Information processing device, control method for information processing device, and control program for information processing device

Abstract

To improve the accuracy of predicting the remaining amount of gas in a gas container. [Solution] An information processing device in a delivery system for delivering gas to consumers includes an acquisition unit that acquires information regarding the consumer's gas usage during past meter reading dates; a prediction unit that calculates the predicted gas usage from a specified meter reading date to a target prediction date using a first prediction model obtained by machine learning information regarding gas usage during past meter reading dates over a specified period; a determination unit that determines the reliability of the predicted gas usage according to the first prediction model by comparing it with the true gas usage according to a probability model based on the gas usage during past meter reading dates; and a calculation unit that, if the determination unit determines that the reliability is low, calculates the predicted gas usage using a second prediction model based on the gas usage during past meter reading dates, which is the gas usage during a specified period in the past corresponding to the specified period, and the gas usage during multiple meter reading dates including the specified meter reading date.

Description

[Technical Field]

[0001] The present invention relates to an information processing device, a control method for an information processing device, and a control program for an information processing device. [Background technology]

[0002] Conventionally, when delivering gas cylinders to consumers, gas cylinders to be replaced are identified based on gas meter reading data, and a delivery route is determined. For example, the applicant has provided a delivery system that calculates a predicted future gas consumption amount from the consumer's past gas consumption amount, and extracts consumers with a high priority for gas delivery (e.g., Patent Documents 1 and 2). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7437358 [Patent Document 2] Patent No. 7437542 Summary of the Invention [Means for solving the problem]

[0004] According to one embodiment of the present invention, an information processing device in a delivery system for delivering gas to consumers comprises an acquisition unit that acquires information regarding the consumer's gas usage during past meter reading dates; a prediction unit that calculates the predicted gas usage from a specified meter reading date to a target prediction date using a first prediction model obtained by machine learning information regarding the gas usage during past meter reading dates over a specified period; a determination unit that determines the reliability of the predicted gas usage based on the first prediction model by comparing it with the true gas usage based on a probability model based on the gas usage during past meter reading dates; and a calculation unit that, if the determination unit determines that the reliability is low, calculates the predicted gas usage using a second prediction model based on the gas usage during past meter reading dates, which includes the gas usage during a specified period in the past corresponding to the specified period, and the gas usage during multiple meter reading dates including the specified meter reading date.

[0005] In an information processing device according to one embodiment of the present invention, the judgment unit may judge that the reliability is low when a predetermined condition indicating a large deviation between the predicted gas usage based on the first prediction model and the true gas usage based on the probabilistic model is met.

[0006] In an information processing device according to one embodiment of the present invention, the probabilistic model is defined from the average and variance of the gas usage for a specified period in the past corresponding to the specified period and the gas usage for multiple meter reading days including the specified meter reading day, and as a specified condition, if the predicted gas usage by the first prediction model does not fall within a specified range determined by the average and variance, it may be determined that the reliability is low.

[0007] In the information processing device according to one embodiment of the present invention, the determination unit may change the size of the predetermined range based on the accuracy of the first prediction model obtained during machine learning.

[0008] In the information processing device according to one embodiment of the present invention, the determination unit may increase the predetermined range when the accuracy of the first prediction model is high, and may decrease the predetermined range when the accuracy of the first prediction model is low.

[0009] In an information processing device according to one embodiment of the present invention, the calculation unit may calculate the predicted gas usage based on the second prediction model as the average of the gas usage for a specified period in the past corresponding to the specified period and the gas usage for multiple meter reading days including the specified meter reading day.

[0010] In one embodiment of the present invention, a control method of an information processing device in a delivery system for delivering gas to consumers includes the steps of: the information processing device acquiring information regarding the consumer's gas usage for past meter reading dates; calculating a predicted gas usage from a specified meter reading date to a target prediction date using a first prediction model obtained by machine learning information regarding gas usage for past meter reading dates over a specified period; determining the reliability of the predicted gas usage based on the first prediction model by comparing it with the true gas usage based on a probability model based on the gas usage for past meter reading dates; and, if the reliability is determined to be low in the determining step, calculating the predicted gas usage using a second prediction model based on the gas usage for a specified period in the past corresponding to the specified period, among multiple usages for past meter reading dates, and the gas usage for multiple meter reading dates including the specified meter reading date.

[0011] A control program of an information processing device in a delivery system for delivering gas to consumers according to one embodiment of the present invention enables the information processing device to implement the following functions: acquire information regarding the consumer's gas usage during past meter reading dates; calculate a predicted gas usage from a specified meter reading date to a target prediction date using a first prediction model obtained by machine learning information regarding gas usage during past meter reading dates over a specified period; determine the reliability of the predicted gas usage according to the first prediction model by comparing it with the true gas usage according to a probability model based on the gas usage during past meter reading dates; and, if the reliability is determined to be low by the determination function, calculate the predicted gas usage using a second prediction model based on the gas usage during a specified period in the past corresponding to the specified period, among multiple usage amounts during past meter reading dates, and the gas usage during multiple meter reading dates including the specified meter reading date. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a schematic diagram of a delivery system configuration according to one embodiment of the present invention. [Figure 2] FIG. 2 is an example of a data table. [Figure 3] 3(a) and (b) are examples of data tables. [Figure 4] FIG. 4 is a schematic diagram illustrating the creation of explanatory variables (features) used for learning when generating a prediction model. [Figure 5] FIG. 5 is a flowchart of an example of a prediction process according to an embodiment of the present invention. [Figure 6] FIG. 6 shows an example of the hardware configuration of a computer that can realize a server according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, an embodiment of the invention according to the present disclosure (also referred to as the present invention) will be described using the drawings. Note that the drawings are merely examples, and the present invention is not limited to those shown in the drawings. For example, the illustrated delivery system, configuration diagram of the information processing device (server), meter, communication device, number of gas containers, data set (table), and flowchart are merely examples, and the present invention is not limited to these.

[0014] As described above, the applicant provides a machine learning-based gas remaining quantity prediction model for efficient gas delivery in a gas supply service in which gas containers are replaced according to the amount of gas used by a consumer. Efficient delivery of gas containers requires accurate prediction of the consumer's remaining gas quantity. Currently, so-called "smart meters" equipped with communication devices are becoming more widespread for meters that measure gas usage, and meter readings are frequently transmitted to a designated management server. This allows for highly accurate learning when generating a gas remaining quantity prediction model. However, there are also consumers (hereinafter referred to as "non-equipped consumers") whose meters do not have communication devices and require regular meter readings by meter readers. The gas usage of non-equipped consumers is obtained by meter readers directly checking the meter readings of the meter connected to the gas container. For non-equipped consumers, meter readings by meter readers occur, for example, once a month, which limits the amount of data available for learning, potentially resulting in low accuracy of the learning. According to one embodiment of the present invention, for a gas container whose remaining gas amount is measured by a meter that does not have a communication device and requires a meter reader (hereinafter also referred to as a "non-smart meter"), the accuracy of a machine learning prediction model for the remaining gas amount is determined, and the remaining gas amount prediction method is changed depending on the determination result, making it possible to maintain the accuracy of the prediction.

[0015] <System configuration> FIG. 1 is a diagram illustrating an example of the configuration of a delivery system according to one embodiment of the present invention. The delivery system 800 may include a server (processing server) 100, a management database (DB) 210, a provider DB 220, a communication device 310 installed on the meter 320 of the gas containers 330A and 330B, a meter 420 of the gas containers 430A and 430B, and a communication terminal 441 of a meter reader 440 who checks the meter readings of the meter 420. Note that when there is no need to distinguish between consumers with smart meters installed (hereinafter also referred to as "customers with smart meters") and consumers without smart meters, they may be collectively referred to as "consumers." Note that one embodiment of the present invention relates to predicting the remaining gas volume of consumers without smart meters, and therefore gas prediction for consumers with smart meters will not be described in detail here. However, FIG. 1 also illustrates consumers with smart meters installed to clearly illustrate the overview of the delivery system. In addition, in Figure 1, both the installed consumer and the non-installed consumer are shown as one detached house, but the consumer is not limited to this and may be an apartment building, a corporation, etc., and there may be multiple installed consumers and multiple non-installed consumers.

[0016] The server 100 may be a server that executes various types of arithmetic processing in the delivery system 800. Although only one server 100 is shown in FIG. 1, this is not limiting. That is, each function described as being provided by the server 100 may be realized by multiple servers. The server 100 may also be, for example, a distributed server system that cooperates by communicating via a network, or may be a so-called cloud server. That is, the server 100 is not limited to a physical server, and may also include a software-based virtual server or edge server.

[0017] The server 100 and the communication devices 310, 441 may be connected via a network 600. The communication method in the network 600 may be, for example, LTE, LTE-Advanced, fourth generation communication (4G), fifth generation communication (5G), sixth generation communication (6G) or later communication methods, CDMA, etc. Also, for example, the communication method may be a wireless communication method for IoT, such as Category M, Category M1, or NB-IoT (Narrow Band IoT), which is an extension of LTE. Note that the communication method is not limited to these examples.

[0018] The management DB 210 may include a meter reading data DB 212 that stores (stores) meter reading data from each meter, and a prediction model DB 213 that stores various prediction models. The provider DB 220 is a database related to providers that provide gas delivery services, and may include a consumer information DB 221 related to each consumer. While FIG. 1 illustrates a configuration in which the management DB 210 and the provider DB 220 are connected to the server 100 via a network 600, this is not limiting. For example, the management DB 210 and the provider DB 220 may be connected to the server 100 via a network separate from the network 600. Furthermore, the information stored in the management DB 210 and the provider DB 220 may be stored in a storage unit of the server 100.

[0019] FIG. 2 shows an example of consumer information stored in the consumer information DB 221. The consumer information table TB1 may store a consumer ID (Identifier) ​​(an example of an identifier) ​​that uniquely identifies each consumer, associated with a meter ID, capacity information, etc. The meter ID may be identification information that uniquely identifies a meter. The capacity information may be information that indicates the total capacity of the gas containers supplied to the consumer. Note that any type of identifier may be used as long as it uniquely identifies the object to which it is assigned. For example, referring to the consumer data table TB1, a consumer with consumer ID "CS0001" has a meter with meter ID "mt0001" installed, the gas container has a capacity of "50 L," and there are "two" gas containers consisting of a main container and a spare container, with a total capacity of 100 L. Furthermore, a consumer with consumer ID "NCS0001" is a consumer without a meter installed, and a meter ID may not be associated with the meter.

[0020] Note that table TB1 in FIG. 2 is an example and is not limited to this. For example, the capacity information is not limited to the format shown in the figure, as long as it is information that can determine the total capacity of the gas container. Furthermore, customer data may be stored in multiple tables, distinguishing between installed customers and non-installed customers. Furthermore, customer data table TB1 may store, for installed customers, the identifier of communication device 310, the installation date of meters 320, 420, information about the gas delivery service provider (the service provider used by the customer), information about the delivery destination, contact information for the customer, and other data.

[0021] FIG. 3 shows an example of information related to gas usage. Table TB2 in FIG. 3(a) may be a table related to gas usage by installed consumers, and table TB3 in FIG. 3(b) may be a table related to gas usage by non-installed consumers. Information indicating whether or not a gas container has been replaced may also be stored. Each table may be stored in the meter reading data DB 212.

[0022] Next, we will explain how to generate a prediction model for gas usage by non-installed consumers using machine learning, using Figure 4. Note that from here on, the term "prediction model" refers to a prediction model for gas usage by non-installed consumers, and does not include installed consumers.

[0023] Figure 4 may be an example of features used in machine learning when generating a prediction model through machine learning. In Figure 4, the circles represent meter reading data acquired when a meter reader visits, and may be, for example, actual monthly gas usage. In one embodiment of the present invention, the explanatory variables (features) used to generate a prediction model through machine learning may be 19-dimensional. Specifically, the number of days from the second most recent meter reading to the one immediately preceding meter reading is defined as "interval a," the amount of gas usage from the second most recent meter reading to the one immediately preceding meter reading is defined as "change a," the number of days from the one immediately preceding meter reading to the most recent meter reading is defined as "interval b," the amount of gas usage from the one immediately preceding meter reading to the most recent meter reading is defined as "change b," the number of days from the most recent meter reading to the prediction target date is defined as "interval c," and a 12-dimensional one-hot vector representing the month of the prediction target is defined as "prediction target month," where the features (explanatory variables) may be "interval a," "change a," "change a / interval a," "interval b," "change b," "change b / interval b," "interval c," and "month of the prediction target date," as shown in Figure 4. Then, a prediction model may be generated by learning using a past dataset, with "change c / interval c," i.e., daily usage (hereinafter referred to as "daily usage"), as the objective variable.

[0024] The prediction model may be generated by an algorithm using at least one of Linear Regression, Support Vector Regression, Random Forest Regression, Gradient Boosting Regression, Gradient Boosting Decision Tree, and LightGBM (Light Gradient Boosting Machine).

[0025] Note that the above-described machine learning prediction model may have low prediction accuracy for some meters. In this case, it is desirable to use a rule-based prediction model that follows preset rules rather than a machine learning prediction model. Therefore, according to one embodiment of the present invention, the accuracy (reliability) of the machine learning prediction model may be determined in advance, and if the accuracy is determined to be low, gas usage may be predicted using a rule-based prediction model. Specifically, the distribution of true gas usage based on a probabilistic model may be compared with the predicted gas usage based on machine learning, and the validity of the output of the machine learning prediction model (i.e., the predicted gas usage based on machine learning) may be determined based on the probability that the predicted value based on machine learning occurs within the distribution of true gas usage. Note that the true usage based on the probabilistic model may refer to the true value of actual gas usage, i.e., the correct data itself that will be observed in the future. Note that, hereinafter, the predicted gas usage based on the machine learning prediction model and the rule-based prediction model, as well as the true gas usage based on the probabilistic model, may be daily gas usage. This is explained below.

[0026] First, if the true gas usage according to the probabilistic model is y, then y can be expressed by the following equation:

[0027]

number

[0028] Note that α1, α2,…, α n Specifically, n>=3 and may be the daily usage calculated from the meter reading two readings before, the daily usage calculated from the meter reading one reading before, and the daily usage for the same month of the previous year for the month to be predicted.

[0029] If the predicted gas usage amount predicted using the machine learning prediction model does not satisfy the following formula, it may be determined that the output of the machine learning prediction model is abnormal.

[0030]

number

[0031] If the difference between the predicted value (y) of the machine learning model and the portion of the probabilistic model explained by past daily usage is outside the range of ±kσ, it means that the probability of obtaining a predicted value by the machine learning model is very low, considering the distribution of true gas usage by the probabilistic model. In other words, it means that there is a large discrepancy between the predicted usage by machine learning and the true gas usage by the probabilistic model. Therefore, if formula (2) is not satisfied, it can be determined that the accuracy (reliability) of the machine learning prediction model is low.

[0032] In one embodiment of the present invention, if the accuracy of the machine learning prediction model is low, the gas usage may be predicted by a rule-based prediction model. The rule-based prediction model uses past daily usage (α1, α2, ..., α n ), for example, the past daily usage (α1, α2, ..., α n ) More specifically, the rule-based prediction model may be calculated using the average of the daily usage calculated from the two most recent meter readings, the daily usage calculated from the one most recent meter reading, and the daily usage for the same month of the previous year for the month to be predicted. The value of n may be determined according to the number of meter reading days in the month to be predicted; if there are four meter reading days, data for four reading days may be used. In other words, the true gas usage calculated by the probabilistic model may be expressed using the predicted usage of the rule-based prediction model and an error term.

[0033] In addition, the past daily usage (α1, α2, ..., α n ) is almost constant, it is highly appropriate to estimate gas consumption by the average of past daily consumption. n) is almost constant, σ in equation (2) becomes infinitesimally small. In other words, there is a high probability that a rule-based prediction model will be selected to predict gas usage. Conversely, if the past daily usage (α1, α2, ..., α n ) fluctuates greatly, it is not appropriate to estimate gas consumption by the average of past daily consumption. n ) fluctuates significantly, σ in equation (2) becomes large. In other words, there is a high probability that a machine learning prediction model will be selected to predict gas usage.

[0034] Furthermore, according to one embodiment of the present invention, the selected prediction model may be tuned by the value of the hyperparameter k in Equation (1). Specifically, the larger the value of k, the more likely it is that a prediction value from a machine learning prediction model will be selected, and the smaller the value of k, the more likely it is that a prediction value from a rule-based prediction model will be selected. For example, when Equation (2) is not satisfied when k = 2, the probability that the deviation between the average true gas usage obtained from the probabilistic model and the predicted value from the machine learning model will be greater than the distribution of true usage represented by the probabilistic model is less than approximately 5%. Therefore, for example, the selected prediction model can be adjusted based on evaluation values ​​during learning, etc., by setting the value of k to a larger value if the accuracy of the machine learning prediction model is high, or setting the value of k to a smaller value if the accuracy is low.

[0035] As described above, according to one embodiment of the present invention, a rule-based prediction model is appropriately selected depending on the accuracy of a machine learning prediction model, the accuracy of which varies depending on the features used for learning, making it possible to output a correct prediction value.

[0036] <Functional configuration> 1, the following describes the functions of each functional unit of the server (processing server) 100. The server 100 may include a communication unit 110, a prediction unit 120, a determination unit 130, and a calculation unit 140.

[0037] The communication unit 110 may function as an acquisition unit that acquires information regarding the amount of gas used by the consumer. The information regarding the amount of gas used by the consumer is information regarding the amount of gas used by each consumer during the past meter reading dates, transmitted from the communication device 310 or the communication terminal 441 of the meter reader 440, and may be the meter reading value stored in the meter reading data DB 212 described above.

[0038] The prediction unit 120 may calculate the predicted gas usage from a predetermined meter reading date to a prediction target date using a first prediction model obtained by machine learning information on gas usage between past meter reading dates. The predicted usage may be daily usage. The first prediction model is as described with reference to FIG. 4. The predetermined meter reading date may be, for example, the most recent meter reading date.

[0039] The determining unit 130 may determine the reliability of the predicted gas usage amount according to the first prediction model by comparing it with the true gas usage amount according to a probability model based on the gas usage amount during past meter reading dates. Note that, if a predetermined condition indicating a large deviation between the predicted gas usage amount according to the first prediction model and the true gas usage amount according to the probability model is met, the determining unit 130 may determine that the reliability of the predicted gas usage amount according to the first prediction model is low.

[0040] Here, the above-mentioned probabilistic model may be defined by the average and variance of the gas usage amount for a predetermined period in the past corresponding to the predetermined period and the gas usage amount for multiple meter reading dates including the predetermined meter reading date. Here, the "predetermined period" may be the period during which data used for learning was acquired when generating a prediction model using machine learning. In the above example, the predetermined period may be one month of the target prediction month. The "predetermined past period corresponding to the predetermined period" may be one month of the same month of the previous year of the target prediction month. The "predetermined meter reading date" may be the most recent (latest) meter reading date. The reliability may be determined to be low if the predicted gas usage amount by the first prediction model does not fall within a predetermined range determined by the average and variance as a predetermined condition. That is, the predetermined condition may be the above-mentioned formula (2), and the predetermined range may be ±kσ in formula (2).

[0041] If the determination unit 130 determines that the reliability is low, the calculation unit 140 may calculate the predicted gas usage using a second prediction model based on the gas usage for a predetermined period in the past corresponding to the predetermined period among the gas usage for past meter reading dates and the gas usage for multiple meter reading dates including the predetermined meter reading date. That is, the second prediction model may refer to the rule-based prediction model described above. Specifically, the calculation unit 140 may calculate the predicted gas usage using the second prediction model as the average of the gas usage for a predetermined period in the past corresponding to the predetermined period and the gas usage for multiple meter reading dates including the predetermined meter reading date. That is, as described above, the calculation unit 140 may calculate the gas usage using the rule-based prediction model by averaging the daily usage calculated from the two most recent meter readings, the daily usage calculated from the most recent meter reading, and the average daily usage for the same month of the previous year for the month to be predicted.

[0042] Furthermore, the determination unit 130 may change the size of the predetermined range based on the accuracy of the first prediction model obtained during machine learning. That is, the value of the hyperparameter k may be adjusted depending on the accuracy of the machine learning. Specifically, the determination unit 130 may increase the predetermined range (increase k) when the accuracy of the first prediction model is high, and may decrease the predetermined range (decrease k) when the accuracy of the first prediction model is low.

[0043] <Prediction processing> Next, a prediction process according to an embodiment of the present invention will be described. FIG. 5 is a flowchart of the prediction process according to an embodiment of the present invention. First, the server 100 may acquire information about the consumer's gas usage during past meter reading dates (step S11). The server 100 may calculate a predicted gas usage from a predetermined meter reading date to a prediction target date using a first prediction model obtained by machine learning information about gas usage during past meter reading dates over a predetermined period (step S12). The server 100 may then determine the reliability of the predicted gas usage according to the first prediction model by comparing it with the true gas usage according to a probability model based on the gas usage during past meter reading dates (step S13). If the reliability is determined to be low (low in step S13), the server 100 may calculate the predicted gas usage using a second prediction model based on the gas usage during a predetermined period corresponding to the predetermined period among the gas usage during past meter reading dates and the gas usage during multiple meter reading dates including the predetermined meter reading date (step S14). Then, the predicted gas usage amount according to the second prediction model may be used as the predicted gas usage amount. Note that if the reliability is determined to be high in step S13 (high in step S13), the predicted gas usage amount according to the first prediction model may be output.

[0044] <Hardware configuration> The hardware configuration of the server 100 will be described. FIG. 6 shows an example of the hardware configuration of a computer capable of realizing the server 100 of this embodiment. The server 100 includes a processor 101, a storage 102, a memory 103, an input / output interface (input / output I / F) 104, and a communication interface (communication I / F) 105. The components are connected to each other via a bus B. The server 100 realizes the functions and methods described in this embodiment through cooperation of these components. For example, when each functional unit of the server 100 is realized by software, it is realized by the processor 101 executing instructions included in a program loaded from the storage 102 into the memory 103. That is, the server 100 according to this embodiment functions as a communication unit 110, a prediction unit 120, a determination unit 130, and a calculation unit 140 by the processor 101 executing the program loaded into the memory 103.

[0045] The processor 101 may include, for example, a central processing unit (CPU), a micro processing unit (MPU), a graphics processing unit (GPU), a microprocessor, a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc., and may be realized by a logic circuit (hardware) formed in an integrated circuit (an integrated circuit (IC) chip, a large scale integration (LSI)), etc., or a dedicated circuit. Note that the server 100 preferably has a processor 101 with high computing power for processing the above-mentioned large amounts of data.

[0046] The communication I / F 105 is implemented as hardware such as a network adapter, communication software, or a combination of these, and transmits and receives various data to and from external devices. The communication may be performed either wired or wirelessly, and any communication protocol may be used as long as mutual communication is possible.

[0047] The input / output I / F 104 includes an input device for inputting various operations to the server 100, and an output device for outputting processing results processed by the server 100. The input device includes, for example, a touch panel, a touch display, hardware keys such as a keyboard, a pointing device such as a mouse, a camera (for inputting operations via images), and a microphone (for inputting operations via voice). The output device outputs processing results processed by the processor 101. The output device includes, for example, a touch panel, a speaker, etc.

[0048] Although the present invention has been described based on the drawings and examples, it should be noted that those skilled in the art would easily be able to make various modifications and alterations based on this disclosure. Therefore, it should be noted that these modifications and alterations are within the scope of the present invention. For example, the functions included in each component, step, etc. can be rearranged so as not to cause logical inconsistencies, and multiple components, steps, etc. can be combined into one or divided. Furthermore, the configurations shown in the above embodiments may be combined as appropriate.

[0049] Although the above description has been given of an embodiment in which gas usage is predicted, the present invention is not limited to gas usage, and may also be used to predict usage of oil, water and sewerage, etc.

[0050] The program of each embodiment of the present disclosure may be provided in a state stored on a storage medium readable by an information processing device. The storage medium may store the program on a "non-transitory tangible medium." The storage medium may include, where appropriate, one or more semiconductor-based or other integrated circuits (ICs) (e.g., field programmable gate arrays (FPGAs), application-specific ICs (ASICs), etc.), hard disk drives (HDDs), hybrid hard drives (HHDs), optical disks, optical disk drives (ODDs), magneto-optical disks, magneto-optical drives, floppy diskettes, floppy disk drives (FDDs), magnetic tape, solid-state drives (SSDs), RAM drives, secure digital cards or drives, any other suitable storage media, or any suitable combination of two or more of these. The storage medium may, where appropriate, be volatile, non-volatile, or a combination of volatile and non-volatile.

[0051] Furthermore, the program of the present disclosure may be provided to the server 100 via any transmission medium (such as a communication network or broadcast waves) capable of transmitting the program.

[0052] Furthermore, each embodiment of the present disclosure may also be realized in the form of a data signal embedded in a carrier wave in which a program is embodied by electronic transmission.

[0053] The program of the present disclosure may be implemented using, for example, a scripting language such as JavaScript (registered trademark) or Python (registered trademark), C language, Go language, Swift (registered trademark), Koltin (registered trademark), Java (registered trademark), or the like.

[0054] According to each aspect of the present disclosure described above, gas usage can be accurately predicted, providing a more livable environment for people, thereby contributing to the achievement of Goal 11 of the Sustainable Development Goals (SDGs), "Sustainable cities and communities." [Explanation of symbols]

[0055] 100 servers (processing servers) 110 Communications Department 120 Prediction Department 130 Judgment section 140 Calculation Unit 210 Management DB (Database) 212 Meter Reading Data DB 213 Prediction Model DB 220 Business DB 221 Consumer information DB 310 Communication Equipment 320,420 meters 330A, 330B, 430A, 430B gas cylinders 440 Meter Reader 441 Communication Terminals 800 Delivery System

Claims

1. An information processing device in a gas delivery system for delivering gas to consumers, an acquisition unit that acquires information about the consumer's gas usage amount during past meter reading dates; a prediction unit that calculates a predicted gas usage amount from a predetermined meter reading date to a prediction target date after a predetermined period using a first prediction model obtained by machine learning of information about the gas usage amount during the past meter reading dates; a determination unit that determines the reliability of the predicted gas usage amount according to the first prediction model by comparing it with a true gas usage amount according to a probability model based on the gas usage amount during the past meter reading days; a calculation unit that, when the determination unit determines that the reliability is low, calculates the predicted gas usage amount using a second prediction model based on the gas usage amount for a past predetermined period corresponding to the predetermined period among the gas usage amounts for the past meter reading days and the gas usage amounts for a plurality of meter reading days including the predetermined meter reading day; An information processing device comprising:

2. the determination unit determines that the reliability is low when a predetermined condition indicating a large deviation between the predicted gas usage amount according to the first prediction model and the true gas usage amount according to the probabilistic model is satisfied. The information processing device according to claim 1 .

3. The probabilistic model is defined by the average and variance of the gas usage amount for the past predetermined period corresponding to the predetermined period and the gas usage amount for the plurality of meter reading days including the predetermined meter reading day, and the reliability is determined to be low when the predicted gas usage amount by the first prediction model is not located within a predetermined range determined by the average and the variance as the predetermined condition. The information processing device according to claim 2 .

4. the determination unit changes the size of the predetermined range based on accuracy of the first prediction model obtained during the machine learning. The information processing device according to claim 3 .

5. the determination unit increases the predetermined range when the accuracy of the first prediction model is high, and decreases the predetermined range when the accuracy of the first prediction model is low. The information processing device according to claim 4 .

6. The calculation unit calculates the predicted gas usage amount using the second prediction model as an average of the gas usage amount for the past predetermined period corresponding to the predetermined period and the gas usage amount for the plurality of meter reading days including the predetermined meter reading day. The information processing device according to claim 1 .

7. A method for controlling an information processing device in a gas distribution system for distributing gas to consumers, comprising: The information processing device acquiring information about the amount of gas used by the customer during past meter reading dates; calculating a predicted gas usage amount from a predetermined meter reading date to a prediction target date a predetermined period later using a first prediction model obtained by machine learning information on the gas usage amount during the past meter reading dates; determining the reliability of the predicted gas usage amount by the first prediction model by comparing it with a true gas usage amount by a probability model based on the gas usage amount during the past meter reading days; If the reliability is determined to be low in the determining step, calculating the predicted gas usage amount using a second prediction model based on the gas usage amount for a predetermined past period corresponding to the predetermined period among the multiple usage amounts for the past meter reading days and the gas usage amounts for multiple meter reading days including the predetermined meter reading day; A control method for an information processing device that executes the above.

8. A control program for an information processing device in a gas delivery system for delivering gas to consumers, In the information processing device, A function of acquiring information about the customer's gas usage during past meter reading dates; a function of calculating a predicted gas usage amount from a predetermined meter reading date to a prediction target date after a predetermined period using a first prediction model obtained by machine learning information on the gas usage amount during the past meter reading dates; a function of determining the reliability of the predicted gas usage amount by the first prediction model by comparing it with a true gas usage amount by a probability model based on the gas usage amount during the past meter reading days; a function of calculating, when the determining function determines that the reliability is low, the predicted gas usage amount using a second prediction model based on the gas usage amount for a predetermined past period corresponding to the predetermined period among the multiple usage amounts for the past meter reading days and the gas usage amounts for multiple meter reading days including the predetermined meter reading day; A control program for an information processing device that realizes the above.

Images