data center and operation data for the operation of a plurality of facilities in the data center, based on the sensor data Clustering a plurality of sensors to create a plurality of first groups for indoor environments; Generating a plurality of second groups for system environments by clustering the plurality of facilities based on the driving data; Analyzing the influence between the plurality of first groups and the plurality of second groups based on a fractal analysis between the first group and the plurality of second groups, and based on the analysis result of the influence A method for determining a control policy including determining a control policy for automatically controlling a facility may be provided.

Another embodiment of the present invention, when the computer program is executed by a computing device, sensor data sensed by a plurality of sensors from the building management system of the data center and driving data for the operation of a plurality of facilities in the data center. and clustering the plurality of sensors based on the sensor data to create a plurality of first groups for an indoor environment, and clustering the plurality of facilities based on the driving data to generate a plurality of second groups for a system environment. A group is created, an influence between the plurality of first groups and the plurality of second groups is analyzed based on a fractal analysis between the plurality of first groups and the plurality of second groups, and the influence It is possible to provide a computer program stored in a computer readable recording medium including a sequence of instructions for determining a control policy for automatically controlling equipment based on the analysis result of FIG.

The above-described means for solving the problems is only illustrative and should not be construed as limiting the present invention. In addition to the exemplary embodiments described above, there may be additional embodiments described in the drawings and detailed description.

According to any one of the above-described problem solving means of the present invention, a plurality of first groups for an indoor environment are generated by clustering a plurality of sensors based on sensor data, and a plurality of facilities are clustered based on driving data to create a system. By generating a plurality of second groups for the environment and analyzing the influence between the plurality of first groups and the plurality of second groups based on the fractal analysis between the plurality of first groups and the plurality of second groups, It is possible to provide a server, method, and computer program that can optimally control the system by solving the air flow uncertainty problem caused by the raised floor.

It is possible to provide a server, method, and computer program for estimating hyperparameters for clustering of a plurality of sensors and clustering of a plurality of facilities at predetermined intervals using a reinforcement learning-based model and self-learning the hyperparameters.

1 is an exemplary diagram for explaining a process of performing an excessive cooling operation in a conventional data center.
2 is a configuration diagram of a server according to an embodiment of the present invention.
3A and 3B are exemplary diagrams for explaining a process of generating a plurality of first groups based on sensor data and generating a plurality of second groups based on driving data, according to an embodiment of the present invention. .
4A to 4C are exemplary diagrams for explaining a process of analyzing the degree of influence between a plurality of first groups and a plurality of second groups according to an embodiment of the present invention.
5 is a flowchart of a method for determining a control policy for automatically controlling facilities in a data center in a server according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail so that those skilled in the art can easily practice the present invention with reference to the accompanying drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case where it is "directly connected" but also the case where it is "electrically connected" with another element interposed therebetween. . In addition, when a part "includes" a certain component, this means that it may further include other components, not excluding other components, unless otherwise stated, and one or more other characteristics. However, it should be understood that it does not preclude the possibility of existence or addition of numbers, steps, operations, components, parts, or combinations thereof.

In this specification, a "unit" includes a unit realized by hardware, a unit realized by software, and a unit realized using both. Further, one unit may be realized using two or more hardware, and two or more units may be realized by one hardware.

In this specification, some of the operations or functions described as being performed by a terminal or device may be performed instead by a server connected to the terminal or device. Likewise, some of the operations or functions described as being performed by the server may also be performed in a terminal or device connected to the corresponding server.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

2 is a configuration diagram of a server according to an embodiment of the present invention. Referring to FIG. 2 , the server 200 includes a data collection unit 210, a preprocessing unit 220, a clustering unit 230, an influence analysis unit 240, a control policy determination unit 250, and a parameter estimation unit. (260).

The data collection unit 210 may collect sensor data sensed by a plurality of sensors and operation data about the operation of a plurality of facilities in the data center from a facility maintenance service (FMS) of the data center. For example, the data collection unit 210 collects sensor data and operation data related to power or facilities of the data center from the building management system, sets standard information for the collected sensor data and operation data, and controls tags. You can perform point setting and extraction, etc.

The preprocessing unit 220 may perform preprocessing to remove outliers and noise from at least one of sensor data and driving data.

To this end, the preprocessing unit 220 may first determine whether missing values and noise are included in at least one of the sensor data and the driving data by a threshold value or more. For example, if it is determined that at least one of sensor data and driving data contains missing values and noise above a threshold value, the transmission unit (not shown) informs the administrator terminal that the use of data including missing values and noise above a threshold value is impossible. can be notified by

Thereafter, the preprocessing unit 220 may secondarily determine whether at least one of the sensor data and the driving data corresponds to stationary time series data or non-stationary time series data. For example, the preprocessing unit 220 uses a statistical verification method (KPSS, Kwiatkowski-Philips-Schmidt-Shin test) to determine whether at least one of sensor data and driving data is either stationary time series data or non-stationary time series data. It can be judged whether it corresponds to Here, the statistical verification method means a verification method in which acceptance or rejection of the null hypothesis is determined based on the null hypothesis set to “the time series is a stationary time series”.

For example, the preprocessing unit 220 adopts the null hypothesis if the P-value derived through the statistical verification method is greater than or equal to a significant value (eg, 0,05), and rejects the null hypothesis if it is less than or equal to a significant value (eg, 0,05). can do. At this time, if the null hypothesis is rejected, the preprocessing unit 220 may adopt the opposite hypothesis of "the time series is a non-stationary time series" that corresponds to the opposite of the null hypothesis. In this statistical verification method, verification may be performed using predetermined data (eg, data corresponding to 2 minutes) every predetermined period (eg, every hour) for at least one of sensing data and driving data.

According to the present invention, by determining whether at least one of the above-described sensor data and driving data corresponds to either stationary time series data or non-stationary time series data, a suitable filtering algorithm may be applied according to the time series characteristics of the data. That is, in the case of stationary time series data, the average, variance, and covariance are constant over time, and non-stationary time series data (in general, data in the real physical environment are non-stationary time series data as trends or seasonality are reflected). In the case of high probability), the mean, variance, and covariance over time are not constant. Therefore, according to the present invention, sensor data and driving data corresponding to raw data may be filtered by applying an optimal filtering algorithm according to time-series characteristics.

For example, when at least one of sensor data and driving data corresponds to stationary time series data, the preprocessing unit 220 applies a Fast Fourier Transform Denoiser to the sensor data and driving data to filtering can be performed.

For another example, when at least one of the sensor data and the driving data corresponds to non-stationary time series data, the preprocessing unit 220 applies a Kalman filter to at least one of the sensor data and the driving data for filtering. can be performed.

Returning to FIG. 2 , the clustering unit 230 clusters a plurality of sensors based on the sensor data to generate a plurality of first groups for the indoor environment, and clusters a plurality of facilities based on the driving data to create a system environment. It is possible to create a plurality of second groups for. A process of generating a plurality of first groups for the indoor environment and generating a plurality of second groups for the system environment will be described in detail with reference to FIGS. 3A and 3B.

3A and 3B are exemplary diagrams for explaining a process of generating a plurality of first groups based on sensor data and generating a plurality of second groups based on driving data, according to an embodiment of the present invention. .

Referring to FIG. 3A , the clustering unit 230 calculates the environmental enthalpy of the indoor environment 100 based on sensor data composed of time-series data (eg, data related to temperature and humidity sensed by a plurality of sensors). can be calculated For example, the clustering unit 230 uses sensor data of sensor 1 311 and sensor 2 312 to determine the environmental enthalpy 322 of the indoor environment 100 based on the temperature 320 and humidity 321. ) can be calculated. Here, the closer the environmental enthalpy 322 calculated using sensor data of sensor 1 311 and sensor 2 312 is to the diagonal 323, the higher the similarity may be considered.

In the same way, the clustering unit 230 may calculate the system enthalpy for the system environment 110 based on operation data composed of time-series data (eg, data related to ventilation temperature and humidity of multiple thermostats). .

According to the present invention, by calculating the environmental enthalpy 322 and the system enthalpy, dimensions can be matched for analysis of mutual influence between the indoor environment 100 and the system environment 110 .

The clustering unit 230 may cluster a plurality of sensors by performing dynamic time warping (DTW) on environmental enthalpy and cluster a plurality of facilities by performing dynamic time warping on system enthalpy. Here, clustering for a plurality of sensors and clustering for a plurality of facilities is defined as a physical environment consisting of a structurally stable system that does not consider the air flow uncertainty of the raised floor 120 in the data center, so time-series data suitable for these physical environments Dynamic time warping can be performed with a clustering algorithm applicable to . Here, dynamic time warping is an algorithm that verifies whether patterns match to evaluate the similarity between vectors in temporal order, and has the advantage of deriving better performance indicators than existing RMSE (Root Mean Square Error) or MAE (Mean Absolute Error) have

For example, the clustering unit 230 clusters a plurality of sensors having similar patterns through similarity evaluation on time-series data through dynamic time warping performed for each of the indoor environment 100 and the system environment 110, Multiple facilities can be clustered.

The clustering of the plurality of sensors and the clustering of the plurality of facilities may be updated every predetermined time period (eg, 1 hour).

Referring to FIG. 3B , the clustering unit 230 clusters a plurality of sensors based on sensor data to generate a plurality of first groups 330 for the indoor environment 100 and a plurality of facilities based on driving data. A plurality of second groups 340 for the system environment 110 may be generated by clustering.

For example, the clustering unit 230 performs clustering on a total of 81 sensors belonging to the indoor environment 100 in the data center based on sensor data to generate a total of 17 first groups 330, and Clustering is performed on a total of 15 thermostats belonging to my system environment 110 based on operation data to determine the first subgroup 341, the second subgroup 342, the third subgroup 343, and the fourth subgroup. A total of four second groups 340 composed of groups 344 can be created.

Returning to FIG. 2 , the influence analyzer 240 determines the influence between the plurality of first groups and the plurality of second groups based on the fractal analysis between the plurality of first groups and the plurality of second groups. can be analyzed. A process of analyzing the degree of influence between a plurality of first groups and a plurality of second groups will be described in detail with reference to FIGS. 4A to 4C.

Referring to FIG. 4A , the influence analyzer 240 analyzes a plurality of first groups and systems for the indoor environment 100 through fractal analysis based on the results of generating the plurality of first groups and the plurality of second groups. A degree of influence between a plurality of second groups on the environment 110 may be analyzed. This is because the data center is composed of the physical environment of the structurally unstable system 400 due to the air flow uncertainty of the raised floor 120 in the data center, so that the plurality of first groups for the indoor environment 100 for the disordered time series data This is because it is required to analyze the degree of influence (ie, sensitivity) between the plurality of second groups on the system environment 110 and to select an algorithm considering this.

To this end, the influence analysis unit 240 performs fractal analysis based on emergent phenomena to discover orderly rules through mutual influence analysis under uncertainty (disordered world) to determine a plurality of first groups and Sensitivity 410 between a plurality of second groups may be calculated. Fractal analysis is a method for explaining Complex Theory, which is expressed in the form of a power function law, a scale-invariant structure with autosimilarity characteristics in which small structures repeat in a similar form within the structure of the entire time series data. The power law shows that patterns exist even in irregular data, and at the same time, the pattern similarity between the indoor environment 100 and the system environment 110 can be expressed through the Hurst index and the fractal spectrum width difference between the indoor environment and the system environment. there is. For example, through fractal analysis of a plurality of first groups and a plurality of second groups, it is possible to determine that sufficiently orderly data exists even in disordered time-series data between the indoor environment 100 and the system environment 110. In this case, the lower the fractal index, the higher the sensitivity 410 index.

For example, the influence analysis unit 240 calculates the sensitivity 410 for a plurality of second groups for the system environment 110 based on the plurality of first groups for the indoor environment 100, and It may be normalized so that the sum of the sensitivity indices between a plurality of second groups for any one of the first groups of is '1'. Based on this, the sensitivity 410 between the plurality of first groups for one second group may be calculated through the reverse of the calculated sensitivity 410 .

This fractal analysis can proceed in a total of four steps. Hereinafter, each step of fractal analysis will be explained.

In step 1, the influence analysis unit 240 may change the noise time-series data to accumulated difference data in order to analyze them.

In step 2, the influence analysis unit 240 may calculate the Hurst index by analyzing the time series data in a local dimension. For example, the influence analysis unit 240 calculates a Hurst Exponent between a plurality of second groups for the system environment 110 based on a plurality of first groups for the indoor environment 100, Fractal characteristics between a plurality of first groups and a plurality of second groups may be analyzed based on the calculated Hurst index.

In step 3, the influence analysis unit 240 may analyze whether noise, fractal, or random walk characteristics are present through the Hurst index.

In step 4, the influence analyzer 240 may calculate a spectrum width difference between a plurality of first groups for an indoor environment and a plurality of second groups for a system environment through multi-fractal spectrum analysis. For example, the influence analysis unit 240 calculates a difference in spectrum width between a plurality of first groups and a plurality of second groups through multi-fractal spectrum analysis based on fractal characteristics, and based on the difference in spectrum width, It is possible to calculate the degree of influence between the first group and the plurality of second groups. In this case, the closer the spectral width difference is, the higher the index of the sensitivity 410 may be, and the sum of the index of the sensitivity 410 may be '1' through a normalization process.

Referring to FIG. 4B , for example, the influence analysis unit 240 determines subgroup A 431 and subgroup B 432 of the plurality of second groups 430 for thermostats belonging to the system environment 110. , the fractal indices between the D subgroup 421 among the plurality of first groups 420 for the plurality of sensors belonging to the C subgroup 433 and the indoor environment 100 are '0.1', '0.2', and '0.4', respectively. ', and by normalizing it, it can be derived as '0.43', '0.36', and '0.21', respectively.

In addition, the influence analysis unit 240 determines the A subgroup 431, B subgroup 432, and C subgroup 433 among the plurality of second groups 430 for thermostats belonging to the system environment 110. The process of calculating the fractal index between the E subgroups 422 among the plurality of first groups 420 for the plurality of sensors belonging to the indoor environment 100 and then calculating the influence between the F subgroups 423 can be repeated.

Referring to FIG. 4C , a result of analysis of influence between a plurality of first groups and a plurality of second groups is shown, and four groups 441 for thermostats belonging to the system environment 110 and the indoor environment 100 are shown. Sensitivity indicators 440 between 17 groups 442 for the belonging sensor may be calculated. Here, the sensitivity indicator 440 may be calculated every hour.

Returning to FIG. 2 again, the control policy determination unit 250 may determine a control policy for automatically controlling the facility based on the analysis result of the degree of influence.

The parameter estimator 260 may estimate hyperparameters for clustering of a plurality of sensors and clustering of a plurality of facilities by using a reinforcement learning-based model. Here, the presence of hyperparameters for generating a plurality of first groups and a plurality of second groups and calculating the degree of influence between the plurality of first groups and the plurality of second groups causes uncertainty in the prediction result. .

Therefore, in the present invention, by estimating hyperparameters every predetermined period (eg, every week) to estimate optimal parameters, a model suitable for the space of the indoor environment and the space of the system environment in the data center based on sensor data and driving data can be self-learned.

Such a server 200 may be executed by a computer program stored on a medium containing a sequence of instructions for determining a control policy for automatically controlling equipment in a data center. When the computer program is executed by a computing device, it collects sensor data sensed by a plurality of sensors from the building management system of the data center and driving data about the operation of a plurality of facilities in the data center, and collects data based on the sensor data. A plurality of first groups for indoor environments are generated by clustering sensors, a plurality of second groups are generated for system environments by clustering a plurality of facilities based on driving data, and a plurality of first groups and a plurality of second groups are generated. Analyzing the degree of influence between a plurality of first groups and a plurality of second groups based on fractal analysis between two groups, and determining a control policy for automatically controlling equipment based on the analysis result of the degree of influence It can contain a sequence of instructions.

5 is a flowchart of a method for determining a control policy for automatically controlling facilities in a data center in a server according to an embodiment of the present invention. Referring to FIG. 5 , a method of determining a control policy for automatically controlling facilities in a data center performed by a server 200 includes steps processed in time series according to the embodiment shown in FIGS. 2 to 4C. . Therefore, even if the content is omitted below, it is also applied to the method of determining the control policy for automatically controlling facilities in the data center performed in the server 200 according to the embodiments shown in FIGS. 2 to 4C.

In step S510, the server 200 may collect sensor data sensed by a plurality of sensors and driving data about the operation of a plurality of facilities in the data center from the building management system of the data center.

In step S520, the server 200 may create a plurality of first groups for the indoor environment by clustering a plurality of sensors based on the sensor data.

In step S530, the server 200 may create a plurality of second groups for the system environment by clustering a plurality of facilities based on the operation data.

In step S540, the server 200 may analyze the degree of influence between the plurality of first groups and the plurality of second groups based on the fractal analysis between the plurality of first groups and the plurality of second groups.

In step S550, the server 200 may determine a control policy for automatically controlling the facility based on the analysis result of the degree of influence.

In the above description, steps S510 to S550 may be further divided into additional steps or combined into fewer steps, depending on an embodiment of the present invention. Also, some steps may be omitted as needed, and the order of steps may be switched.

A method for determining a control policy for automatically controlling facilities in a data center performed by a server described with reference to FIGS. 2 to 5 is a computer program stored in a medium executed by a computer or a record including instructions executable by the computer. It can also be implemented in the form of a medium. In addition, the method for determining a control policy for automatically controlling facilities in a data center performed by a server described with reference to FIGS. 2 to 5 may be implemented in the form of a computer program stored in a medium executed by a computer.

Computer readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. Also, computer readable media may include computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.

The above description of the present invention is for illustrative purposes, and those skilled in the art can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, the embodiments described above should be understood as illustrative in all respects and not limiting. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and equivalent concepts should be construed as being included in the scope of the present invention. do.

200: server
210: data collection unit
220: preprocessing unit
230: clustering unit
240: influence analysis unit
250: control policy decision unit
260: parameter estimation unit

Claims (17)

In a server that determines a control policy for automatically controlling facilities in a data center,
a data collection unit that collects sensor data sensed by a plurality of sensors from the building management system of the data center and driving data about the operation of a plurality of facilities in the data center;
Based on the sensor data, a plurality of first groups for indoor environments are generated by clustering the plurality of sensors, and a plurality of second groups for system environments are generated by clustering the plurality of facilities based on the driving data. a clustering unit that does;
an influence analyzer configured to analyze an influence between the plurality of first groups and the plurality of second groups based on a fractal analysis between the plurality of first groups and the plurality of second groups; and
A control policy determination unit for determining a control policy for automatically controlling equipment based on the analysis result of the influence.
A server that includes a.
According to claim 1,
Wherein the clustering unit calculates environmental enthalpy for the indoor environment based on the sensor data, and calculates system enthalpy for the system environment based on the driving data.
According to claim 2,
The clustering unit clusters the plurality of sensors by performing dynamic time warping (DTW) on the environmental enthalpy, and clusters the plurality of facilities by performing the dynamic time warping on the system enthalpy. , server.
According to claim 1,
The influence analysis unit calculates a Hurst Exponent between the plurality of second groups for the system environment based on the plurality of first groups for the indoor environment,
And analyzing fractal characteristics between the plurality of first groups and the plurality of second groups based on the calculated Hurst index.
According to claim 4,
The influence analyzer calculates a difference in spectrum width between the plurality of first groups and the plurality of second groups through multi-fractal spectrum analysis based on the fractal characteristics, and based on the difference in spectrum width, the plurality of second groups. And calculating the degree of influence between the first group and the plurality of second groups.
According to claim 1,
The server further comprises a pre-processing unit configured to perform pre-processing to remove outliers and noise from at least one of the sensor data and the driving data.
According to claim 6,
The pre-processing unit determines whether at least one of the sensor data or the driving data corresponds to either steady time series data or non-stationary time series data;
When at least one of the sensor data or the driving data corresponds to stationary time-series data, performing filtering by applying a fast Fourier transform denoiser to the sensor data,
When at least one of the sensor data or the driving data corresponds to non-stationary time-series data, filtering is performed by applying a Kalman filter to the sensor data.
According to claim 1,
The server further comprising a parameter estimator for estimating hyperparameters for clustering of the plurality of sensors and clustering of the plurality of facilities using a reinforcement learning-based model.
A method for determining a control policy for automatically controlling facilities in a data center in a server,
Collecting sensor data sensed by a plurality of sensors from a building management system of a data center and operation data for operation of a plurality of facilities in the data center;
generating a plurality of first groups for an indoor environment by clustering the plurality of sensors based on the sensor data;
generating a plurality of second groups for system environments by clustering the plurality of facilities based on the operation data;
analyzing an influence between the plurality of first groups and the plurality of second groups based on a fractal analysis between the plurality of first groups and the plurality of second groups; and
Determining a control policy for automatically controlling equipment based on the analysis result of the influence
To include, a control policy determination method.
According to claim 9,
calculating environmental enthalpy for the indoor environment based on the sensor data; and
Further comprising calculating a system enthalpy for the system environment based on the operation data, the control policy determination method.
According to claim 10,
The step of generating the plurality of first groups,
Clustering the plurality of sensors by performing dynamic time warping (DTW) on the environmental enthalpy;
The step of generating the plurality of second groups,
and clustering the plurality of facilities by performing the dynamic time warping on the system enthalpy.
According to claim 9,
In the step of analyzing the influence,
calculating a Hurst Exponent between the plurality of second groups for the system environment based on the plurality of first groups for the indoor environment; and
and analyzing fractal characteristics between the plurality of first groups and the plurality of second groups based on the calculated Hurst index.
According to claim 12,
In the step of analyzing the influence,
calculating a difference in spectrum width between the plurality of first groups and the plurality of second groups through multi-fractal spectrum analysis based on the fractal characteristics; and
and calculating an influence between the plurality of first groups and the plurality of second groups based on the difference in spectrum width.
According to claim 9,
The control policy determination method further comprising performing preprocessing to remove outliers and noise from at least one of the sensor data and the driving data.
15. The method of claim 14,
Performing the preprocessing step,
determining whether at least one of the sensor data and the driving data corresponds to either stationary time-series data or non-stationary time-series data;
performing filtering by applying a Fast Fourier Transform Denoiser to the sensor data when at least one of the sensor data and the driving data corresponds to stationary time-series data; and
And performing filtering by applying a Kalman filter to the sensor data when at least one of the sensor data or the driving data corresponds to non-stationary time-series data.
According to claim 9,
The control policy determination method further comprising estimating hyperparameters for clustering of the plurality of sensors and clustering of the plurality of facilities using a reinforcement learning-based model.
A computer program stored on a computer readable recording medium comprising a sequence of instructions for determining a control policy for automatically controlling equipment in a data center,
When the computer program is executed by a computing device,
Collecting sensor data sensed by a plurality of sensors from the building management system of the data center and operation data for the operation of a plurality of facilities in the data center;
Based on the sensor data, a plurality of first groups for indoor environments are generated by clustering the plurality of sensors, and a plurality of second groups for system environments are generated by clustering the plurality of facilities based on the driving data. do,
Analyzing the degree of influence between the plurality of first groups and the plurality of second groups based on a fractal analysis between the plurality of first groups and the plurality of second groups;
A computer program stored on a computer readable recording medium comprising a sequence of instructions for determining a control policy for automatically controlling a facility based on an analysis result of the degree of influence.

Images