TWI910348B - Automatic suggestion ordering system and automatic suggestion ordering method - Google Patents

Abstract

The present disclosure provides an automatic suggestion ordering system, which includes a cloud server and a computer device, and the computer device is connected to the cloud server. The cloud server uses historical sales data and external factor data to train and test multiple different machine learning models, so as to select a most accurate machine learning model from the multiple different machine learning models, and then use the most accurate machine learning model to generate a prediction result. The computer device obtains the prediction result from the cloud server, and then calculates the information of the suggestion order quantity of the product according to the prediction result and the inventory quantity of the product.

Description

Automated suggestion ordering system and automated suggestion ordering method

This invention relates to an automation system and method, and more particularly to an automation suggestion ordering system and method.

In a narrow sense, fresh food refers to ready-to-eat foods provided by manufacturers in conjunction with convenience store operators, such as bento boxes, rice balls, sandwiches, cold noodles, steamed buns, hot dogs, etc. These are mostly products that manufacturers tailor-make for convenience stores.

However, fresh food products have a short shelf life, and there are complex environmental variables that affect sales results. Currently, they are ordered manually, making it impossible to objectively consider all the influencing variables, which leads to lost sales opportunities or losses due to improper disposal.

This invention proposes an automated suggestion ordering system and an automated suggestion ordering method to improve upon the problems of prior art.

In one embodiment of the present invention, the automated suggested ordering system includes a cloud server and a computer device connected to the cloud server. The cloud server trains and tests multiple different machine learning models using historical sales data and external factor data, selecting the most accurate model from among them, and then generating a prediction result using the most accurate model. The computer device obtains the prediction result from the cloud server and calculates the suggested order quantity information based on the prediction result and the product inventory quantity.

In one embodiment of the present invention, the cloud server includes a storage device, a processor, and a communication device. The communication device is electrically connected to the processor, and the processor is electrically connected to the storage device. The storage device stores historical sales and external factor data, multiple different machine learning models, and future external factor information. The processor adjusts the prediction results based on the future external factor information, and the communication device transmits the prediction results to a computer device.

In one embodiment of the invention, the processor is a central processing unit, and a plurality of different machine learning models include an exponential smoothing model. The processor uses the exponential smoothing model to perform differentiated weighted averaging on historical sales data of fewer than a preset number of days according to holidays, summer and winter vacations, and different weeks to produce a prediction.

In one embodiment of the invention, the processor is a central processing unit, and a plurality of different machine learning models include generalized additive models. The processor uses the generalized additive models to decompose historical sales data and external factor data of more than a preset number of days into trend modules, seasonal modules and external factor modules in a time series manner, and then superimposes or multiplies the outputs of trend modules, seasonal modules and external factor modules to produce a prediction.

In one embodiment of the present invention, the computer device includes a network device, a storage device, a processing device, and a display device. The display device is electrically connected to the processing device, and the processing device is electrically connected to the network device and the storage device. The network device is connected to a cloud server. The network device receives prediction results from the cloud server. The storage device stores the inventory quantity of goods. Based on the prediction results and the inventory quantity of goods, the processing device calculates the suggested order quantity information for goods. The display device displays the suggested order quantity information for goods. The processing device transmits the suggested order quantity information for goods and the corresponding actual order quantity of goods back to the cloud server through the network device.

In one embodiment of the present invention, the automated suggested ordering method proposed by the present invention includes the following steps: training and testing multiple different machine learning models using historical sales data and external factor data through a cloud server, selecting the most accurate machine learning model from the multiple different machine learning models, and then generating a prediction result through the most accurate machine learning model; obtaining the prediction result from the cloud server through a computer device, and then calculating the suggested order quantity information of the goods based on the prediction result and the quantity of goods in stock.

In one embodiment of the present invention, the automated suggested ordering method further includes: obtaining information on future external factors through a cloud server; adjusting the prediction results based on the information on future external factors through a cloud server; and transmitting the prediction results to a computer device through a cloud server.

In one embodiment of the invention, the plurality of different machine learning models include an exponential smoothing model, and the automated suggested ordering method further includes: using the exponential smoothing model to perform differentiated weighted averaging of historical sales data for fewer than a preset number of days by holidays, summer and winter vacations and different weeks through a cloud server to generate a prediction.

In one embodiment of the present invention, a plurality of different machine learning models include generalized additive models, and the automated suggested ordering method further includes: using a cloud server to decompose historical sales data and external factor data of more than a preset number of days into trend modules, seasonal modules and external factor modules in a time series manner, and then superimposing or multiplying the outputs of trend modules, seasonal modules and external factor modules to produce predictions.

In one embodiment of the present invention, the automated suggested ordering method further includes: displaying suggested order quantity information of goods through a computer device; and transmitting the suggested order quantity information of goods and the corresponding actual order quantity of goods back to a cloud server through a computer device.

In summary, the technical solution of this invention has significant advantages and beneficial effects compared with existing technologies. Through the automated suggested ordering system and method of this invention, an algorithmic model is established using artificial intelligence machine learning capabilities to automatically predict future sales and calculate the optimal suggested order quantity, significantly reducing the disadvantages of purely manual ordering.

The above description will be given in detail below by way of implementation, and the technical solution of the present invention will be further explained.

To make the description of this invention more detailed and complete, reference can be made to the accompanying drawings and the various embodiments described below, in which the same numbers represent the same or similar elements. On the other hand, well-known elements and steps are not described in the embodiments to avoid unnecessarily limiting the invention.

Referring to Figure 1, the present invention is an automated suggested ordering system 100, which can be applied in convenience stores or widely used in related technical aspects. This automated suggested ordering system 100 represents a significant technological advancement and has broad industrial applicability. The specific implementation of the automated suggested ordering system 100 will be explained below with reference to Figure 1.

It should be understood that various implementations of the automated suggested ordering system 100 are described in conjunction with Figure 1. In the following description, for ease of explanation, numerous specific details are further defined to provide a comprehensive explanation of one or more implementations. However, this technology can be implemented without these specific details. In other examples, known structures and devices are shown in block diagram form to effectively describe these implementations. The term "for example" is used herein to mean "as an example, instance, or illustration." Any implementation described herein as "for example" should not be interpreted as better or superior to other implementations.

Figure 1 is a block diagram of an automated suggested ordering system 100 according to an embodiment of the present invention. As shown in Figure 1, the automated suggested ordering system 100 includes a cloud server 110 and a computer device 120. Architecturally, the computer device 120 is connected to the cloud server 110. It should be understood that in the embodiments and scope of the claims, the description of "connection" can refer to one component directly or indirectly connected to another component. For example, the cloud server 110 can be indirectly connected to the computer device 120 through other computer devices, or the cloud server 110 can be directly connected to the computer device 120 without the need for other computer devices; those skilled in this art should flexibly choose according to their needs.

In practice, for example, computer device 120 can refer to the main computer equipment of the convenience store headquarters and/or the computer equipment of the back-end of each convenience store. Those familiar with this technology should choose flexibly according to the needs at the time.

In practice, for example, a cloud server 110 can be considered a server. Regarding servers, many technologies have been developed or are under development to manage the operation of computer servers, generally providing accessibility, consistency, and efficiency. Remote management allows for the removal of server input/output interfaces (e.g., monitors, mice, keyboards, etc.) and the need for network administrators to physically access each server. For instance, large data centers containing many computer servers typically use various remote management tools to manage, configure, monitor, and debug server hardware and software.

It should be understood that the terms “approximately,” “about,” or “roughly” used in this document are intended to modify any quantity that may vary slightly, but such slight variations do not alter its nature. Unless otherwise specified in the implementation, it means that the error range of the values modified by “approximately,” “about,” or “roughly” is generally permissible within 20 percent, preferably within 10 percent, and even more preferably within 5 percent.

When the automated suggested ordering system 100 is operating, the cloud server 110 trains and tests multiple different machine learning models using historical sales data and external factor data. Based on this, it selects the most accurate machine learning model from among the multiple models, and then generates a prediction result using the most accurate model. The computer device 120 obtains the prediction result from the cloud server 110 and calculates the suggested order quantity information based on the prediction result and the product inventory quantity. In this way, the automated suggested ordering system 100 uses artificial intelligence machine learning to build a computational model, automatically predict future sales, and calculate the optimal suggested order quantity, significantly reducing the disadvantages of purely manual ordering.

To further illustrate the architecture of the cloud server 110, please refer to Figure 1. As shown in Figure 1, the cloud server 110 includes a storage device 111, a processor 112, and a communication device 113. Architecturally, the communication device 113 is electrically connected to the processor 112, and the processor 112 is electrically connected to the storage device 111. It should be understood that, in the embodiments and scope of the patent application, the description involving "electrical connection" can generally refer to one component being indirectly electrically coupled to another component through other components, or one component being directly electrically connected to another component without needing to go through other components. For example, storage device 111 may be a built-in data storage device directly electrically connected to processor 112, or storage device 111 may be an external data storage device indirectly connected to processor 112 via communication device 113.

For example, processor 112 can be a central processing unit, and communication device 113 can be a network communication device (such as a network card) or a communication connection device. The storage device 111 described above can be implemented as different data storage devices or as the same data storage device, such as a computer hard drive, external server, flash memory, external hard drive, USB flash drive, or other computer-readable recording media, etc. Furthermore, those skilled in the art will understand that integrating multiple data storage devices into one storage device 111, or transferring other data content to storage device 111, still falls under the embodiments of this invention.

In use, storage device 111 stores historical sales and external factor data, along with multiple different machine learning models. Processor 112 uses the historical sales data and external factor data to train and test the multiple different machine learning models, thereby selecting the most accurate prediction model from among them, and then generating prediction results using the most accurate prediction model. Communication device 113 transmits the prediction results to computer device 120.

In practice, for example, historical sales data and external factor data can include ordering cycles, dynamic inventory, logistics and delivery time data, product matching files, ordering and receiving, inventory quantity, county/city weather data, sales quantity, store manager orders, store master files, temporary closures, product master files, product activities, early arrival of goods, quantity of discarded items, promotional details, etc.

Correspondingly, in practice, for example, the processing performed by processor 112 may optionally include: deriving the arrival time of each product at each store based on the daily logistics arrival time data; generating a configuration for preparing inventory one day in advance and writing it into a database table, with the system defaulting to preparing inventory one day in advance (e.g., if a store has a special need for advance inventory preparation for certain 3-code product groups, the configuration can be modified by the back-end system maintenance personnel); setting the data time interval used for modeling based on the current date (e.g., the past 24 months), and providing data within the set interval. Relevant holiday and group-level promotional information is used for subsequent model training and prediction; identification codes are generated based on store and 3-code product group codes to configure parallel computing and accelerate the model training and prediction process; the sales data of individual products in the last 60 days are used to calculate the sales ratio of individual products to their respective product groups. For example, the sales ratio of individual products will be used to allocate group-level predictions to individual products; product codes belonging to new products are extracted. During the testing period, after new products arrive in stores, stores will independently reorder for 14 days, and then the model will take over the ordering; the sales ratio of each store for each product group is calculated. The safety stock coefficient will be set differently based on the sales proportion of each product group, ensuring that service levels are proportional to sales. Based on product ordering cycle data, a list of products available for ordering from each store on a specified ordering date will be selected, and the model's group-level predictions will be allocated to these available products. Store-specific data will be extracted for model training and prediction; this data includes temperature, probability of rainfall, and temporary business closures. The latest on-site inventory data will be extracted for safety stock settings and on-site inventory adjustment calculations. The latest imported sales data will be used to measure the model's prediction accuracy. The difference between model predictions and actual sales will guide the selection of the safety stock coefficient.

In one embodiment of the present invention, storage device 111 stores information on future external factors. Processor 112 adjusts forecast results based on the information on future external factors, and communication device 113 transmits the forecast results to computer device 120. In practice, for example, processor 112 can obtain future weather forecasts from an external website through communication device 113 to incorporate the information on future external factors stored in storage device 111. Processor 112 flexibly adjusts forecast results based on the relationship between historical external factors and product sales.

To further illustrate the architecture of the computer device 120, please continue referring to Figure 1. As shown in Figure 1, the computer device 120 includes a network device 123, a storage device 121, a processing device 122, a display device 125, and an input device 126. Architecturally, the display device 125 is electrically connected to the processing device 122, the processing device 122 is electrically connected to the network device 123, the storage device 121, and the input device 126, and the network device 123 is connected to the cloud server 110. For example, storage device 121 may be a hard drive, flash storage device or other storage medium, processing device 122 may be a central processing unit, network device 123 may be a network communication device (such as a network card), display device 125 may be a built-in display or an external monitor, and input device 126 may be a keyboard, mouse, etc.

In use, network device 123 receives prediction results from cloud server 110. Storage device 121 stores the product inventory quantity. Processing device 122 calculates the suggested order quantity information based on the prediction results and product inventory quantity (e.g., the quantity obtained by subtracting the product inventory quantity from the predicted quantity). Display device 125 displays the suggested order quantity information for store managers to view as a basis for actual product order quantity. After the store manager inputs the actual product order quantity through input device 126, processing device 122 sends the product inventory quantity, suggested order quantity information, and corresponding actual order quantity back to cloud server 110 through network device 123. In this way, the cloud server 110 can use the product inventory quantity, product suggested order quantity information and/or product actual order quantity to perform transfer learning on the above multiple different machine learning models to enhance the accuracy of subsequent predictions.

In one embodiment of the invention, a plurality of different machine learning models include an exponential smoothing model. The processor 112 of the cloud server 110 uses the exponential smoothing model to perform differentiated weighted averaging on historical sales data of less than a preset number of days (e.g., about 180 days) according to holidays, summer and winter vacations and different weeks to produce a prediction. In practice, for example, the exponential smoothing model algorithm produces predictions by weighting historical data. In practical applications, in order to differentiate the predictions for holidays, summer and winter vacations, and different days of the week, the following processing methods are used in selecting the data for weighting the average: holiday predictions are based on past holiday sales; make-up work predictions are based on past non-weekend, non-holiday sales; predictions for a specific day of the week are based on past sales of the same week; at the same time, in order to differentiate the treatment of summer and winter vacations and seasonal sales, the following three different methods are used to select data of the same week. 1. Historical sales data for the same week; 2. Sales forecasts based on whether the delivery date falls during summer or winter vacation, selecting the corresponding historical sales data for the same week (whether it falls during summer or winter vacation); 3. Historical sales data for the same week of the same month (to obtain seasonal sales differences). Additionally, the number of historical sales data points less than the preset number of days is an adjustable parameter, for example, it can be set to use the past 6 sales volumes for a weighted average.

In one embodiment of the invention, a plurality of different machine learning models include a Generalized Additive Model (GAM). The processor 112 of the cloud server 110 uses the GAM to decompose historical sales data and external factor data for more than a preset number of days (e.g., approximately 180 days) into trend modules, seasonal modules, and external factor modules in a time series manner. The outputs of the trend module, seasonal module, and external factor module are then superimposed or multiplied to produce a prediction. In practice, for example, the algorithm of the GAM decomposes the time series model into trend, seasonal, and external factor modules for separate modeling. The final model prediction is obtained by superimposing or multiplying the outputs of each module. External factors used in GAM modeling may include: summer and winter vacations, general holidays, major holidays, average temperature, probability of rainfall, and number of promotional items within the group. These external factors and seasonal effects are multiplied together with the output of the trend module, and the final prediction is obtained by the following calculation: Model prediction = Trend module × (1 + Weekday module + Month module + Temperature module + Rainfall module + Promotional module + Public holiday module + Major holiday module + Summer and winter vacation module).

In summary, the cloud server 110 can simultaneously employ generalized additive models suitable for relatively long-term analysis and exponentially smoothed models suitable for relatively short-term analysis to perform computationally comprehensive machine learning. The cloud server 110 splits historical sales data and external factor data into training data and test data. Using the training data, the cloud server 110 builds multiple different machine learning models (e.g., generalized additive models, exponentially smoothed models, and/or other models) for each store and each product category. After the machine learning models are trained, the cloud server 110 uses test data to measure the accuracy of each model, and selects the most accurate machine learning model from the multiple different models. The cloud server 110 generates prediction results by using the most accurate machine learning model and adjusts the prediction results based on information about future external factors.

To further explain the method performed by the automated suggestion ordering system 100 described above, please refer to Figures 1 and 2. Figure 2 is a flowchart of an automated suggestion ordering method 200 according to an embodiment of the present invention. As shown in Figure 2, the automated suggestion ordering method 200 includes steps S201 to S202 (it should be understood that, unless otherwise specified, the order of the steps mentioned in this embodiment can be adjusted according to actual needs, and they can even be performed simultaneously or partially simultaneously).

The automated recommended ordering method 200 may take the form of a computer program product on a non-transient computer-readable recording medium having a plurality of computer-readable instructions contained in the medium. Suitable recording media may include any of the following: non-volatile memory, such as read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electronically eraseable programmable read-only memory (EEPROM); volatile memory, such as static memory (SRAM), dynamic memory (SRAM), double data rate random access memory (DDR-RAM); optical storage devices, such as read-only optical discs (CD-ROM), read-only digital multifunction discs (DVD-ROM); magnetic storage devices, such as hard disk drives and floppy disk drives.

In step S201, multiple different machine learning models are trained and tested using historical sales data and external factor data through cloud server 110. The most accurate machine learning model is then selected from these models to generate a prediction result. In step S202, the prediction result is obtained from the cloud server via computer device 120. Based on the prediction result and the product inventory quantity, the recommended order quantity information is calculated.

In one embodiment of the present invention, the automated suggested ordering method 200 further includes: obtaining future external factor information through cloud server 110; adjusting the prediction results based on the future external factor information through cloud server 110; and transmitting the prediction results to computer device 120 through cloud server 110.

In one embodiment of the invention, the plurality of different machine learning models include an exponential smoothing model, and the automated suggested ordering method 200 further includes: using the exponential smoothing model to perform differentiated weighted averaging processing on historical sales data of fewer than a preset number of days by holidays, summer and winter vacations and different weeks through a cloud server 110 to generate a prediction.

In one embodiment of the present invention, a plurality of different machine learning models include generalized additive models, and the automated suggested ordering method 200 further includes: using a cloud server 110 to decompose historical sales data and external factor data of more than a preset number of days into trend modules, seasonal modules and external factor modules in a time series manner, and then superimposing or multiplying the outputs of trend modules, seasonal modules and external factor modules to produce a prediction.

In one embodiment of the present invention, the automated suggested ordering method 200 further includes: displaying suggested order quantity information of goods through computer device 120; and transmitting the suggested order quantity information of goods and the corresponding actual order quantity of goods back to cloud server 110 through computer device 120.

In summary, the technical solution of this invention has significant advantages and beneficial effects compared with the prior art. Through the automated suggested ordering system 100 and automated suggested ordering method 200 of this invention, an algorithmic model is established using artificial intelligence machine learning functions to automatically predict future sales and calculate the optimal suggested order quantity, significantly reducing the disadvantages of manual ordering.

Although the invention has been disclosed above by way of implementation, it is not intended to limit the invention. Anyone skilled in the art may make various modifications and alterations without departing from the spirit and scope of the invention. Therefore, the scope of protection of the invention shall be determined by the appended patent application.

To make the above and other objects, features, advantages and embodiments of the present invention more apparent and understandable, the symbols are explained as follows: 100: Automation suggestion ordering system; 110: Cloud server; 111: Storage device; 112: Processor; 113: Communication device; 120: Computer device; 121: Storage device; 122: Processing device; 123: Network device; 125: Display device; 126: Input device; 200: Automation suggestion ordering method; S201-S202: Steps

To make the above and other objects, features, advantages and embodiments of the present invention more apparent, the accompanying drawings are explained as follows: Figure 1 is a block diagram of an automated suggestion ordering system according to an embodiment of the present invention; and Figure 2 is a flowchart of an automated suggestion ordering method according to an embodiment of the present invention.

100: Automated Recommended Ordering System

110: Cloud Server

111: Storage Equipment

112: Processor

113: Communication Equipment

120: Computer Device

121: Storage Device

122: Processing device

123: Network Devices

125: Display Device

126: Input Device

Claims (9)

An automated order placement system includes: a cloud server that trains and tests multiple different machine learning models using historical sales data and external factor data, selects the most accurate machine learning model from these models, and generates a prediction result using the most accurate model. The cloud server includes a processor, and the different machine learning models include a generalized additive model. The processor uses the generalized additive model to decompose the historical sales data and external factor data (for more than a preset number of days) into a trend module, a seasonal module, and an external factor module in a time series manner. The outputs of the trend module, the seasonal module, and the external factor module are then superimposed or multiplied to produce a prediction. The seasonality module and the external factors module include a weekday module, a month module, a temperature module, a rainfall module, a promotion module, a public holiday module, a major holiday module, and a summer/winter break module. The processor, using the generalized additive model, generates a prediction as: the output of the trend module × (1 + the output of the weekday module + the output of the month module + the output of the temperature module + the output of the rainfall module + the output of the promotion module + the output of the public holiday module + the output of the major holiday module + the output of the summer/winter break module). A computer device is connected to the cloud server. The computer device obtains the prediction results from the cloud server and then calculates a suggested order quantity for a product based on the prediction results and a product inventory quantity. The automated ordering system as described in claim 1, wherein the cloud server comprises: a storage device storing historical sales and external factor data, the various machine learning models, and future external factor information, wherein a processor is electrically connected to the storage device and adjusts the prediction results based on the future external factor information; and a communication device electrically connected to the processor, which transmits the prediction results to the computer device. The automated suggested ordering system as described in claim 2, wherein the processor is a central processing unit. The automated order suggestion system described in claim 1, wherein the different machine learning models include an exponential smoothing model, and the processor uses the exponential smoothing model to perform differentiated weighted averaging of historical sales data for fewer than the preset number of days, categorized by holidays, summer/winter breaks, and different weeks, to generate a forecast. The automated suggested ordering system as described in claim 1, wherein the computer device comprises: a network device connected to the cloud server, the network device receiving the prediction result from the cloud server; a storage device storing the inventory quantity of the goods; a processing device electrically connected to the network device and the storage device, the processing device calculating the suggested order quantity information of the goods based on the prediction result and the inventory quantity of the goods; and a display device electrically connected to the processing device, the display device displaying the suggested order quantity information of the goods, the processing device transmitting the suggested order quantity information of the goods and a corresponding actual order quantity of a goods back to the cloud server through the network device. An automated suggested ordering method includes the following steps: training and testing multiple different machine learning models using historical sales data and external factor data on a cloud server; selecting the most accurate machine learning model from these models; generating a prediction result using the most accurate model; and retrieving the prediction result from the cloud server using a computer device; and calculating a suggested order quantity based on the prediction result and a product inventory quantity. The different machine learning models include a generalized additive model. The automated suggested ordering method further includes: using the generalized additive model on the cloud server to process more than a preset number of historical sales data and external factor data... The time series is decomposed into a trend module, a seasonal module, and an external factor module. The outputs of these modules are then superimposed or multiplied to generate a forecast. The seasonal and external factor modules include a week-by-week module, a month-by-month module, a temperature module, a rainfall module, a promotion module, a public holiday module, a major holiday module, and a summer/winter break module. The forecast generated by the cloud server using this generalized additive model is: (Output of the trend module × (1 + Output of the week-by-week module + Output of the month-by-month module + Output of the temperature module + Output of the rainfall module + Output of the promotion module + Output of the public holiday module + Output of the major holiday module + Output of the summer/winter break module)). The automated suggested ordering method as described in claim 6 further includes: obtaining information about future external factors through the cloud server; adjusting the prediction result based on the future external factor information through the cloud server; and transmitting the prediction result to the computer device through the cloud server. The automated ordering method described in claim 6, wherein the different machine learning models include an exponential smoothing model, further comprises: using the exponential smoothing model on a cloud server to perform differentiated weighted averaging of historical sales data for fewer than the preset number of days, categorized by holidays, summer/winter breaks, and different weeks, to generate a forecast. The automated suggested ordering method as described in claim 6 further includes: displaying the suggested order quantity information of the product through the computer device; and transmitting the suggested order quantity information of the product and the corresponding actual order quantity of a product back to the cloud server through the computer device.