Information processing device, program and method

Abstract

To achieve effective time-series prediction in terms of prediction accuracy and calculation speed.SOLUTION: An information processing device comprises: a dataset creation unit for creating a plurality of datasets from time-series data consisting of a plurality of element data; a time-series prediction unit for performing time-series prediction using each of the plurality of datasets; and a synthesis unit for obtaining a prediction result by synthesizing a plurality of partial results of the time-series prediction obtained by the time-series prediction unit, wherein the datasets include two or more element data which are discrete in the time-series data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

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

[0002] In recent years, time series prediction has been performed for various purposes. Time series prediction is a process of analyzing time series data having causal relationships along a time series and predicting the future. Regarding such time series prediction, Patent Document 1 discloses a technology for speeding up time series prediction. More specifically, Patent Document 1 discloses a technology for dividing time series data into certain chronologically consecutive intervals and performing time series prediction for each of the divided data obtained by the division. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2006-252394 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the above technology, the time series data is divided into fixed intervals, which means that the long-term trend components of the time series data are lost.Furthermore, if the division width is too fine, the regression components necessary for time series prediction are lost, making it difficult to perform time series prediction.

[0005] Therefore, the present invention has been made in consideration of the above problems, and an object of the present invention is to provide a technology that realizes effective time series prediction in terms of prediction accuracy and calculation speed. [Means for solving the problem]

[0006] In order to solve the above problem, according to one aspect of the present invention, there is provided an information processing device including: a dataset creation unit that creates a plurality of datasets from time series data composed of a plurality of element data; a time series prediction unit that performs time series prediction using each of the plurality of datasets; and a synthesis unit that obtains a prediction result by synthesizing a plurality of partial results of the time series prediction obtained by the time series prediction unit, wherein the datasets include two or more element data that are discrete in the time series data.

[0007] The apparatus may further include a moving average processing unit that generates the time series data by performing moving average processing on the original time series data.

[0008] The method may further include an acquisition unit that acquires a number of periodic elements indicating the number of element data constituting one period of the original time series data, and the dataset creation unit may create the plurality of datasets such that element data in the time series data, each having an interval corresponding to the number of periodic elements, constitute the same dataset.

[0009] The number according to the number of periodic elements may be a divisor of the number of periodic elements.

[0010] The moving average processor may perform the moving average process using a number corresponding to the number of periodic elements as a moving average number.

[0011] The information processing device may further include a verification data creation unit that creates verification data from the time series data, the verification data including element data different from the element data that constitutes the plurality of data sets, wherein the time series prediction unit and the synthesis unit each perform processing in which a different number of data sets are used as the plurality of data sets, and the information processing device may further include a verification unit that verifies the accuracy of each of the two or more prediction results obtained by the synthesis unit using the verification data.

[0012] The verification unit may determine the number of datasets to be used as the plurality of datasets in subsequent processing based on the results of the verification, and the time series processing unit and the synthesis unit may perform processing in which the number of datasets determined by the verification unit is used as the plurality of datasets in subsequent processing.

[0013] The verification unit may determine the number of data sets from which, among two or more prediction results, the prediction result with the shortest calculation time among prediction results whose accuracy exceeds a threshold, or the prediction result with the highest accuracy among prediction results whose calculation time is less than a threshold, is obtained.

[0014] The synthesis unit may synthesize the plurality of partial results of the time series prediction obtained by the time series prediction unit, and further obtain the prediction result by interpolating data between element data constituting the synthesis result.

[0015] In order to solve the above-mentioned problem, according to another aspect of the present invention, there is provided a program for causing a computer to function as an information processing device, comprising: a dataset creation unit that creates a plurality of datasets from time series data composed of a plurality of element data; a time series prediction unit that performs time series prediction using each of the plurality of datasets; and a synthesis unit that obtains a prediction result by synthesizing results of the plurality of time series predictions obtained by the time series prediction unit, wherein the datasets include two or more element data that are discrete in the time series data.

[0016] In order to solve the above-mentioned problem, according to another aspect of the present invention, there is provided a computer-executable method including: creating a plurality of data sets from time series data composed of a plurality of element data; performing time series prediction using each of the plurality of data sets; and obtaining a prediction result by combining the results of the plurality of time series predictions obtained by the time series prediction, wherein the data sets include two or more element data that are discrete in the time series data. [Effects of the Invention]

[0017] According to the present invention as described above, it is possible to realize effective time series prediction in terms of prediction accuracy and calculation speed. [Brief explanation of the drawings]

[0018] [Figure 1] FIG. 1 is an explanatory diagram showing a configuration of an information processing device 10 according to a first embodiment of the present invention. [Figure 2] 10 is an explanatory diagram showing a specific example of moving average processing by the moving average processing unit 140. FIG. [Figure 3] FIG. 2 is an explanatory diagram showing an example of a dataset created by a dataset creating unit 150. [Figure 4] 10 is an explanatory diagram showing a specific example of time series prediction by the time series prediction unit 160. FIG. [Figure 5] 10 is an explanatory diagram showing a specific example of synthesis and interpolation by a synthesis unit 170. FIG. [Figure 6] 3 is a flowchart showing the operation of the information processing device 10 according to the first embodiment of the present invention. [Figure 7] FIG. 2 is an explanatory diagram showing the accuracy and calculation time of the proposed method according to the first embodiment of the present invention. [Figure 8] FIG. 10 is an explanatory diagram showing the transition results of prediction results, etc. [Figure 9] FIG. 10 is an explanatory diagram showing the transition results of prediction results, etc. [Figure 10] FIG. 10 is an explanatory diagram showing the configuration of an information processing device 10-2 according to a second embodiment of the present invention. [Figure 11] 10 is a flowchart showing the operation of an information processing device 10-2 according to the second embodiment of the present invention. [Figure 12] FIG. 1 is a block diagram showing an example of a hardware configuration 90. DETAILED DESCRIPTION OF THE INVENTION

[0019] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0020] Embodiments of the present invention relate to an information processing device that performs time series prediction based on time series data. For example, an information processing device 10 according to a first embodiment performs time series prediction using multiple data sets created from time series data. An information processing device 10-2 according to a second embodiment appropriately determines the number of data sets to be used as the multiple data sets in the time series prediction. The first and second embodiments of the present invention will be described in detail below.

[0021] <<First embodiment>> <Configuration of information processing device 10> 1 is an explanatory diagram showing the configuration of an information processing device 10 according to a first embodiment of the present invention. As shown in Fig. 1, the information processing device 10 according to one embodiment of the present invention includes a storage unit 110, a time-series data acquisition unit 120, a period estimation unit 130, a moving average processing unit 140, a data set creation unit 150, a time-series prediction unit 160, a synthesis unit 170, and a prediction result output unit 180.

[0022] (Storage unit 110) The storage unit 110 stores various information and data used in the operation of the information processing device 10. For example, as shown in FIG. 1, the storage unit 110 stores time-series data, the number of periodic elements, multiple data sets, and multiple partial results. Time-series data is data in which multiple element data are arranged in a chronological order. In an embodiment of the present invention, the main focus may be on time-series data in which the content of each element data has a causal relationship with a time position. Examples of such time-series data include time-series data having element data such as temperature or humidity measured every hour at a certain location. Other examples of time-series data include stock prices, business sales, traffic volume, or the number of visits to a website.

[0023] The number of periodic elements, the data set, and the partial results will be mentioned in the explanation of other configurations.

[0024] (Time series data acquisition unit 120) The time series data acquisition unit 120 is configured to acquire time series data and store the acquired time series data in the storage unit 110. The time series data acquired by the time series data acquisition unit 120 and stored in the storage unit 110 is an example of original time series data. The method by which the time series data acquisition unit 120 acquires time series data is not particularly limited. For example, the time series data acquisition unit 120 may be connected to an observation means such as a sensor via a wired or wireless connection and acquire time series data from the sensor. Alternatively, the time series data acquisition unit 120 itself may be a sensor. Alternatively, the time series data acquisition unit 120 may acquire time series data input manually. Furthermore, the time series data acquisition unit 120 may acquire time series data by communicating with another device that stores time series data.

[0025] (Period estimation unit 130) The period estimation unit 130 estimates periodic components suitable for time series prediction using the time series data stored in the storage unit 110. For example, the period estimation unit 130 may estimate periodic components based on an index (e.g., AIC (Akaike Information Criterion)) for searching for periodic components suitable for time series prediction. The period estimation unit 130 stores the estimated periodic components in the storage unit 110. Hereinafter, the periodic component is defined as the number of periodic elements indicating the number of element data constituting one period of the time series data. For example, if the time series data stored in the storage unit 110 is time series data of temperature measured every hour, the number of periodic elements may be estimated to be 365 × 24 = 8760, which corresponds to one year's worth of element data, and stored in the storage unit 110.

[0026] However, if the number of periodic elements of the time series data stored in storage unit 110 is known, the known number of periodic elements may be used without estimating the number of periodic elements by period estimation unit 130. For example, if the time series data stored in storage unit 110 is time series data of temperature in a certain city measured every hour, it is known that the temperature period is also one year because the Earth's orbital period is one year, so 365 × 24 = 8760, which corresponds to one year's worth of element data, may be used as the number of periodic elements.

[0027] (Moving average processing unit 140) The moving average processing unit 140 reads out the time series data from the storage unit 110, performs moving average processing on the time series data, and obtains the time series data after the moving average. For example, the moving average processing unit 140 performs the moving average processing using the number of datasets (N) created by the dataset creation unit 150 (described later) as the moving average number.

[0028] Fig. 2 is an explanatory diagram showing a specific example of moving average processing by the moving average processor 140. When the number of moving averages is 24, the moving average processor 140 calculates the average for each of 24 consecutive element data as shown in Fig. 2. For example, the average value of group G1 consisting of 24 element data becomes element data 0 in the time series data after the moving average, the average value of group G2 becomes element data 1 in the time series data after the moving average, and the average value of group G3 becomes element data 2 in the time series data after the moving average.

[0029] (Dataset Creation Unit 150) The dataset creation unit 150 creates two or more datasets from the time series data after moving averaging obtained by the moving average processing unit 140. For example, the dataset creation unit 150 creates two or more datasets such that element data having a certain interval between each other in the time series data after moving averaging constitute the same dataset. Note that the two or more datasets created by the dataset creation unit 150 include multiple datasets used for time series prediction by the time series prediction unit 160, which will be described later.

[0030] The above-mentioned constant interval used to create two or more data sets may be a number corresponding to the number of periodic elements. In this case, the data set creation unit 150 may create the same data set for data elements that have the same remainder when the data number of each data element in the time series data after moving averaging is divided by the number corresponding to the number of periodic elements. Note that the number corresponding to the number of periodic elements may be a divisor of the number of periodic elements. For example, if the number of periodic elements is 8760, the number corresponding to the number of periodic elements may be 24.

[0031] 3 is an explanatory diagram showing an example of dataset creation by the dataset creation unit 150. When the number corresponding to the number of periodic elements is 24, the dataset creation unit 150 creates dataset 0, which is made up of element data such as element data 0 and element data 24, whose data numbers are divided by 24 and leave a remainder of "0", as shown in FIG. 3. Similarly, the dataset creation unit 150 creates dataset 1, which is made up of element data such as element data 1 and element data 25, whose data numbers are divided by 24 and leave a remainder of "1". The dataset creation unit 150 creates datasets 0 to 23 by similar processing.

[0032] The data set creating unit 150 stores the two or more data sets that it has created in the storage unit 110.

[0033] (Time series prediction unit 160) The time series prediction unit 160 performs time series prediction using the dataset created by the dataset creation unit 150. The time series prediction unit 160 does not need to perform time series prediction on all of the two or more datasets created by the dataset creation unit 150, but performs time series prediction on each of a set number K of datasets among the two or more datasets. The set number K may be stored in the storage unit 110, may be input by the user of the information processing device 10, or may be determined randomly.

[0034] Which dataset is used as the set number K of datasets may be selected manually by the user of the information processing device 10, or may be selected by the time series prediction unit 160. For example, since equal interval prediction is preferable, the time series prediction unit 160 may select a dataset that satisfies Equation 1 from among dataset numbers A (A=0, 1, 2,...N). A≡M(mod(N / K)) (Formula 1)

[0035] In the above formula 1, N is the number of datasets, two or more, created by the dataset creation unit 150. M may be any value among M=0, 1, 2...((N / K)-1). According to the above formula 1, when the set number K is 2 and N=24, the time series prediction unit 160 may select dataset 0 and dataset 12.

[0036] The time series prediction unit 160 performs time series prediction using the selected set number K of data sets. First, the time series prediction unit 160 estimates a model and parameters suitable for the data set. The time series prediction unit 160 may use any model as the prediction model depending on the characteristics of the data. Basically, any time series prediction model that outputs a prediction result for input time series data can be used. Furthermore, the time series prediction unit 160 may use any parameter as the parameter depending on the prediction model. Then, the time series prediction unit 160 performs time series prediction for each of the set number K of data sets using the estimated model and parameters.

[0037] 4 is an explanatory diagram showing a specific example of time series prediction by the time series prediction unit 160. As shown in FIG. 4, the time series prediction unit 160 can perform time series prediction on each of data set 0 and data set 12. As a result of time series prediction on data set 0, element data 8760, 8784, 8808, etc. corresponding to future time points in data set 0 are obtained as partial prediction results. Similarly, as a result of time series prediction on data set 12, element data 8772, 8796, 8820, etc. corresponding to future time points in data set 12 are obtained as partial prediction results. Note that the number of element data to be predicted by the time series prediction unit 160 may be determined arbitrarily depending on the accuracy of the prediction model.

[0038] When the time series prediction unit 160 has completed the time series prediction for all of the set number K of data sets, it stores in the storage unit 110 the partial results obtained by the time series prediction for each data set.

[0039] (Synthesizer 170) The synthesis unit 170 reads out the multiple partial results stored in the storage unit 110 and synthesizes the multiple partial results to obtain a prediction result. Specifically, the synthesis unit 170 combines the multiple partial results and arranges each element data in ascending chronological order. Furthermore, if the set number K is smaller than the number N of created data sets, the synthesis unit 170 performs interpolation processing. Synthesis and interpolation will be described below with reference to FIG. 5.

[0040] Fig. 5 is an explanatory diagram showing a specific example of synthesis and interpolation by the synthesis unit 170. As shown in the upper part of Fig. 5, when a partial result (0) based on data set 0 and a partial result (12) based on data set 12 are obtained, the synthesis unit 170 synthesizes these partial results (0) and (12) to obtain the synthesis result shown in the middle part of Fig. 5.

[0041] However, because the synthesis result is obtained based on a portion of the data sets, the time intervals of the element data in the synthesis result are longer than the time intervals of the element data in the time-series data after moving averaging created by the moving average processing unit 140. For this reason, the synthesis unit 170 obtains a prediction result by interpolating (N / K)-1 pieces of element data between the element data of the synthesis result, as shown in the lower part of Fig. 5. In the lower part of Fig. 5, for example, element data 8761 to element data 8771 are added by interpolation between element data 8760 and element data 8772 that make up the synthesis result.

[0042] As a specific example of the interpolation process, linear interpolation will be explained. t and the t+1th element data x t+1 When interpolating data between x' and x', the data created by linear interpolation is t’ Then, linear interpolation data can be created using Equation 2. x' t’ ={t' / ((N / K)+1)}×(x t+1 -x t )+x t (Formula 2)

[0043] Here, t' is the data to be additionally generated, and t'=1, 2, . . . (N / K). The synthesis unit 170 synthesizes the additionally generated data as the t-th element data x t and the t+1th element data x t+1 Insert between.

[0044] The synthesis unit 170 may perform interpolation using other methods such as cubic spline interpolation and LOCF (Last Observation Carried Forward). When the set number K is manually selected, the synthesis unit 170 may perform interpolation as appropriate depending on the time interval between predictions.

[0045] (Prediction result output unit 180) The prediction result output unit 180 outputs the prediction result obtained by the synthesis unit 170. For example, the prediction result output unit 180 may output the prediction result obtained by the synthesis unit 170 by display or sound, may transmit the prediction result to another device, or may store it in the storage unit 110.

[0046] <Operation of the information processing device 10> The configuration of the information processing device 10 according to the first embodiment of the present invention has been described above. Next, the operation of the information processing device 10 according to the first embodiment of the present invention will be summarized with reference to FIG.

[0047] 6 is a flowchart showing the operation of the information processing device 10 according to the first embodiment of the present invention. As shown in FIG. 6, first, the time series data acquisition unit 120 acquires time series data and stores the time series data in the storage unit 110 (S410). Next, the period estimation unit 130 estimates the number of periodic elements in the time series data and stores the estimation result in the storage unit 110 (S420).

[0048] Then, the moving average processing unit 140 reads out the time series data from the storage unit 110, performs moving average processing on the time series data, and obtains the time series data after the moving average (S430).

[0049] Thereafter, the data set creation unit 150 creates two or more data sets from the time series data after moving average obtained by the moving average processing unit 140, and stores the two or more created data sets in the storage unit 110 (S440).

[0050] Next, the time series prediction unit 160 selects a set number K of data sets, and estimates a model and parameters suitable for the data sets (S452).The time series prediction unit 160 then performs time series prediction on each of the set number K of data sets using the estimated model and parameters, and stores the partial results obtained by time series prediction on each data set in the storage unit 110 (S454).

[0051] Thereafter, the synthesis unit 170 reads out the plurality of partial results stored in the storage unit 110 and synthesizes the plurality of partial results (S460). Furthermore, if the set number K is smaller than the number of created data sets N (S470 / YES), the synthesis unit 170 performs interpolation processing on the synthesis result (S472). Then, the prediction result output unit 180 outputs the prediction result obtained by the synthesis unit 170 (S480), and the operation ends.

[0052] <Action and effect> According to the first embodiment of the present invention described above, various operational effects can be obtained. For example, according to the first embodiment of the present invention, time series prediction is performed on each of two or more data sets created so that element data having a fixed interval between each other constitute the same data set in time series data, and a prediction result is obtained by combining multiple partial results of the time series prediction. This makes it possible to perform predictions with high accuracy and at time intervals equivalent to the resolution of the data while reducing the amount of calculation. Specific examples of such effects will be described below.

[0053] The original time series data was hourly temperature data for Tokyo from January 1, 2018 to December 31, 2021, with the number of periodic elements being 365 x 24 and the number of datasets N set to 24, and each of the processes described above was performed. For this reason, a moving average process (number of moving averages = 24) was performed on the original time series data, and the time series data after the moving average contained element data that was the average value of the 24 element data of the original time series data. Furthermore, 24 datasets were created from the time series data after the moving average.

[0054] Then, the time series prediction unit 160 uses the AR model to perform time series prediction from January 1, 2022 to December 31, 2022. At that time, the accuracy is verified for each of the set numbers K = 2, 4, 6, 12, and 24. When the set number K is smaller than the number of created datasets N, the method shown in Equation 2 is used for the interpolation process.

[0055] FIG. 7 is an explanatory diagram showing the accuracy and calculation time of the proposed method of the first embodiment of the present invention. FIGS. 8 and 9 are explanatory diagrams showing the transition results of the forecast results, etc. In FIGS. 7 and 8, the results of verifying time series data after moving average using the ARMA model are also shown for comparison. In the ARMA model, the autoregressive component was 365 × 24, and the moving average component was 24. The accuracy was calculated using MAPE (Mean Absolute Percentage Error) using actual time series data of temperature from January 1, 2022 to December 31, 2022 as verification data. Note that when N>K, the accuracy may fluctuate depending on the dataset used. Therefore, in the verification, when N>K, the value of M in Equation 1 was set to M=0, 1, 2, ... We verified all patterns ((N / K)-1), and Figure 7 shows the average MAPE value for each pattern. Figures 8 and 9 show the prediction results for M=0.

[0056] Referring to Figure 7, the ARMA model for comparison has a MAPE of 24.560%, while the proposed method shows a deterioration in accuracy regardless of the number of datasets used (setting number K). This is thought to be due to the fact that each dataset consists of element data that is a portion of the time series data after moving averaging, and therefore the amount of information in each dataset is reduced compared to the time series data after moving averaging.

[0057] However, the degree of accuracy degradation is limited to approximately 0.1 to 1.5%. Looking at the trend graphs in Figures 8 and 9, it can be seen that there is no significant difference between the trends of the proposed method and the ARMA model, indicating that the degree of accuracy degradation is small. Furthermore, the proposed method significantly reduces the calculation time compared to the ARMA model, demonstrating that the amount of calculation has been reduced.

[0058] Next, we compare the prediction results for different settings of the number K using the proposed method. It can be seen that when the setting number K is reduced, the area where linear interpolation processing is performed increases, and so there is a basic tendency for accuracy to deteriorate. However, as shown in Figures 8 and 9, there is no significant difference in the prediction progress. This is thought to be because the time series data used here was data that could be sufficiently predicted even with an ARMA model, so the deterioration in accuracy due to linear interpolation was not evident.

[0059] Furthermore, when the setting number K=2, accuracy improved compared to the other cases, even though the number of data sets used was smaller than in the other cases. This is thought to be because the time series data used was data that could be sufficiently predicted even with an ARMA model. Because the time series data used showed a monotonous trend, the interpolated prediction results did not deviate from the true value, and no significant deterioration in accuracy was observed. Regarding calculation time, it can be confirmed that changing the setting number K reduces the amount of calculation depending on the number of data sets. However, in this verification, predictions were made using iterative processing. It is possible that calculation time can be further reduced by making predictions parallel processing rather than iterative processing.

[0060] As described above, the proposed method significantly reduces calculation time compared to other prediction methods such as the ARMA model, although there is a decrease in accuracy. Furthermore, changing the setting number K changes the relationship between calculation time and accuracy.

[0061] Furthermore, according to the first embodiment of the present invention, the dataset creation unit 150 creates a dataset of a divisor of the number of periodic elements in the time series data after moving averaging so that element data having intervals of a divisor of the number of periodic elements from each other constitute the same dataset. This configuration allows element data of the same phase in each period of the time series data to be included in the same dataset. Therefore, the accuracy of partial results obtained from such datasets and prediction results obtained from multiple partial results can be improved.

[0062] Furthermore, according to the first embodiment of the present invention, the moving average processing unit 140 performs moving average processing on the time series data to obtain the time series data after moving average. With this configuration, the influence of noise on the prediction result is suppressed, which is thought to improve the accuracy of the prediction result.

[0063] Furthermore, the moving average processor 140 performs moving average processing on the original time series data using a divisor of the number of periodic elements used by the dataset creator 150 as the moving average number. In this case, there are no gaps in the element data in the original time series data that serve as the averaging elements for the element data of each dataset. In the example shown in FIGS. 2 and 3, the averaging element for element data 0 in dataset 0 is element data 0 to 23 in the original time series data, and the averaging element for the next element data in dataset 0 is element data 24 to 47 in the original time series data, with no gaps between the respective averaging elements. With this configuration, further improvement in the accuracy of the prediction results is expected.

[0064] <<Second embodiment>> The first embodiment of the present invention has been described above. As mentioned above, changing the set number K changes the relationship between calculation time and accuracy. Therefore, it is useful to appropriately determine the set number K. Therefore, a second embodiment relating to a method for appropriately determining the set number K will be described below.

[0065] <Configuration of information processing device 10-2> Fig. 10 is an explanatory diagram showing the configuration of an information processing device 10-2 according to the second embodiment of the present invention. As shown in Fig. 10, the information processing device 10-2 according to the second embodiment of the present invention includes a storage unit 110, a time-series data acquisition unit 120, a period estimation unit 130, a moving average processing unit 140, a data set creation unit 150, a time-series prediction unit 160, a synthesis unit 170, and a prediction result output unit 180, as well as a verification data creation unit 210 and a verification unit 220. The following mainly describes the configurations of the verification data creation unit 210 and the verification unit 220.

[0066] (Verification data creation unit 210) The verification data creation unit 210 creates verification data from the time-series data after moving averaging obtained by the moving average processing unit 140. For example, the verification data creation unit 210 may divide the time-series data after moving averaging into two pieces at a certain element data, and use the latter of the two pieces of data obtained by the division as verification data. In this case, the dataset creation unit 150 creates two or more datasets from the former data. In other words, the element data that make up the verification data is different from the element data that make up the two or more datasets created by the dataset creation unit 150.

[0067] The time series prediction unit 160 and the synthesis unit 170 perform the time series prediction and synthesis described in the first embodiment for each of the different numbers of data sets, and obtain two or more prediction results.

[0068] (Verification unit 220) The verification unit 220 verifies the accuracy of each of the two or more prediction results obtained by the time series prediction unit 160 and the synthesis unit 170 by comparing each prediction result with verification data. For example, the verification unit 220 calculates the accuracy using MAPE (Mean Absolute Proportional Error) or RMSE (Root Mean Squared Error). Then, the verification unit 220 determines the set number K based on the verification result. The set number K determined by the verification unit 220 is used by the time series prediction unit 160, the synthesis unit 170, etc. in subsequent processing.

[0069] For example, the verification unit 220 may determine, as the set number K, the number of data sets that have yielded the prediction result with the shortest calculation time among prediction results whose accuracy exceeds a preset threshold, or the prediction result with the highest accuracy among prediction results whose calculation time is less than a preset threshold. The verification unit 220 stores the determined set number K in the storage unit 110.

[0070] <Operation of information processing device 10-2> The configuration of the information processing device 10-2 according to the second embodiment of the present invention has been described above. Next, the operation of the information processing device 10-2 according to the second embodiment of the present invention will be summarized.

[0071] Fig. 11 is a flowchart showing the operation of the information processing device 10-2 according to the second embodiment of the present invention. In Fig. 11, the processes of S410 to S430 are the same as those described with reference to Fig. 6, and therefore detailed description thereof will be omitted here.

[0072] 11, when the moving average processing unit 140 obtains the time series data after the moving average, the verification data creating unit 210 creates verification data from the time series data after the moving average (S510). Thereafter, the dataset creating unit 150 creates two or more datasets from the time series data after the moving average, excluding the verification data, and stores the two or more created datasets in the storage unit 110 (S440).

[0073] Thereafter, the time series prediction unit 160 and the synthesis unit 170 perform the processes of S452 to S472 shown in FIG. 6 for each of the different numbers of data sets, thereby obtaining two or more prediction results (S520).

[0074] Next, the verification unit 220 verifies the accuracy of each of the two or more prediction results obtained by the time series prediction unit 160 and the synthesis unit 170 by comparing each prediction result with the verification data (S530).Then, the verification unit 220 determines a set number K based on the verification results and stores the set number K in the storage unit 110 (S540).

[0075] <Action and effect> The second embodiment of the present invention described above provides a variety of advantageous effects. For example, the verification unit 220 can verify two or more prediction results obtained using different numbers of data sets using verification data that are true values, thereby appropriately determining the set number K to be used in subsequent processing.

[0076] For example, the verification unit 220 may determine, as the set number K, the number of data sets that have yielded the prediction result with the shortest calculation time among prediction results whose accuracy exceeds a preset threshold, or the prediction result with the highest accuracy among prediction results whose calculation time is less than a preset threshold. With this configuration, it is possible to achieve both accuracy and calculation time.

[0077] <<Hardware configuration>> The above describes the embodiments of the present invention. The above-described information processing, such as the creation of a data set and time series prediction, is realized by a combination of software and hardware. Below, we will explain an example of a hardware configuration that can be applied to the information processing device 10 and the information processing device 10-2.

[0078] 12 is a block diagram showing an example of a hardware configuration 90. The hardware configuration 90 includes a CPU (Central Processing Unit) 901, a ROM (Read Only Memory) 902, a RAM (Random Access Memory) 903, and a host bus 904. The hardware configuration 90 also includes a bridge 905, an external bus 906, an interface 907, an input device 908, a display device 909, an audio output device 910, a storage device (HDD) 911, a drive 912, and a network interface 915.

[0079] The CPU 901 functions as an arithmetic processing unit and a control unit, and controls overall operation in accordance with various programs. Alternatively, the CPU 901 may be a microprocessor. The ROM 902 stores programs used by the CPU 901, calculation parameters, etc. The RAM 903 temporarily stores programs used in the execution of the CPU 901, parameters that change as appropriate during the execution, etc. These are interconnected by a host bus 904, which is comprised of a CPU bus or the like. Cooperation between the CPU 901, ROM 902, and RAM 903 and software can realize functions such as a period estimation unit 130, a moving average processing unit 140, a data set creation unit 150, a time series prediction unit 160, a synthesis unit 170, a prediction result output unit 180, a verification data creation unit 210, and a verification unit 220.

[0080] The host bus 904 is connected to an external bus 906, such as a PCI (Peripheral Component Interconnect / Interface) bus, via a bridge 905. It is not necessary to configure the host bus 904, bridge 905, and external bus 906 separately, and these functions may be implemented on a single bus.

[0081] The input device 908 is composed of input means such as a mouse, keyboard, touch panel, buttons, microphone, sensors, switches, and levers that allow the user to input information, and an input control circuit that generates an input signal based on the user's input and outputs it to the CPU 901. By operating the input device 908, the user can input various types of data and instruct processing operations.

[0082] The display device 909 includes, for example, a display device such as a liquid crystal display (LCD) device, a projector device, an OLED (Organic Light Emitting Diode) device, a lamp, etc. The audio output device 910 includes an audio output device such as a speaker and a headphone.

[0083] The storage device 911 is a data storage device configured as an example of the storage unit 110 according to this embodiment. The storage device 911 may include a storage medium, a recording device that records data on the storage medium, a reading device that reads data from the storage medium, and a deleting device that deletes data recorded on the storage medium. The storage device 911 is configured, for example, by an HDD (Hard Disk Drive) or an SSD (Solid State Drive), or a memory having equivalent functions. This storage device 911 drives storage and stores programs executed by the CPU 901 and various data.

[0084] The drive 912 is a reader / writer for a storage medium, and is built into or externally attached to the hardware configuration 90. The drive 912 reads information recorded on a removable storage medium 84, such as an attached magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, and outputs the information to the RAM 903 or the storage device 911. The drive 912 can also write information to the removable storage medium 84.

[0085] The network interface 915 is, for example, a communication interface configured with a communication device for connecting to a network, etc. The network interface 915 may be a wireless LAN (Local Area Network) compatible communication device or a wired communication device that performs wired communication.

[0086] <<Supplementary Information>> Although the preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, the present invention is not limited to these examples. It is clear that a person skilled in the art to which the present invention pertains can conceive of various modifications and alterations within the scope of the technical ideas set forth in the claims, and it is understood that these also naturally fall within the technical scope of the present invention.

[0087] For example, the steps in the processing of information processing device 10 and information processing device 10-2 in this specification do not necessarily have to be processed in chronological order according to the order described in the flowcharts. For example, the steps in the processing of information processing device 10 and information processing device 10-2 may be processed in an order different from the order described in the flowcharts, or may be processed in parallel.

[0088] It is also possible to create a computer program that causes hardware such as a CPU, ROM, and RAM built into the information processing device 10 and the information processing device 10-2 to perform functions equivalent to those of the above-described information processing device 10 and the information processing device 10-2. A non-transitory storage medium storing the computer program is also provided.

[0089] Furthermore, the functions of the storage unit 110, the time-series data acquisition unit 120, the period estimation unit 130, the moving average processing unit 140, the dataset creation unit 150, the time-series prediction unit 160, the synthesis unit 170, the prediction result output unit 180, the verification data creation unit 210, and the verification unit 220 may be implemented in a single housing or may be distributed across multiple housings. When the functions are implemented in a single housing, the device having that housing may correspond to the information processing device 10 or the information processing device 10-2. When the functions are distributed across multiple housings, a combination of one or more of the multiple devices having the housings may correspond to the information processing device 10 or the information processing device 10-2. When the functions are distributed across multiple housings, an interface for inputting and outputting information between the devices having the housings is also implemented. [Explanation of symbols]

[0090] 10. Information processing equipment 110 Storage section 120 Time series data acquisition section 130 Period estimation part 140 Moving average processing unit 150 Dataset Creation Department 160 Time Series Prediction Unit 170 Synthesis Section 180 Prediction result output section 210 Verification Data Creation Department 220 Verification Department

Claims

1. a dataset creation unit that creates a plurality of datasets from time series data that is composed of a plurality of element data; a time series prediction unit that performs time series prediction using each of the plurality of data sets; a synthesis unit that synthesizes a plurality of partial results of the time series prediction obtained by the time series prediction unit to obtain a prediction result; Equipped with The data set includes two or more discrete element data in the time series data.

2. The information processing device according to claim 1 , further comprising a moving average processing unit that generates the time series data by performing moving average processing on original time series data.

3. an acquisition unit that acquires a period element number indicating the number of element data constituting one period of the original time series data, The information processing device according to claim 2 , wherein the data set creation unit creates the plurality of data sets such that element data having intervals corresponding to the number of periodic elements in the time series data constitute the same data set.

4. The information processing device according to claim 3 , wherein the number according to the number of periodic elements is a divisor of the number of periodic elements.

5. The information processing device according to claim 4 , wherein the moving average processing unit performs the moving average processing using a number corresponding to the number of periodic elements as a moving average number.

6. a verification data creation unit that creates verification data from the time-series data, the verification data including element data different from element data constituting the plurality of data sets; the time series prediction unit and the synthesis unit perform processes in which different numbers of data sets are used as the plurality of data sets, The information processing apparatus according to claim 5 , further comprising a verification unit that verifies the accuracy of each of the two or more prediction results obtained by the synthesis unit using the verification data.

7. The verification unit determines the number of data sets to be used as the plurality of data sets in subsequent processing based on a result of the verification; The information processing device according to claim 6 , wherein the time series processing unit and the synthesis unit perform processing in which the number of data sets determined by the verification unit is used as the plurality of data sets in subsequent processing.

8. 8. The information processing device according to claim 7, wherein the verification unit determines the number of the data sets from which a prediction result having the shortest calculation time among prediction results whose accuracy exceeds a threshold, or a prediction result having the highest accuracy among prediction results whose calculation time is less than a threshold, is obtained, among two or more prediction results.

9. 9. The information processing device according to claim 8, wherein the synthesis unit synthesizes the plurality of partial results of the time series prediction obtained by the time series prediction unit, and further interpolates data between element data constituting the synthesis result to obtain the prediction result.

10. Computer, a dataset creation unit that creates a plurality of datasets from time series data that is composed of a plurality of element data; a time series prediction unit that performs time series prediction using each of the plurality of data sets; a synthesis unit that synthesizes the results of the time series predictions obtained by the time series prediction units to obtain a prediction result; Equipped with The data set includes two or more discrete element data in the time series data.

11. Creating multiple data sets from time series data consisting of multiple element data; performing a time series prediction using each of the plurality of data sets; obtaining a prediction result by combining the results of the plurality of time series predictions obtained by the time series prediction; Including, A computer-implemented method, wherein the data set includes two or more element data that are discrete in the time series data.

Images