Information processing device, information processing system, and model learning method

Abstract

To provide an information processor, an information processing system, and a method for learning a model that can reduce the amount of calculation while preventing reduction of the accuracy of a model which predicts a target.SOLUTION: An information processor learning device includes: a clustering unit for grouping an aggregate of numerical data including a plurality of factors and a value varied according to the factors, into a plurality of clusters; an acquisition unit for acquiring at least one piece of the numerical data in each cluster from each cluster; and a learning unit for learning a model for calculating a value corresponding to input data including the factors, on the basis of the numerical data acquired from each cluster, by a regression analysis method using a Kernel method.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to an information processing device, an information processing system, and a model learning method. [Background technology]

[0002] For example, there is a technology that uses a model constructed by machine learning to predict the amount of power demand in a predetermined system or the characteristics of products manufactured in a plant (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-192014 Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, models for predicting power demand and product characteristics are sometimes constructed using a regression method such as support vector regression. When constructing a model based on a regression method such as support vector regression, increasing the number of training data items generally improves the accuracy of the constructed model. However, in such cases, the calculation time required for constructing the model increases. Furthermore, when predicting a target using a model constructed using a large amount of training data, the amount of calculation required by the information processing device increases, resulting in longer processing times.

[0005] The present invention has been made in consideration of the above-mentioned conventional problems, and aims to provide an information processing device that can reduce the amount of calculation while suppressing a decrease in the accuracy of a model that predicts a target. [Means for solving the problem]

[0006] The main present invention for solving the above-mentioned problems is an information processing device including: a clustering unit that groups a set of numerical data including a plurality of factors and values ​​that change depending on the plurality of factors into a plurality of clusters; an acquisition unit that acquires at least one of the numerical data included in each of the plurality of clusters from each of the plurality of clusters; and a learning unit that learns a model for calculating a value corresponding to input data including each of the plurality of factors based on the numerical data acquired from each of the plurality of clusters using a regression analysis method using a kernel method. [Effects of the Invention]

[0007] According to the present invention, it is possible to provide an information processing device that can reduce the amount of calculation while suppressing a decrease in the accuracy of a model that predicts a target. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram showing a configuration of a demand forecasting system 10. [Figure 2] FIG. 2 is a diagram illustrating an example of a hardware configuration of a learning device 20. [Figure 3] FIG. 2 is a diagram showing an example of information stored in a storage device 32. [Figure 4] FIG. 2 is a diagram showing an example of a data set 41. [Figure 5] FIG. 2 is a diagram illustrating an example of functional blocks realized in the learning device 20. [Figure 6] 10 is a flowchart showing an example of processing executed by the learning device 20. [Figure 7] FIG. 10 is a diagram showing an example of a learning model 44 construction process. [Figure 8] FIG. 10 is a diagram illustrating an example of data grouped into multiple clusters. [Figure 9] FIG. 10 illustrates an example of a change process. [Figure 10] FIG. 10 is a diagram showing an example of a state in which a data set 41 is classified. [Figure 11] FIG. 10 is a diagram showing an example of a learning model 44 construction process. [Figure 12] FIG. 10 is a diagram illustrating an example of multiple clusters for each pattern. [Figure 13] FIG. 2 is a diagram illustrating an example of a hardware configuration of a prediction device 21. [Figure 14] FIG. 2 is a diagram showing an example of functional blocks realized in the prediction device 21. [Figure 15] 10 is a flowchart showing an example of processing executed by the prediction device 21. DETAILED DESCRIPTION OF THE INVENTION

[0009] At least the following matters will become clear from the description of this specification and the accompanying drawings.

[0010] =====This embodiment===== <<<Configuration of Demand Forecasting System 10>>> Fig. 1 is a diagram showing the configuration of a demand forecasting system 10 according to one embodiment of the present invention. The demand forecasting system 10 is a system for forecasting the power demand of consumers in, for example, a predetermined power system (hereinafter referred to as "power system A"), and includes a learning device 20 and a prediction device 21. Consumers are, for example, those who use power from the power system, such as factories and ordinary homes.

[0011] Here, the power system A of this embodiment is provided with a data processing device 15 that outputs data x1 to x10 used to predict the power demand of consumers of the power system A. The data processing device 15 outputs, at predetermined intervals, ten pieces of data x1 to x10, including data indicating the temperature, humidity, amount of sunlight, etc. of the power system A, and data indicating the attributes of the measurement date (for example, whether it is a "weekday" or a "holiday").

[0012] The learning device 20 (first information processing device) constructs a model for predicting the power demand in the power system A by machine learning, based on the past data x1 to x10.

[0013] Prediction device 21 (second information processing device) predicts the power demand of power system A based on data x1 to x10 output from data processing device 15 and the model constructed by learning device 20. Note that learning device 20 and prediction device 21 are connected via network 25. Furthermore, demand prediction system 10 corresponds to an information processing system.

[0014] <<<About the learning device 20>>> ==Configuration of Learning Device 20== 2 is a diagram showing an example of the hardware configuration of the learning device 20. The learning device 20 is a computer including a CPU (Central Processing Unit) 30, a memory 31, a storage device 32, an input device 33, a display device 34, and a communication device 35.

[0015] The CPU 30 executes programs stored in the memory 31 and the storage device 32 to realize various functions of the learning device 20 .

[0016] The memory 31 is, for example, a RAM (Random Access Memory) and is used as a temporary storage area for programs, data, and the like.

[0017] The storage device 32 is a non-volatile storage device that stores various types of information such as programs executed by the CPU 30 and data sets.

[0018] The input device 33 is a device that accepts commands and data input by the user, and includes an input interface such as a keyboard and a touch sensor that detects a touch position on a touch panel display.

[0019] The display device 34 is, for example, a device such as a display, and the communication device 35 exchanges various programs and data with the prediction device 21 and other computers via the network 25.

[0020] 3 is a diagram showing an example of information stored in the storage device 32. The storage device 32 stores a control program 40, a data set 41, setting data 42, cluster data 43, a learning model 44, test data 45, and target value data 46.

[0021] The control program 40 is a program for realizing various functions of the learning device 20, and includes, for example, an OS (Operating System).

[0022] The data set 41 is data used when constructing a learning model 44 (described later), and includes past data x1 to x10 and y output from the data processing device 15, as shown in FIG.

[0023] Here, "data x1" is, for example, data indicating the temperature of power system A, and "data x2" is data indicating the humidity of power system A. Furthermore, "data x9" is data indicating the attributes of the day on which data x1 etc. was acquired (for example, whether it was a "weekday" or a "holiday"), and "data x10" is, for example, data indicating the amount of sunlight on power system A. Furthermore, "data y" is the amount of power usage on power system A corresponding to data x1 to x10. Note that in this embodiment, in data x9, "1" indicates a weekday, and "0" indicates a holiday.

[0024] Furthermore, data x3 to x8, like x1, x2, etc., are data relating to a predetermined physical quantity (for example, wind speed) of power system A, and therefore detailed explanations thereof will be omitted here.

[0025] The data set 41 is a set of numerical data including i pieces (for example, 10,000 pieces) of data x1 to x10 at different times. The first data of the data set 41 is acquired at time t1, for example, and data x1 indicating temperature is "27°C", data x2 indicating humidity is "62%", data x9 indicating the attribute of the acquisition date is "1 (weekday)", and data x10 indicating the amount of sunlight is "0.6 MJ / m 2". Also, data y indicating the amount of power consumption at this timing is "2443 kW". In this embodiment, "numeric data" refers to data in which data y indicates a numerical value, and is also called continuous value data or continuous data.

[0026] The setting data 42 is data that indicates conditions for classifying the data included in the dataset 41 into multiple patterns. For example, if the amount of power consumption in the power system A changes significantly around the boundary of "30°C," it is preferable that the learning model 44 (described later) be constructed using both data included in the dataset 41 that is equal to or greater than "30°C" and data that is less than "30°C." As will be described in detail later, the learning device 20 of this embodiment can classify the dataset 41 into multiple desired patterns (e.g., data that is equal to or greater than "30°C" and data that is less than "30°C") based on the setting data 42.

[0027] The condition set in the setting data 42 is a predetermined temperature, but is not limited to this, and may be, for example, a condition indicating a predetermined humidity or a predetermined amount of sunlight, or a condition indicating whether it is a "weekday" or a "holiday." In other words, the data may be data indicating the condition of any of the data x1 to x10.

[0028] The cluster data 43 is data indicating the number of clusters when grouping the data of the dataset 41. Here, the cluster data 43 in this embodiment includes a variable "k" indicating the number of clusters of the dataset 41. Note that, as will be described in detail later, if the dataset 41 is not classified into multiple patterns, the cluster data 43 becomes a variable "k0" indicating the number of clusters of the dataset 41.

[0029] On the other hand, when the data set 41 is classified into multiple patterns, the cluster data 43 becomes a variable "kn" indicating the number of clusters for each of the n patterns. Specifically, when the data set 41 is classified into two patterns, the cluster data 43 includes a variable "k1" indicating the number of clusters for pattern 1 and a variable "k2" indicating the number of clusters for pattern 2.

[0030] The learning model 44 is a model for predicting the power demand of consumers in the power system A based on the data x1 to x10 output from the data processing device 15. In this embodiment, when the learning model 44 is trained, the coefficients of the function of the learning model 44 are adjusted. The learning model 44 is constructed based on, for example, a support vector regression technique. In this embodiment, the function of the learning model 44 is expressed as y=f(x1, x2, ~, x10) for convenience.

[0031] The test data 45 is data for evaluating the prediction accuracy of the learning model 44, and includes, for example, a plurality of numerical data similar to the data set 41.

[0032] The target value data 46 is data indicating a "target value" of the accuracy of the learning model 44. Here, the "target value" in this embodiment is, for example, the prediction accuracy (e.g., 85%) obtained based on the learning model 44 constructed using the dataset 41 as training data. However, the "target value" may be a value determined by the user, regardless of the prediction accuracy of the learning model 44 constructed using the dataset 41 as training data.

[0033] ==Functional Blocks of Learning Device 20== 5 is a diagram showing an example of functional blocks realized in learning device 20. When CPU 30 of learning device 20 executes control program 40, learning device 20 realizes a classification unit 50, an initial value setting unit 51, a clustering unit 52, a calculation unit 53, an acquisition unit 54, a learning unit 55, an accuracy calculation unit 56, a determination unit 57, and a processing unit 58.

[0034] The classification unit 50 determines whether or not setting data 42 for classifying the data set 41 is stored in the storage device 32. Furthermore, if setting data 42 is stored in the storage device 32, the classification unit 50 classifies the data of the data set 41 into a plurality of patterns based on the setting data 42.

[0035] The initial value setting unit 51 sets an initial value for the number of clusters when grouping the data in the data set 41. For example, when the setting data 42 is not stored in the storage device 32, the initial value setting unit 51 sets a variable "k0" indicating the number of clusters (for example, 1 / 100 of the number of data in the data set 41) based on the number of data in the data set 41.

[0036] Furthermore, when the setting data 42 is stored in the storage device 32, the initial value setting unit 51 sets a variable "kn" indicating the number of clusters based on the number of data included in each of the multiple patterns. For example, the variable "kn" is set to, for example, 1 / 100 of the number of data included in the nth pattern. Note that although the initial value of the number of clusters is determined in accordance with the number of data, it may be a predetermined value.

[0037] The clustering unit 52 acquires the data set 41 and groups it into clusters the number of which is indicated by the variable "k" (here, k clusters) using, for example, the k-means method. Note that "grouping" here refers to dividing data into, for example, a predetermined number of groups.

[0038] The calculation unit 53 calculates the center of each of the multiple clusters grouped by the clustering unit 52. Here, the "center" of a cluster is, for example, the average value of multiple data included in the cluster.

[0039] The acquisition unit 54 acquires the centers of the calculated clusters, and the learning unit 55 uses the data acquired by the acquisition unit 54 as training data to construct the learning model 44 using support vector regression. Therefore, in this embodiment, the centers of the clusters grouped by the clustering unit 52 serve as training data. The "training data" refers to, for example, data used to construct the learning model 44, and is also referred to as training data or model construction data. While the learning unit 55 constructs the learning model 44 using support vector regression, this is not limiting, and any regression analysis method using a kernel method may be used. Here, the "regression analysis method using a kernel method" refers to, for example, a method of mapping training data to a feature space and performing linear data analysis in the feature space. The "regression analysis method using a kernel method" includes, in addition to support vector regression, kernel regression analysis and kernel multiple regression, for example.

[0040] The accuracy calculation unit 56 calculates the prediction accuracy of the learning model 44 using the test data 45, and the determination unit 57 determines whether the difference between the calculated prediction accuracy and a "target value" of the prediction accuracy is greater than a "predetermined value." For example, if the target value is "80%" and the "predetermined value" is "5%, the determination unit 57 determines whether the calculated classification accuracy is within the range of "75%" to "85%."

[0041] When the calculated prediction accuracy is higher than the "target value," the processing unit 58 changes the number of clusters that the clustering unit 52 groups so that the prediction accuracy becomes the "target value" (in this embodiment, so that the prediction accuracy falls within the range of the "target value" ± "predetermined value"). Specifically, when the prediction accuracy calculated by the accuracy calculation unit 56 is lower than the "target value," the processing unit 58 increases the number of clusters that the clustering unit 60 groups. On the other hand, when the prediction accuracy calculated by the accuracy calculation unit 61 is higher than the "target value," the processing unit 65 decreases the number of clusters that the clustering unit 60 groups.

[0042] <<Learning Process S10>> 6 is a flowchart showing an example of processing executed by each functional block of the learning device 20. First, the classification unit 50 determines whether or not the storage device 32 stores setting data 42 for classifying the dataset 41 (S20). In this embodiment, the processing to be executed differs depending on whether or not the setting data 42 is stored in the storage device 32. Therefore, first, a case where the setting data 42 is not stored in the storage device 32 will be described.

[0043] ==If the setting data 42 is not stored in the storage device 32== If the setting data 42 is not stored in the storage device 32 (S20: No), the initial value setting unit 51 sets the initial value of the number of clusters as the variable "k0" (S21) based on the number of data in the data set 41. As a result, the cluster data 43 in the storage device 32 includes information about the variable "k0".

[0044] After the process S21 is executed, the clustering unit 52 executes a process of constructing a learning model 44 (S22). Fig. 7 is a flowchart showing an example of the process S22. First, the clustering unit 52 acquires cluster data 43 indicating the number of clusters (S40).

[0045] Furthermore, the clustering unit 52 acquires the dataset 41 stored in the storage device 32, and groups the dataset 41 into a number of "k0" clusters (i.e., the number equal to the number of variables "k0") (S41).

[0046] 8 is a diagram illustrating a state in which a data set 41 is grouped into a plurality of clusters. Each data in the data set 41 includes ten pieces of data (x1 to x10), but for convenience, this embodiment illustrates the data as two pieces of data, x1 and x2. Here, the data set 41 is grouped into, for example, eight clusters A1 to A8 (variable "k0"=8).

[0047] Next, the calculation unit 53 calculates the center O of each of the eight clusters (S42). Specifically, in the case of cluster A1 in Fig. 8, the calculation unit 53 calculates the average value of the 12 data included in cluster A1 and sets the calculation result as center O1. Similarly, the calculation unit 53 calculates centers O2 to O8 of each of the seven clusters A2 to A8.

[0048] The acquiring unit 54 acquires data on the centers O1 to O8 of each of the multiple clusters calculated by the calculating unit 53 (S43). Furthermore, the learning unit 55 constructs a learning model 44 using support vector regression, using the data on the centers O1 to O8 as training data (S44). As a result, the process S22 in Fig. 6 ends. Note that the acquiring unit 54 may acquire multiple pieces of data included in each of the clusters other than the center O.

[0049] Furthermore, once process S22 is completed and the learning model 44 is constructed, the accuracy calculation unit 56 calculates the prediction accuracy of the learning model 44 using the test data 45 (S23). Then, the determination unit 57 calculates the "difference" between the accuracy calculated by the accuracy calculation unit 56 and the "target value" and determines whether the "difference" is greater than a "predetermined value" (S24). If the "difference" is smaller than the "predetermined value" (S24: No), that is, if the calculated classification accuracy is close to the "target value," the series of processes ends. On the other hand, if the "difference" is greater than the "predetermined value" (S24: Yes), that is, if the calculated accuracy is far from the "target value," a change process (S25) is executed.

[0050] In the change process S25 of FIG. 9, the determination unit 57 determines whether the accuracy calculated by the accuracy calculation unit 61 is higher than the "target value" (S50). If the calculated accuracy is higher than the "target value" (S50: Yes), the processing unit 58 decreases the variable "k0" by, for example, "1" (S51) to reduce the number of clusters grouped by the clustering unit 52. Specifically, the processing unit 58 updates the cluster data 43 so that the variable "k0" is decreased by, for example, "1". This makes it possible to maintain prediction accuracy while preventing the calculation time from becoming longer than necessary.

[0051] On the other hand, if the calculated prediction accuracy is lower than the "target value" (S50: No), the processing unit 58 increases the variable "k0" by, for example, "1" (S52) to increase the number of clusters grouped by the clustering unit 52. Specifically, the processing unit 58 updates the cluster data 43 so that the variable "k0" increases by, for example, "1". As a result, the number of clusters increases, and the prediction accuracy improves.

[0052] Then, when the change process S25 is executed, the processes S22 to S24 are repeated. That is, in this embodiment, the number of clusters is changed so as to reduce the difference between the prediction accuracy and the "target value," so that the prediction accuracy of the constructed learning model 44 falls within the range of the "target value" ± "predetermined value." Note that, in this embodiment, the number of clusters changed in processes S51 and S52 is set to "1," but it may be another number (for example, "2" or more).

[0053] In this way, the learning model 44 of this embodiment is constructed based on the center O, which is representative data of the data set 41, among the data included in the data set 41. As a result, the learning unit 55 can train the learning model 44 using a small amount of data while maintaining information about the distribution of the data in the data set 41. Note that the "distribution of data" refers to, for example, a distribution in a vector space when each of the data x1 to x10 is treated as a vector.

[0054] ==When the setting data 42 is stored in the storage device 32== If the setting data 42 is stored in the storage device 32 (S20: Yes), the classification unit 50 classifies the data set 41 into a plurality of patterns (S30) based on the setting data 42. For example, if the condition of the setting data 42 indicates a temperature Tx (e.g., 30°C), the classification unit 50 classifies the data set 41 into two patterns, Pattern 1, which is equal to or higher than the temperature Tx, and Pattern 2, which is lower than the temperature Tx, as shown in FIG.

[0055] Then, the initial value setting unit 51 sets an initial value for the number of clusters for each of the plurality of patterns based on the number of data included in each of the plurality of patterns (S31). Specifically, the initial value setting unit 51 sets an initial value indicating the number of clusters for the nth pattern as a variable "kn" based on the number of data included in the nth pattern. For example, if there are two patterns, the initial value setting unit 51 sets a variable "k1" as the initial value for pattern 1 and a variable "k2" as the initial value for pattern 2. Therefore, the cluster data 43 in the storage device 32 includes, for example, information on the variables "k1" and "k2."

[0056] After the process S31 is executed, the clustering unit 52 executes a process of constructing a learning model 44 (S32). Fig. 11 is a flowchart showing an example of the process S32. First, the clustering unit 52 acquires cluster data 43 indicating the number of clusters for each pattern (S60).

[0057] Furthermore, the clustering unit 52 groups the data classified into a plurality of patterns into clusters, the number of clusters being "kn" for each of the plurality of patterns (S61).

[0058] 12 is a diagram illustrating a state in which data of a data set 41 classified into two patterns is grouped into a plurality of clusters. Here, the data of pattern 1 is grouped into, for example, three clusters B1 to B3 (variable "k1"=3), and the data of pattern 2 is grouped into, for example, five clusters C1 to C5 (variable "k2"=5).

[0059] Next, the calculation unit 53 calculates the centers of each of the multiple clusters (S62). For example, in the case shown in Fig. 12, the calculation unit 53 calculates the centers P1 to P3 of the three clusters B1 to B3, respectively, and the centers Q1 to Q5 of the five clusters C1 to C5, respectively.

[0060] The acquiring unit 54 acquires data on the centers (e.g., centers P1 to P3, Q1 to Q5) of each of the multiple clusters calculated by the calculating unit 53 (S63). Furthermore, the learning unit 55 constructs a learning model 44 using support vector regression, using the data on the centers (e.g., centers P1 to P3, Q1 to Q5) as training data (S64). As a result, the process S32 in FIG. 11 ends. Note that the acquiring unit 54 may acquire multiple pieces of data included in each of the clusters other than the centers P and Q.

[0061] Furthermore, when process S32 is completed and learning model 44 is constructed, processes S23 and S24 described above in Fig. 6 are executed. As a result, if the "difference" between the accuracy calculated by accuracy calculation unit 56 and the "target value" is smaller than the "predetermined value" (S24: No), the series of processes ends, and if the "difference" is larger than the "predetermined value" (S24: Yes), change process S25 is executed.

[0062] By executing the change process S25 in FIG. 9 described above, the number of clusters (variable "kn") is changed so that the accuracy of the learning model 44 approaches the "target value." If the calculated accuracy is higher than the "target value" (S50: Yes), the processing unit 58 decreases the variable "kn" by, for example, "1" (S51) to reduce the number of clusters grouped by the clustering unit 52. Here, for example, if the number of patterns is "2," the processing unit 58 updates the cluster data 43 so that each of the variables "k1" and "k2" is decreased by "1." This makes it possible to maintain prediction accuracy while preventing calculation time from becoming longer than necessary.

[0063] On the other hand, if the calculated prediction accuracy is lower than the "target value" (S50: No), the processing unit 58 increases the variable "kn" by, for example, "1" (S52) to increase the number of clusters grouped by the clustering unit 52. For example, if the number of patterns is "2," the processing unit 58 updates the cluster data 43 so that each of the variables "k1" and "k2" increases by "1." As a result, the number of clusters increases, improving the prediction accuracy. In this way, when the change process S25 is executed, processes S32, S23, and S24 are repeated, and ultimately, the prediction accuracy of the constructed learning model 44 falls within the range of the "target value" ± "predetermined value."

[0064] <<<About Prediction Device 21>>> ==Configuration of prediction device 21== 13 is a diagram showing an example of the hardware configuration of the prediction device 21. The prediction device 21 is a computer including a CPU 70, a memory 71, a storage device 72, an input device 73, a display device 74, and a communication device 75. Note that the hardware configuration of the prediction device 21 is similar to that of the learning device 20, and therefore a detailed description thereof will be omitted here.

[0065] The storage device 72 stores the learning model 44 constructed in the learning process S10 of FIG.

[0066] Like the control program 40, the prediction program 80 is a general term for programs for realizing various functions of the prediction device 21. The prediction data 81 is data indicating a predicted value of the power demand of the power system A.

[0067] ==Function Blocks== 14 is a diagram illustrating an example of functional blocks realized in the prediction device 21. When the CPU 70 of the prediction device 21 executes the prediction program 80, an acquisition unit 100 and a calculation unit 101 are realized in the prediction device 21.

[0068] The acquiring unit 100 acquires data x1 to x10 output at predetermined intervals from the data processing device 15. The data x1 to x10 output from the data processing device 15 corresponds to "input data."

[0069] The calculation unit 101 calculates the value of the power demand (data y) according to the input data, based on the input data acquired by the acquisition unit 100 and the learning model 44 stored in the storage device 72.

[0070] <<Demand forecast processing S100>> 9, the acquisition unit 100 acquires input data (data x1 to x10) (S200). Then, the calculation unit 101 calculates the value of the power demand (data y) of the power system A corresponding to the input data based on the input data (data x1 to x10) and the learning model 44 (S201).

[0071] Here, the learning model 44 based on support vector regression is a model constructed based on the center, which is representative data of the data set 41, among the data included in the data set 41. In other words, since the learning model 44 is a model constructed based on a small amount of training data, the amount of data that the calculation unit 101 must refer to in process S201 is also small.

[0072] Therefore, in this embodiment, the amount of calculation performed by the calculation unit 101 in process S201 can be reduced compared to when a learning model is constructed using all of the data in the dataset 41. Note that process S201 corresponds to so-called inference processing.

[0073] Then, the calculation unit 101 stores the calculated value of the power demand in the storage device 72 as the predicted data 81 (S202). As a result, the user can grasp the predicted value of the power demand of the power system A.

[0074] ===Summary=== The demand forecasting system 10 of this embodiment has been described above. The clustering unit 52 of the learning device 20 groups the data of the dataset 41 into multiple clusters, and the learning unit 55 constructs the learning model 44 by using support vector regression with a kernel method on the data of each of the multiple clusters. In this way, in this embodiment, data is acquired from clusters, so the amount of training data can be efficiently reduced while maintaining the distribution of the data in the dataset 41, which is the training data. As a result, the learning model 44 can be constructed in a shorter time. Furthermore, the amount of calculation required for inference processing can be reduced while suppressing a decrease in the accuracy of the learning model 44.

[0075] Furthermore, the processing unit 58 changes the number of clusters to be grouped by the clustering unit 52 so that the difference between the accuracy calculated by the accuracy calculation unit 56 and the target accuracy value becomes smaller (S25). As a result, in this embodiment, the prediction accuracy of the learning model 44 can be set to a value close to the target value while suppressing the amount of calculation.

[0076] Furthermore, since the accuracy of the learning model 44 approaches the "target value" when process S25 is executed, the initial value of the number of clusters may be any value. However, for example, setting a small value as the initial value of the number of clusters and increasing the number of clusters can reduce calculation time more than setting a large value as the initial value of the number of clusters and decreasing the number of clusters. Therefore, it is more preferable to improve the accuracy of the learning model 44 by setting a small value as the initial value of the number of clusters and increasing the number of clusters.

[0077] Furthermore, the amount of power consumption (or power demand) of power system A generally varies depending on various factors such as the temperature, humidity, and amount of sunlight of power system A. The classification unit 50 of this embodiment classifies the data set 41 into a plurality of patterns based on, for example, setting data 42 that indicates the conditions of factors that affect the value of the power demand to be predicted. Furthermore, with the data of the data set 41 classified into a plurality of patterns, the clustering unit 52 groups each of the data of the plurality of patterns into a cluster. Therefore, in this embodiment, by clustering by pattern, data with different patterns is not included in the same cluster, thereby improving the accuracy of the learning model 44.

[0078] The clustering unit 52 may also change the number of clusters depending on the number of data included in each of the multiple patterns, thereby allowing a more optimal number of clusters to be set as the initial value when performing grouping.

[0079] Furthermore, the calculation unit 53 calculates the center of each of the multiple clusters, and the learning unit 55 uses the calculated data as training data. This makes it possible to reduce the amount of data while accurately maintaining the distribution state in the vector space of the data set 41, which is the training data.

[0080] The above-described embodiments are intended to facilitate understanding of the present invention and are not intended to limit the present invention. Furthermore, the present invention may be modified or improved without departing from the spirit thereof, and the present invention includes equivalents thereof.

[0081] For example, the prediction target in this embodiment is the amount of electricity demand, but is not limited to this. For example, the quality of a product (e.g., iron) produced in a factory may be predicted based on the operation data of the factory. In such a case, the data x1 to x10 in this embodiment correspond to the operation data of the factory, and the data y indicating the amount of electricity consumption is a value indicating a predetermined quality of the product (e.g., iron).

[0082] Furthermore, although the acquisition unit 54 of the present embodiment acquires the center of a cluster, this is not limiting and actual data included in the cluster may be acquired and used as training data. Furthermore, the acquisition unit 54 may acquire a plurality of data from the centers included in the cluster or from the actual data and use these as training data. In this way, increasing the amount of data acquired by the acquisition unit 54 can further improve the accuracy of the learning model 44.

[0083] Although the determination unit 57 determines whether the calculated prediction accuracy is within the range of "75%" to "85%," the present invention is not limited to this. For example, the determination unit 57 may determine whether the accuracy is greater than a target value of "80%." In such a case, the processing unit 58 changes the number of clusters to be grouped by the clustering unit 52 so that the calculated accuracy matches the "target value," i.e., so that the difference between the prediction accuracy and the target value becomes smaller. Even in such a case, the same effect as that of this embodiment can be obtained. [Explanation of symbols]

[0084] 10 Demand forecasting system 15 Data Processing Device 20 Learning Device 21 Prediction Device 25 Network 30,70 CPU 31,71 memory 32,72 Storage device 33,73 Input devices 34,74 Display device 35,75 Communication equipment 40 Control Program 41 datasets 42 Configuration Data 43 Cluster Data 44 Learning Model 45 test data 46 Target Value Data 50 Classification Department 51 Initial value setting section 52 Clustering Department 53,101 Calculation section 54 Acquisition Department 55 Learning Department 56 Accuracy calculation section 57 Judgment section 58 Processing section 80 Prediction Program 81 Forecast Data 100 Acquisition Department

Claims

1. a clustering unit that groups a set of numerical data including a plurality of factors and values ​​that vary depending on the plurality of factors into a plurality of clusters; an acquisition unit that acquires at least one of the numerical data included in each of the plurality of clusters, but not all of the numerical data, from each of the plurality of clusters; a learning unit that learns a model for calculating values ​​corresponding to input data including each of the plurality of factors based on the numerical data acquired from each of the plurality of clusters, using a regression analysis method that uses a kernel method; a classification unit that classifies the set of numerical data into a plurality of patterns based on information indicating conditions of predetermined factors among the plurality of factors; Including, The learning unit using at least one of the actual numerical data included in each of the plurality of clusters and data included in each of the plurality of clusters calculated from the numerical data included in each of the plurality of clusters as training data, and learning the one model using a regression analysis method using the kernel method; the clustering unit groups the sets of the numerical data included in each of the plurality of patterns into a plurality of clusters for each of the plurality of patterns, the number of clusters corresponding to the number of the numerical data included in each of the plurality of patterns; 1. An information processing device comprising:

2. 2. The information processing device according to claim 1, an accuracy calculation unit that calculates the accuracy of the model based on data for evaluating the accuracy of the model; a processing unit that changes the number of the plurality of clusters that are grouped by the clustering unit so that the accuracy of the calculated model reaches a target value; The clustering unit grouping the set of numerical data based on the modified number of clusters; An information processing device characterized by:

3. 3. The information processing device according to claim 2, The processing unit increasing the number of the plurality of clusters grouped by the clustering unit so that the accuracy of the calculated model reaches the target value; An information processing device characterized by:

4. The information processing device according to any one of claims 1 to 3, a calculation unit for calculating a center of each of the plurality of clusters; The acquisition unit acquiring data indicative of at least the center from each of the plurality of clusters; An information processing device characterized by:

5. a clustering unit that groups a set of numerical data, including a plurality of factors and values ​​that change depending on the plurality of factors, into a plurality of clusters; an acquisition unit that acquires, from each of the plurality of clusters, at least one, but not all, of the numerical data included in each of the plurality of clusters; a learning unit that learns a model for calculating values ​​corresponding to input data including each of the plurality of factors based on the numerical data acquired from each of the plurality of clusters using a regression analysis method that employs a kernel method; and a classification unit that classifies the set of numerical data into a plurality of patterns based on information indicating conditions of a predetermined factor among the plurality of factors; a second information processing device including a calculation unit that calculates a value corresponding to the input data based on the input data and the model; Equipped with The learning unit using at least one of the actual numerical data included in each of the plurality of clusters and data included in each of the plurality of clusters calculated from the numerical data included in each of the plurality of clusters as training data, and learning the one model using a regression analysis method using the kernel method; The clustering unit grouping the sets of the numerical data included in each of the plurality of patterns into a plurality of clusters for each of the plurality of patterns, the number of clusters corresponding to the number of the numerical data included in each of the plurality of patterns; An information processing system comprising:

6. The computer A clustering process for grouping a set of numerical data including a plurality of factors and values ​​that vary depending on the plurality of factors into a plurality of clusters; an acquisition process of acquiring at least one, but not all, of the numerical data included in each of the plurality of clusters from each of the plurality of clusters; a learning process for learning a model for calculating values ​​corresponding to input data including each of the plurality of factors based on the numerical data acquired from each of the plurality of clusters, using a regression analysis method using a kernel method; a classification process for classifying the set of numerical data into a plurality of patterns based on information indicating the conditions of a predetermined factor among the plurality of factors; Run The learning process includes: using at least one of the actual numerical data included in each of the plurality of clusters and data included in each of the plurality of clusters calculated from the numerical data included in each of the plurality of clusters as training data, and learning the one model using a regression analysis method using the kernel method; The clustering process includes: grouping the sets of the numerical data included in each of the plurality of patterns into a plurality of clusters for each of the plurality of patterns, the number of clusters corresponding to the number of the numerical data included in each of the plurality of patterns; How the model is trained.

Images