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

Abstract

An information processing device (1) is provided with a detection unit (31), an extraction unit (32), and a training unit (34). The detection unit (31) detects the lead-lag interval between an objective variable and each of a plurality of explanatory variable candidates. The extraction unit (32) extracts, from among the plurality of explanatory variable candidates, a plurality of explanatory variable candidates for which the lead-lag interval has been detected by the detection unit (31), as explanatory variables. The training unit (34) trains a model that outputs a future value of the objective variable, on the basis of past time-series values of the objective variable and past time-series values of the explanatory variables extracted by the extraction unit (32).

Description

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

[0001] The present invention relates to an information processing device, an information processing method, and an information processing program.

[0002] Conventionally, multivariate time series forecasting is known, which is a task of predicting the future value of a target variable using past values ​​of the target variable and multiple explanatory variables used to predict the target variable. Regarding multivariate time series forecasting, a technique is known that simultaneously captures dependencies between variables in addition to dependencies in the time direction (see, for example, Non-Patent Document 1).

[0003] Yunhao Zhang and 1 others, "Crossformer: Transformer Utilizing Cross-Dimension Dependency for Multivariate Time Series Forecasting", International Conference on Learning Representations (ICLR) 2023

[0004] However, the above-mentioned conventional techniques have a problem in that prediction accuracy decreases when they try to find dependencies between variables that do not have dependencies.

[0005] The present invention has been made in view of the above, and aims to improve the prediction accuracy of a prediction model that outputs future values ​​of a dependent variable.

[0006] In order to solve the above-mentioned problems and achieve the object, an information processing device according to the present invention includes a detection unit, an extraction unit, and a learning unit. The detection unit detects a lead-lag interval between a dependent variable and each of a plurality of explanatory variable candidates. The extraction unit extracts, from the plurality of explanatory variable candidates, explanatory variable candidates for which the detection unit has detected a lead-lag interval as explanatory variables. The learning unit learns a prediction model that outputs future values ​​of the dependent variable based on past time-series values ​​of the dependent variable and past time-series values ​​of the explanatory variables extracted by the extraction unit.

[0007] According to the present invention, it is possible to improve the prediction accuracy of a prediction model that outputs future values ​​of a dependent variable.

[0008] FIG. 1 is a diagram for explaining an overview of information processing by an information processing device according to an embodiment. FIG. 2 is a schematic diagram showing an example of the configuration of an information processing device according to an embodiment. FIG. 3 is a diagram showing an example of time-series values ​​of two explanatory variables and a response variable. FIG. 4 is a diagram showing an example of prediction accuracy depending on the presence or absence of an explanatory variable that does not have a lead-lag relationship. FIG. 5 is a diagram showing an example of MAE when the window width and prediction length are constant and the value of the lead-lag interval is changed. FIG. 6 is a diagram showing an example of MAE when the lead-lag interval and prediction length are constant and the value of the window width is changed. FIG. 7 is a diagram showing an example of MAE when the lead-lag interval and window width are constant and the value of the prediction length is changed. FIG. 8 is a flowchart showing an example of information processing by a processing unit of an information processing device according to an embodiment. FIG. 9 is a diagram showing an example of a computer that executes an information processing program.

[0009] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited to this embodiment. In addition, in the description of the drawings, the same parts are designated by the same reference numerals.

[0010] 1 is a diagram illustrating an overview of information processing by an information processing device according to an embodiment. The information processing device according to the embodiment learns a prediction model that performs a task of predicting a future value of a dependent variable to be predicted using past values ​​of the dependent variable to be predicted and explanatory variables used in predicting the prediction target.

[0011] As shown in FIG. 1, the information processing device stores the time series values ​​y １：ｔ and the time series values ​​of the d candidate explanatory variables {xc （１） １：ｔ , ..., xc （ｄ） １：ｔ} is input, where d is an integer of 2 or more.

[0012] The information processing device calculates a target variable y and d candidate explanatory variables xc based on the input time series values. （１） , ..., xc （ｄ） The lead-lag interval θc between each of １ , ..., θcｄ A detection process is performed to detect the difference (step S1). Lead means that a change in the value of a certain variable precedes a change in the value of another variable in time, and lag means that a change in the value of a certain variable lags a change in the value of another variable in time. The lead-lag interval is the time difference in the lead-lag phenomenon in which a correlation appears with a time lag between the time series values ​​of two variables. In the following, the lead-lag interval θc １ , ..., θc ｄ When each of these is referred to without being distinguished from the others, it may be referred to as the lead-lag interval θc. For explanatory variables for which the lead-lag interval θc is not detected by the detection process in step S1, θc is set to NaN.

[0013] Next, the information processing device selects d candidates for explanatory variables xc （１） , ..., xc （ｄ） Among these, d' candidates for explanatory variables for which the lead-lag interval θc was detected in step S1 are designated as explanatory variables x （１） , ..., x （ｄ’） (step S2).

[0014] As a result, d candidates for explanatory variables xc （１） , ..., xc （ｄ） Among the explanatory variable candidates, those for which the lead-lag interval θc was not detected in step S1 are excluded. （１） , ..., x （ｄ’） Each of the following is a candidate explanatory variable xc （１） , ..., xc （ｄ） The lead-lag interval θc is one of the d' candidates for the explanatory variables. （１） , ..., x （ｄ’） The lead-lag interval between each of the １ , ..., θ ｄ ’ It is written as follows.

[0015] Furthermore, the information processing device calculates the d' explanatory variables x extracted in step S2. （１） , ..., x （ｄ’） Lead-lag interval θ １, ..., θ ｄ’ For example, the information processing device determines the window width w of the prediction model based on the lead-lag interval θ １ , ..., θ ｄ’ The maximum value among these is determined as the window width w of the prediction model.

[0016] Next, the information processing device calculates the past time series values ​​y １：ｔ and the past time series values ​​x of the d' explanatory variables extracted in step S2 （１） １：ｔ , ..., x （ｄ’） １：ｔ Based on the above, a prediction model is learned with a learning window width w (step S4).

[0017] In the information processing device, as described above, the candidate explanatory variables for which the lead-lag interval θc is not detected are excluded from the explanatory variables, and therefore, the prediction accuracy of the prediction model can be improved compared to when all the candidate explanatory variables are used as explanatory variables. １ , ..., θ ｄ’ By determining the window width w of the prediction model based on the above, a prediction model with high prediction accuracy can be generated.

[0018] [Configuration of information processing device] Fig. 2 is a schematic diagram showing a configuration example of an information processing device according to an embodiment. As shown in Fig. 2, the information processing device 1 according to the embodiment is realized by a general-purpose computer such as a personal computer or a server, and includes a communication unit 10, a storage unit 11, and a processing unit 12.

[0019] The communication unit 10 controls communication between an external device and the processing unit 12 via a telecommunication line such as a local area network (LAN) or the Internet. The communication unit 10 is realized by, for example, a communication module or a network interface card (NIC).

[0020] The storage unit 11 stores in advance or temporarily stores each time processing is performed, an information processing program that operates the information processing device 1, and data used during execution of the information processing program. The storage unit 11 includes a learning data storage unit 20 that stores learning data input from an external device, and a prediction model storage unit 21 that stores a prediction model generated by learning by the processing unit 12.

[0021] The processing unit 12 is realized using a processor such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit), and executes a processing program stored in the storage unit 11. As a result, the processing unit 12 functions as a learning data acquisition unit 30, a detection unit 31, an extraction unit 32, a determination unit 33, a learning unit 34, a prediction data acquisition unit 35, a prediction unit 36, and a prediction result output unit 37, as illustrated in FIG. 2 , to perform information processing. Note that these functional units may each be implemented in different hardware, or some of them may be implemented in different hardware. The processing unit 12 may also include other functional units.

[0022] The learning data acquisition unit 30 acquires the time series value y １：ｔ and the time series values ​​of multiple candidate explanatory variables {xc （１） １：ｔ , ..., xc （ｄ） １：ｔ} is acquired from an external device via the communication unit 10.

[0023] The learning data acquisition unit 30 outputs the learning data acquired from the external device to the detection unit 31. The learning data acquisition unit 30 also stores the learning data acquired from the external device in the learning data storage unit 20.

[0024] The detection unit 31 detects the objective variable y and a plurality of explanatory variable candidates xc based on the learning data acquired by the learning data acquisition unit 30. （１） , ..., xc （ｄ） The lead-lag interval θc between each of １ , ..., θc ｄ Detect the lead-lag interval θc １ is the target variable y and the candidate explanatory variables xc（１） is the lead-lag interval between ｄ is the target variable y and the candidate explanatory variables xc （ｄ） is the lead-lag interval between

[0025] The detector 31 calculates the lead-lag interval θc using a cross-correlation function (CCF). １ , ..., θc ｄ The detection unit 31 detects the lead-lag interval θc by, for example, calculating the following equation (1): ｋ where k is kε{1, ..., d}. In the following formula (1), "L" is the detectable lead-lag interval θc ｋ The correlation coefficient corr(xc, y) is expressed by the following equation (2).

[0026]

[0027] The detection unit 31 detects the value y obtained by shifting y by L in the past. １＋Ｌ：ｔ and xc （ｋ） τ：ｔ－Ｌ－１＋τ Correlation coefficient corr(y １＋Ｌ：ｔ , xc （ｋ） τ：ｔ－Ｌ－１＋τ ) is repeatedly calculated while incrementing the value of τ by 1 from 1 to L+1. In this way, the detection unit 31 calculates L+1 correlation coefficients corr(xc, y). Then, the detection unit 31 determines the interval j at which the average value of the calculated L+1 correlation coefficients corr(xc, y) is maximum as the lead-lag interval θc ｋ Detect as.

[0028] The detection unit 31 detects the lead-lag interval θc ｋ If θc cannot be detected, ｋ For example, when the maximum value of the correlation coefficient corr(xc, y) is below a preset threshold, the detection unit 31 determines that the lead-lag interval θc ｋ is considered not to exist, and θc ｋ = NaN.

[0029] For example, y = umbrella sales, xc （１） = precipitation, xc （２） = wind speed, xc（３） = sunshine hours, xc （４） = humidity. Also, precipitation x c （１） The lead-lag interval between the sales of umbrellas and the sales of umbrellas is time T1, and the sunshine hours xc （３） The lead-lag interval between the wind speed xc （２） and humidity xc （４） and are not dependent on the umbrella sales y. In this case, the detection unit 31 １ , θc ２ , θc ３ , θc ４}={T1, NaN, T2, NaN}.

[0030] The lead-lag interval θc detected by the detection unit 31 ｋ The detection method for the lead-lag interval θc may be a detection method other than the detection method using the reciprocal function. For example, the detection unit 31 may detect the lead-lag interval θc using a statistical method such as a moving average or Granger causality test. ｋ The configuration may be such that the signal is detected.

[0031] Furthermore, the detection unit 31 can add or weight the values ​​of two or more candidate explanatory variables to treat the two or more candidate explanatory variables as one candidate explanatory variable and detect the lead-lag interval between the candidate explanatory variable and the objective variable. Weighting is performed, for example, according to the fluctuation range between the values ​​of the two or more candidate explanatory variables, but is not limited to this example.

[0032] For example, the detection unit 31 determines whether or not a lead-lag interval between two or more explanatory variable candidates can be detected. Then, the detection unit 31 adds or weights the values ​​of two or more explanatory variable candidates for which a lead-lag interval cannot be detected, thereby treating the two or more explanatory variable candidates as a single explanatory variable candidate.

[0033] Alternatively, suppose that the lead-lag interval between a specific explanatory variable candidate and the dependent variable cannot be detected, but the lead-lag interval between the specific explanatory variable candidate and another explanatory variable candidate for which a lead-lag interval between the specific explanatory variable candidate and the dependent variable can be detected. In this case, the detection unit 31 can also set the lead-lag interval between the specific explanatory variable candidate and the dependent variable to a value obtained by adding or subtracting the lead-lag interval between the specific explanatory variable candidate and the other explanatory variable candidate to or from the lead-lag interval between the other explanatory variable candidate and the dependent variable.

[0034] The extraction unit 32 extracts d candidates xc （１） , ..., xc （ｄ） Among these, each of the d' candidates for explanatory variables for which the lead-lag interval θc has been detected by the detection unit 31 is referred to as an explanatory variable x （１） , ..., x （ｄ’） For example, the extraction unit 32 extracts the value of d=4 and {θ １ , θ ２ , θ ３ , θ ４}={4, NaN, 8, NaN}, two candidate explanatory variables xc （１） , xc （３） Let x be the explanatory variable （１） , x （２） Extract as.

[0035] The determination unit 33 determines the d' explanatory variables x extracted by the extraction unit 32. （１） , ..., x （ｄ’） Lead-lag interval θ １ , ..., θ ｄ’ The window width w of the prediction model is determined based on the above equation. This allows the determination unit 33 to reduce the search range compared to when the window width is determined by a full search.

[0036] The determining unit 33 determines the lead-lag interval θ as shown in the following equation (3), for example. １ , ..., θ ｄ’ The largest lead-lag interval among the above is determined as the window width w. １ , ..., θ ｄ’ ) ... (3)

[0037] Furthermore, the determining unit 33 determines, for example, the lead-lag interval θ １ , ..., θ ｄ’ The value obtained by adding a fixed value k1 to the largest lead-lag interval among the above is determined as the window width w, or the lead-lag interval θ １ , ..., θ ｄ’ Alternatively, the window width w may be determined by multiplying the largest lead-lag interval by a fixed value k2.

[0038] Furthermore, the determining unit 33 determines, for example, the lead-lag interval θ １ , ..., θ ｄ’ Based on this, the explanatory variable x （１） , ..., x （ｄ’） For example, the determination unit 33 may determine the window width individually for each explanatory variable x （１） The window width for the lead-lag interval θ １ and the explanatory variable x （ｄ’） The window width for the lead-lag interval θ ｄ’ It is also possible to do this.

[0039] In addition to the window width, the determining unit 33 also determines the lead-lag interval θ １ , ..., θ ｄ’ For example, the determining unit 33 may determine the prediction length h of the prediction model based on the lead-lag interval θ １ , ..., θ ｄ’ Alternatively, the predicted length h may be determined to be equal to or less than the largest lead-lag interval among the above.

[0040] The learning unit 34 calculates the past time series values ​​y １：ｔ and the past time series values ​​x of the d' explanatory variables extracted by the extraction unit 32. （１） １：ｔ , ..., x （ｄ’） １：ｔ Based on this, the learning unit 34 learns a prediction model f using the window width w determined by the determination unit 33 as the learning window width. The learning unit 34 stores the learned prediction model f in the prediction model storage unit 21. The learned prediction model f is an example of a trained model.

[0041] The prediction model f is expressed by, for example, the following formula (4). The prediction model f shown in the following formula (4) is a prediction model with a prediction length h. In the following formula (4), ^ ｔ＋１：ｔ＋ｈ " is the predicted value of the predicted length h for the objective variable y. The predicted length h is determined by the determination unit 33 as described above, but is not limited to this example.

[0042]

[0043] The prediction model f shown in the above formula (4) is a prediction model that performs multivariate time series prediction, which is a task of predicting future values ​​of a dependent variable using past values ​​of the dependent variable to be predicted and multiple explanatory variables used to predict the predicted variable. However, the prediction model f may be any prediction model that predicts future values ​​of the dependent variable, and is not limited to a prediction model that performs multivariate time series prediction. For example, the prediction model f may be a prediction model in which the number of explanatory variables used for prediction is one, or may be another prediction model. Furthermore, the prediction model f is, for example, a crossformer-based prediction model, but may also be a transformer-based prediction model, an RNN (recurrent neural network)-based prediction model, or another prediction model.

[0044] The calculation data acquisition unit 35 acquires calculation data to be used for prediction. The calculation data may be, for example, a past time series value y ｔ－ｗ：ｔ and the past time series values ​​x of d' explanatory variables （１） ｔ－ｗ：ｔ , ..., x （ｄ’） ｔ－ｗ：ｔ Includes:

[0045] The prediction unit 36 ​​acquires the trained prediction model f from the prediction model storage unit 21. The prediction unit 36 ​​inputs the prediction data acquired by the prediction data acquisition unit 35 into the acquired trained prediction model f, and calculates the future time-series value y of the dependent variable. ^ ｔ＋１：ｔ＋ｈ By outputting the above to the prediction model f, the time series value y of the objective variable up to h steps in the future can be calculated. ^ ｔ＋１：ｔ＋ｈ Predict.

[0046] The prediction result output unit 37 outputs the time series value y ^ ｔ＋１：ｔ＋ｈ The prediction result output unit 37 outputs information indicating the time series value y predicted by the prediction unit 36 ​​to an external device or the like via the communication unit 10. ^ ｔ＋１：ｔ＋ｈ It is also possible to store information indicating this in the storage unit 11.

[0047] Here, the explanatory variable x shown in the following formula (5) ｉ （１） and the explanatory variable x shown in the following formula (6) ｉ （２） and the objective variable y ｉ Suppose we have a data set containing two explanatory variables x ｉ （１） , x ｉ （２） and the objective variable y ｉ As shown in the following formula (7) and FIG. 3, the explanatory variable x ｉ （１） is the objective variable y ｉ and the explanatory variable x ｉ （２） is the objective variable y ｉ and does not have a lead-lag relationship.

[0048]

[0049] First, we will explain why we exclude explanatory variable candidates for which no lead-lag interval is detected without using them as explanatory variables. ｉ （２） 4 is a diagram showing an example of prediction accuracy depending on the presence or absence of a cross-former-based prediction model. The prediction accuracy shown in FIG. 4 is an example of MAE (Mean Absolute Error) when a cross-former-based prediction model is used, and this prediction model has been modified so that the objective variable has one-dimensional loss during training and prediction.

[0050] As shown in Figure 4, the explanatory variable x ｉ （２） Explanatory variables x includingｉ （１） , x ｉ （２） Using the objective variable y ｉ The MAE for predicting the following is 0.304. On the other hand, the explanatory variable x ｉ （２） One explanatory variable x without ｉ （１） Using the objective variable y ｉ The MAE is 0.277 when predicting the following. The smaller the MAE value, the higher the prediction accuracy. Therefore, ｉ （１） Using the objective variable y ｉ When making a prediction of ｉ （１） , x ｉ （２） Using the objective variable y ｉ The prediction accuracy is higher than when the prediction is made using the

[0051] In this way, the explanatory variable x that does not have a lead-lag relationship ｉ （２） In the above-described prediction model f, explanatory variables that do not have a lead-lag relationship are excluded, and therefore the information processing device 1 can generate a prediction model f with high prediction accuracy.

[0052] Next, the window width w and the lead-lag interval θ １ The effect of the relationship on prediction accuracy will be explained. The window width w and prediction length h are constant, and the lead-lag interval θ １ The prediction accuracy when the value of θ is changed and the lead-lag interval θ １ The prediction accuracy was measured using the time series values ​​shown in the above equations (5) to (7) when the prediction length h was kept constant and the window width w was changed.

[0053] FIG. 5 shows the relationship between the lead-lag interval θ and the window width w and the prediction length h. １ 5 is a diagram showing an example of MAE when the value of θ is changed. As shown in FIG. 5, the window width w and the prediction length h are fixed at w=8 and h=8, and the lead-lag interval θ １ θ １When changing ∈{1, 2, 4, 8, 16, 32}, θ １ The MAE is lowest when the window width w and the lead-lag interval θ １ The prediction accuracy of the prediction model is higher when the two are consistent.

[0054] FIG. 6 shows the lead-lag interval θ １ 6 is a diagram showing an example of MAE when the lead-lag interval θ is changed while the prediction length h is kept constant. １ and the predicted length h is θ １ When the window width w is varied as wε{1, 2, 4, 8, 16, 32} while keeping the values ​​of σ and h constant as σ=8 and σ=8, the MAE is lowest when w=8.

[0055] Therefore, in this case too, the window width w and the lead-lag interval θ １ The prediction accuracy of the prediction model is higher when the window width w is equal to the lead-lag interval θ １ or more and lead-lag interval θ １ If the window width w is within twice the lead-lag interval θ １ The prediction accuracy of the prediction model will be higher than when the value is less than half of the original value.

[0056] In this way, the objective variable y ｉ and explanatory variable x ｉ （１） The lead-lag interval θ between １ As described above, the information processing device 1 generates a prediction model with high prediction accuracy by setting the window width w according to the lead-lag interval θ １ , ..., θ ｄ ’ Since the window width w is determined according to the above, a prediction model with high prediction accuracy can be generated.

[0057] FIG. 7 shows the lead-lag interval θ １ 8 is a diagram showing an example of MAE when the window width w and the lead-lag interval θ are kept constant and the value of the prediction length h is changed. １ With w = 8 and θ １= 8 and the prediction length h is changed as hε{1, 2, 4, 8, 16, 32}, the MAE is reduced by setting the prediction length h≦window width w.

[0058] Therefore, the window width w and the lead-lag interval θ １ and the window width w are the same, the prediction accuracy of the prediction model is high when the prediction length h is equal to or less than the window width w. In this way, the prediction accuracy of the prediction model can be improved by setting the prediction length h to be equal to or less than the window width w in the prediction model f.

[0059] [Information Processing] Next, a procedure of information processing by the processing unit 12 of the information processing device 1 according to the embodiment will be described. Fig. 8 is a flowchart showing an example of information processing by the processing unit 12 of the information processing device 1 according to the embodiment.

[0060] 8, the processing unit 12 of the information processing device 1 determines whether or not it is time to learn (step S10). The learning timing may be, for example, when a user of the information processing device 1 performs a specific operation or at a predetermined timing, but is not limited to such examples. When the processing unit 12 determines that it is time to learn (step S10: Yes), it acquires information on past time-series values ​​of each of the multiple explanatory variable candidates and the target variable from the storage unit 11 (step S11).

[0061] Next, the processing unit 12 detects the lead-lag interval between the objective variable and each of the multiple explanatory variable candidates based on the information acquired in step S11 (step S12).The processing unit 12 then extracts, from the multiple explanatory variable candidates, the explanatory variable candidates for which the lead-lag interval has been detected as explanatory variables (step S13).The processing unit 12 also determines the window width of the prediction model based on the lead-lag interval between the objective variable and each of the multiple explanatory variables extracted in step S13 (step S14).

[0062] The processing unit 12 learns a prediction model with the window width determined in step S14 based on the past time series values ​​of the objective variable and the past time series values ​​of the explanatory variables extracted in step S13 (step S15).

[0063] When the processing of step S15 is completed or when it is determined that the learning timing has not arrived (step S10: No), the processing unit 12 determines whether the prediction timing has arrived (step S16). The prediction timing is, for example, the timing when the user of the information processing device 1 performs a specific operation or a predetermined timing, but is not limited to such examples.

[0064] When the processing unit 12 determines that the prediction timing has arrived (step S16: Yes), it inputs the prediction data into the prediction model trained in step S15, thereby predicting future time-series values ​​of the dependent variable (step S17). In step S17, the processing unit 12 inputs the prediction data into the trained prediction model and causes the trained prediction model to output future time-series values ​​of the dependent variable, thereby predicting future time-series values ​​of the dependent variable. When the processing of step S17 has ended, or when it has determined that the prediction timing has not arrived (step S16: No), the processing unit 12 determines whether the operation end timing has arrived (step S18). The processing unit 12 determines that the operation end timing has arrived, for example, when the power of the information processing device 1 is turned off.

[0065] If the processing unit 12 determines that the operation end time has not yet arrived (step S18: No), it proceeds to step S10, and if it determines that the operation end time has arrived (step S18: Yes), it terminates the processing shown in Figure 8.

[0066] [Effects] As described above, the information processing device 1 according to the embodiment includes a detection unit 31, an extraction unit 32, and a learning unit 34. The detection unit 31 detects a lead-lag interval between a dependent variable and each of a plurality of explanatory variable candidates. The extraction unit 32 extracts, from the plurality of explanatory variable candidates, explanatory variable candidates for which the lead-lag interval has been detected by the detection unit 31 as explanatory variables. The learning unit 34 learns a prediction model that outputs future values ​​of the dependent variable based on past time-series values ​​of the dependent variable and past time-series values ​​of the explanatory variables extracted by the extraction unit 32. The prediction model f described above is an example of a prediction model. This enables the information processing device 1 to improve the prediction accuracy of a model that outputs future values ​​of the dependent variable.

[0067] The information processing device 1 also includes a determination unit 33. The determination unit 33 determines a window width w of the prediction model f based on the lead-lag interval between each of the explanatory variables and the dependent variable extracted by the extraction unit 32. The learning unit 34 performs learning of the prediction model f using the window width w determined by the determination unit 33 as the window width of the prediction model. This allows the information processing device 1 to determine the learning window width, which was previously determined by search, based on the lead-lag interval, making it easier to determine the learning window width and further improving the prediction accuracy of the prediction model that outputs future values ​​of the dependent variable.

[0068] The information processing device 1 also includes a prediction unit 36 ​​that predicts future values ​​of the dependent variable using a trained model that receives time-series values ​​of the dependent variable and time-series values ​​of the explanatory variables as inputs and outputs future values ​​of the dependent variable. The prediction model f described above is an example of a trained model. The trained model is a model trained based on past time-series values ​​of the explanatory variables, each of which has a lead-lag interval detected between it and the dependent variable, and past time-series values ​​of the dependent variable. This allows the information processing device 1 to accurately predict future values ​​of the dependent variable.

[0069] [System Configuration, etc.] The components of each unit shown in the figure are conceptual functional units and do not necessarily have to be physically configured as shown. In other words, the specific form of distribution and integration of each device is not limited to that shown, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc. Furthermore, all or any part of the processing functions performed by each device can be realized by a CPU and a program executed by the CPU, or can be realized as hardware using wired logic.

[0070] Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using a known method.In addition, the information including the processing procedures, control procedures, specific names, various data and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified.

[0071] [Program] A program written in a computer-executable language may be created to execute the processes performed by the information processing device 1 according to the above embodiment. In one embodiment, the information processing device 1 can be implemented by installing an information processing program that executes the above information processing as package software or online software on a desired computer. For example, by executing the above information processing program on the information processing device, the information processing device can function as the information processing device 1. The information processing device referred to here includes desktop and notebook personal computers. Other examples of information processing devices include mobile communication terminals such as smartphones, mobile phones, and PHS (Personal Handyphone Systems), as well as slate terminals such as PDAs (Personal Digital Assistants). The functions of the information processing device 1 may also be implemented on a cloud server.

[0072] 9 is a diagram showing an example of a computer that executes an information processing program. The computer 1000 includes, for example, a memory 1010, a CPU 1020, a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.

[0073] The memory 1010 includes a ROM (Read Only Memory) 1011 and a RAM 1012. The ROM 1011 stores a boot program such as a BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to a hard disk drive 1031. The disk drive interface 1040 is connected to a disk drive 1041. A removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1041. The serial port interface 1050 is connected to a mouse 1051 and a keyboard 1052, for example. The video adapter 1060 is connected to a display 1061, for example.

[0074] Here, the hard disk drive 1031 stores, for example, an OS (Operating System) 1091, an application program 1092, a program module 1093, and program data 1094. The various pieces of information described in the above embodiments are stored in the hard disk drive 1031 or the memory 1010, for example.

[0075] The information processing program is stored in the hard disk drive 1031 as, for example, a program module 1093 in which instructions to be executed by the computer 1000 are written. Specifically, the program module 1093 in which each process executed by the information processing device 1 described in the above embodiment is written is stored in the hard disk drive 1031.

[0076] Furthermore, data used for information processing by the information processing program is stored as program data 1094, for example, in the hard disk drive 1031. Then, the CPU 1020 reads the program module 1093 and the program data 1094 stored in the hard disk drive 1031 into the RAM 1012 as necessary, and executes each of the above-described procedures.

[0077] The program module 1093 and program data 1094 related to the information processing program are not limited to being stored in the hard disk drive 1031, and may be stored in, for example, a removable storage medium and read by the CPU 1020 via the disk drive 1041. Alternatively, the program module 1093 and program data 1094 related to the information processing program may be stored in another computer connected via a network such as a LAN or a WAN (Wide Area Network), and read by the CPU 1020 via the network interface 1070.

[0078] Although the present invention has been described above as an embodiment, the present invention is not limited to the description and drawings that form part of the disclosure of the present invention. In other words, other embodiments, examples, and operational techniques that can be made by those skilled in the art based on the present invention are all included in the scope of the present invention.

[0079] REFERENCE SIGNS LIST 1 Information processing device 10 Communication unit 11 Storage unit 12 Processing unit 20 Learning data storage unit 21 Prediction model storage unit 30 Learning data acquisition unit 31 Detection unit 32 Extraction unit 33 Determination unit 34 Learning unit 35 Prediction data acquisition unit 36 ​​Prediction unit 37 Prediction result output unit

Claims

1. An information processing device comprising: a detection unit that detects a lead-lag interval between a dependent variable and each of a plurality of candidate explanatory variables; an extraction unit that extracts, from the plurality of candidate explanatory variables, a candidate explanatory variable for which the lead-lag interval has been detected by the detection unit, as an explanatory variable; and a learning unit that learns a model that outputs a future value of the dependent variable based on past time series values ​​of the dependent variable and the past time series values ​​of the explanatory variables extracted by the extraction unit.

2. An information processing device comprising: a prediction unit that predicts future values ​​of a dependent variable using a trained model that receives time series values ​​of a dependent variable and time series values ​​of an explanatory variable as input and outputs future values ​​of the dependent variable, wherein the trained model is a model trained based on past time series values ​​of explanatory variables, each of which has a lead-lag interval detected between it and the dependent variable, and past time series values ​​of the dependent variable.

3. An information processing method executed by an information processing device, comprising: a detection step of detecting a lead-lag interval between a dependent variable and each of a plurality of candidate explanatory variables; an extraction step of extracting, from the plurality of candidate explanatory variables, a candidate explanatory variable for which the lead-lag interval has been detected by the detection step, as an explanatory variable; and a learning step of learning a model that outputs a future value of the dependent variable based on the past time series values ​​of the dependent variable and the past time series values ​​of the explanatory variables extracted by the extraction step.

4. An information processing program that causes a computer to execute the following steps: a detection step that detects a lead-lag interval between a dependent variable and each of a plurality of candidate explanatory variables; an extraction step that extracts, from the plurality of candidate explanatory variables, a candidate explanatory variable for which the lead-lag interval has been detected by the detection step, as an explanatory variable; and a learning step that learns a model that outputs a future value of the dependent variable based on past time series values ​​of the dependent variable and the past time series values ​​of the explanatory variables extracted by the extraction step.

Images