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

Abstract

To avoid determination of an inappropriate set of model parameters caused by sudden abnormal data when evaluating a parameter set used for a process of predicting an autoregression model.SOLUTION: An information processing device 10 includes an index value calculation unit 12a and a parameter determination unit 12b. The index value calculation unit 12a calculates a value of an index set in advance for each of adjustment parameter sets for calculating a regression coefficient used to obtain a predicted value from past data. For each of the adjustment parameter sets, the parameter determination unit 12b selects the adjustment parameter set, based on the calculated index value and the index values of the adjustment parameter sets in the vicinity of the adjustment parameter set.SELECTED DRAWING: Figure 4

Description

Technical Field

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

Background Art

[0002] Conventionally, in a manufacturing site handling hazardous substances, etc., in order to prevent accidents in advance, time-series data has been analyzed and data prediction has been performed based on it. And, in order to make this prediction more accurate, it is necessary to adjust the parameters of the model for prediction.

[0003] However, since the parameter adjustment work is carried out by repeating trial and error based on the tendency of the prediction trend and the knowledge of engineers, it is a problem that it takes a lot of labor and time. Therefore, a method of performing the parameter adjustment work by AI (Artificial Intelligence) technology is known.

[0004] For example, a method of estimating dynamic parameters based on non-stationary input data and output data with the calculated static parameters as constraint conditions based on stationary input data and corresponding stationary output data (see, for example, Patent Document 1), or a method of optimizing a parameter set defining a plurality of non-harmonic signals for time-series data by an exhaustive search method (see, for example, Patent Document 2) is known as the prior art.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0006] However, the conventional techniques described above have the problem that they cannot avoid the possibility of calculating evaluation values ​​using outliers or other abnormal data when evaluating estimated parameters or each parameter set, and that an inappropriate parameter set may be determined due to sudden abnormal data. [Means for solving the problem]

[0007] To solve the above-mentioned problems and achieve the objective, the information processing apparatus of the present invention is characterized by comprising: an index value calculation unit that calculates a preset index value for each of the adjustment parameter sets used to calculate regression coefficients used to obtain predicted values ​​from past data; and a parameter determination unit that selects each of the adjustment parameter sets based on the calculated index value and the index values ​​of neighboring adjustment parameter sets. [Effects of the Invention]

[0008] According to the present invention, by introducing the concept of "distance (temporal position)" to the parameter set or dataset during the evaluation of the parameter set, it is possible to avoid determining an inappropriate set of model parameters due to sudden anomaly data. [Brief explanation of the drawing]

[0009] [Figure 1] Figure 1 is a diagram illustrating the overview of the method for calculating predicted values ​​and regression coefficients in the information processing according to the embodiment. [Figure 2] Figure 2 is a diagram illustrating the overview of the adjustment parameters in the information processing according to the embodiment. [Figure 3] Figure 3 is an explanatory diagram showing an overview of the information processing method according to Embodiment 1. [Figure 4] Figure 4 shows an example of the configuration of the information processing device according to Embodiment 1. [Figure 5] Figure 5 is a diagram showing an overview of the index values ​​in the information processing according to Embodiment 1. [Figure 6] Figure 6 is a diagram showing an overview of the index values ​​in the information processing according to Embodiment 1. [Figure 7] Figure 7 is a diagram showing an overview of the index values ​​in the information processing according to Embodiment 1. [Figure 8] Figure 8 is a diagram illustrating the overview of the process for selecting an adjustment parameter set based on a neighboring adjustment parameter set in the information processing according to Embodiment 1. [Figure 9] Figure 9 is a diagram showing the overall system flow in a specific example of information processing according to Embodiment 1. [Figure 10] Figure 10 shows an example of the settings that serve as prerequisites in a specific example of information processing according to Embodiment 1. [Figure 11] Figure 11 shows an example of the selection result of the adjustment parameter set in a specific example of information processing according to Embodiment 1. [Figure 12] Figure 12 shows an example of the selection result of the adjustment parameter set in a specific example of information processing according to Embodiment 1. [Figure 13] Figure 13 shows an example of the selection result of the adjustment parameter set in a specific example of information processing according to Embodiment 1. [Figure 14] Figure 14 shows an example of the selection result of the adjustment parameter set in a specific example of information processing according to Embodiment 1. [Figure 15] Figure 15 is a flowchart showing an example of the processing procedure according to Embodiment 1. [Figure 16] Figure 16 shows an example of a prediction result obtained by information processing according to Embodiment 1. [Figure 17] Figure 17 is an explanatory diagram showing an overview of the information processing method according to Embodiment 2. [Figure 18] Figure 18 shows an example of the configuration of an information processing device according to Embodiment 2. [Figure 19] Figure 19 is a diagram showing the overall system flow in a specific example of information processing according to Embodiment 2. [Figure 20]FIG. 20 is a diagram showing an example of a method for calculating an average RMS error in a specific example of information processing according to Embodiment 2. [Figure 21] FIG. 21 is a flowchart showing a processing procedure according to Embodiment 2. [Figure 22] FIG. 22 is a diagram showing an example of a prediction result by the information processing according to Embodiment 2. [Figure 23] FIG. 23 is a diagram showing an example of a hardware configuration. MODE FOR CARRYING OUT THE INVENTION

[0010] Hereinafter, embodiments of the information processing apparatus, information processing method, and information processing program according to the present application will be described in detail based on the drawings. Note that, for the information processing apparatus according to the present application, two embodiments will be described, but the information processing apparatus, information processing method, and information processing program according to the present application are not limited by these embodiments.

[0011] [Principle] [1. Introduction] Hereinafter, the functions of the information processing apparatus according to the present embodiment and the autoregressive model which is its output destination will be described. The autoregressive model calculates future predicted values at an arbitrary number of steps ahead by recursively shifting the future predicted value one step ahead for given time series data to the past time series. Further, the information processing apparatus according to the present embodiment searches for an adjustment parameter set, which will be described later, that enables appropriate prediction at each time for the given time series data, and outputs it to an autoregressive model or the like.

[0012] [2. Prediction value calculation method] Hereinafter, a method for calculating a predicted value by the autoregressive model according to the present embodiment and adjustment parameters for calculating regression coefficients used therefor will be described with reference to mathematical formulas and drawings. FIG. 1 is a diagram showing an outline of a method for calculating a predicted value and a regression coefficient in the information processing according to the embodiment. FIG. 2 is a diagram showing an outline of adjustment parameters in the information processing according to the embodiment.

[0013] (2-1. Predicted values ​​and regression coefficients) First, let's explain how to calculate the predicted value. The predicted value is calculated by adding a weighted sum (p) of any number of past data points from the current value using regression coefficients. Here, the method for calculating the predicted value using p past data points is expressed as shown in Figure 1(1). The regression coefficient is calculated using the regularized least squares method to find the value that best fits a weighted sum (q) of past data points. The method for calculating the regression coefficient is expressed as shown in Figure 1(2).

[0014] For example, the example in equation (2) in Figure 1 represents a normal equation for calculating the regression coefficients (a1, a2, a3) when "p=3, q=6". In other words, the predicted value is calculated using 3(p) past data points from the baseline data and the regression coefficients (a1, a2, a3), and the regression coefficients a1, a2, a3 are estimated using 6 past data points from the baseline data.

[0015] (2-2. Adjustment Parameters) Next, we will explain the adjustment parameters used to calculate the regression coefficients mentioned above, referring to their formulas. In addition to p and q mentioned earlier, there are four types of adjustment parameters: λ and n. A set of these four combinations constitutes an adjustment parameter set.

[0016] Here, the adjustment parameter λ is the regularization weight in the calculation process of the regression coefficients described above, and is a robust parameter for stabilizing the calculation. The least squares estimation by ridge regression, which incorporates this λ to ensure robustness, is expressed as shown in equation 2(3). The adjustment parameter n is an arbitrary value that stretches or compresses past data in the calculation of the predicted values, and is a parameter that adjusts the sampling period of the data used for prediction (see Figure 2).

[0017] For example, in the example in Figure 2, there are 13 time points that serve as the basis for calculating the predicted values ​​in the time series data. However, because the adjustment parameter "n=4", 4 time points are extracted at equal intervals from the 13 time points, and the prediction calculation is performed using these extracted time points.

[0018] [Embodiment 1] [1. Overview of Information Processing Methods] First, with reference to Figure 3, an overview of the information processing method performed by the information processing device according to Embodiment 1 will be described. Figure 3 is an explanatory diagram showing an overview of the information processing method according to Embodiment 1. In Figure 3, an example of information processing is described in which an appropriate set of adjustment parameters is selected at a reference time point when calculating predicted values ​​within the evaluation target for given time series data, and output to the outside.

[0019] In the example shown in Figure 3, the information processing device 10 is an information processing device that selects the optimal set of adjustment parameters in response to inputs of time-series data, labels for normal intervals and abnormal prediction intervals, and alarm thresholds, and outputs the selected result. This device is implemented by a computer or a cloud system.

[0020] The information processing device 10 receives time-series data, labels for normal and abnormal prediction intervals, and alarm thresholds as input. Based on the received information, it evaluates the prediction results for all candidate adjustment parameter sets for all time points within the evaluation target, selects the optimal adjustment parameter set, and outputs it externally.

[0021] Specifically, the information processing device 10 first receives time-series data, labels for normal and abnormal prediction intervals for the time-series data, and alarm thresholds, which are input by engineers or other personnel. For example, the information processing device 10 receives past time-series data of the device for which prediction values ​​are to be calculated as time-series data.

[0022] Furthermore, for example, the information processing device 10 receives labels assigned to normal intervals and abnormal prediction intervals in time series data. Here, a normal interval is an interval in the time series data where the time series data after any given time interval does not exceed the alarm threshold described later. In contrast, an abnormal prediction interval is an interval in the time series data where the time series data after any given time interval exceeds the alarm threshold.

[0023] Furthermore, for example, the information processing device 10 receives alarm thresholds set by engineers or the like. Here, the alarm thresholds are set in the process of determining the optimal adjustment parameters such that the prediction will not exceed the alarm threshold in the normal interval and will exceed the alarm threshold in the abnormal prediction interval. Note that the aforementioned labeling and alarm threshold setting are performed in advance by engineers or the like before the information processing device 10 performs any information processing.

[0024] Next, the information processing device 10 calculates the values ​​of pre-set indicators for all candidate adjustment parameter sets for all time points, in response to the acceptance of input time-series data, labels for normal intervals and abnormal prediction intervals, and alarm thresholds. For example, the information processing device 10 calculates the values ​​of four indicators, described later, for all candidate adjustment parameter sets for all time points, in accordance with the input information described above: RMS error, threshold judgment result, abnormal detection success rate, and normal interval prediction success rate.

[0025] Subsequently, the information processing device 10 performs pre-configured processing based on the calculated indicator values ​​to select the optimal adjustment parameter set. For example, the information processing device 10 selects the optimal adjustment parameter set by sequentially performing processes such as selecting the parameter set with the highest anomaly detection success rate based on the four calculated indicator values, and parameter effectiveness determination processing, which will be described later. Finally, the information processing device 10 outputs the selected adjustment parameter set to an external device, the autoregressive model 20.

[0026] In this way, the information processing device 10 selects the optimal set of adjustment parameters in response to the input of time-series data, labels for normal and abnormal prediction intervals, and alarm thresholds, and outputs the selected result externally. As a result, the autoregressive model 20 can obtain the optimal set of adjustment parameters for calculating the predicted values.

[0027] [2. Configuration of the information processing device 10] Next, with reference to Figure 4, the configuration of the information processing device 10 according to Embodiment 1 will be described. Figure 4 is a diagram showing an example of the configuration of the information processing device 10 according to Embodiment 1. As shown in Figure 4, the information processing device 10 according to Embodiment 1 has a communication unit 11, a control unit 12, and a storage unit 13. The information processing device 10 and the autoregressive model 20 are connected to each other via wired or wireless means so that they can communicate with each other.

[0028] The communication unit 11 is implemented, for example, by a NIC (Network Interface Card). The communication unit 11 is connected to the autoregressive model 20 by wire or wireless connection and transmits and receives information with the autoregressive model 20. In addition, the communication unit 11 also receives, for example, time-series data, label information, and alarm thresholds from external sources.

[0029] The memory unit 13 is implemented by a storage device such as RAM (Random Access Memory) or a hard disk. The memory unit 13 stores data and programs necessary for various processes performed by the control unit 12, and is particularly closely related to this embodiment 1, and includes a time-series data storage unit 13a, a label information storage unit 13b, an alarm threshold storage unit 13c, and an index value storage unit 13d.

[0030] The time-series data storage unit 13a stores time-series data obtained by monitoring the target device over time. For example, the time-series data storage unit 13a stores data as time-series data, which includes measured values ​​such as the temperature of the target device obtained by monitoring the target device over time, along with the time of measurement.

[0031] The label information storage unit 13b stores information about the labels attached to time series data that fall under normal intervals and abnormal prediction intervals. For example, in the time series data of a target device, measurements taken when the device is operating normally are labeled as normal intervals, and measurements taken immediately before the device malfunctions are labeled as abnormal prediction intervals. The label information storage unit 13b stores the information about the labels attached to each of these time series data.

[0032] The alarm threshold storage unit 13c stores alarm thresholds set to determine whether or not to issue an alarm for a given measurement. For example, the alarm threshold storage unit 13c stores alarm thresholds set for the purpose of determining whether or not to issue an alarm for a given measurement such as the temperature of a target device.

[0033] Furthermore, the aforementioned time-series data, label information, and alarm thresholds must be stored in advance before the control unit 12 selects the optimal adjustment parameter set. These three data sets are stored in their respective memory units via the communication unit 11. Moreover, this storage may be performed manually by an engineer or the like, or automatically by a device monitoring the target device.

[0034] The index value storage unit 13d stores each index value calculated by the index value calculation unit 12a, which will be described later. For example, the index value storage unit 13d stores the index values ​​calculated for all candidate adjustment parameter sets for all time points of the received time series data.

[0035] The control unit 12 is implemented by a CPU (Central Processing Unit) or MPU (Micro Processing Unit), etc., which executes various programs stored in the memory device inside the information processing device 10 using RAM as the working area. Alternatively, the control unit 12 can be implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). The control unit 12 includes an index value calculation unit 12a and a parameter determination unit 12b.

[0036] The index value calculation unit 12a calculates the value of a pre-set index for each of the adjustment parameter sets used to calculate regression coefficients used to obtain predicted values ​​from past data. The index value calculation unit 12a then stores the calculated index values ​​in the index value storage unit 13d.

[0037] For example, the index value calculation unit 12a calculates the index value for all candidate adjustment parameter sets for all time points of the time series data from the data stored in the time series data storage unit 13a, the label information storage unit 13b, and the alarm threshold storage unit 13c, and stores it in the index value storage unit 13d.

[0038] Furthermore, the index value calculation unit 12a may calculate the following as index values: the mean squared error between the predicted value corresponding to the time point that serves as the basis for calculating the predicted value in past data and the actual data from the past; the difference between the predicted value corresponding to the time point and a preset threshold; the anomaly detection success rate, which is the ratio of anomaly prediction intervals in past data that include a time point corresponding to a predicted value for which anomaly prediction was successful, for all anomaly prediction intervals in past data; and the normal interval prediction success rate, which is the ratio of the number of time points corresponding to a predicted value for which normal prediction was successful, to the total number of time points in the normal interval in past data.

[0039] Here, an overview of each of the aforementioned index values ​​will be explained with reference to Figures 5 to 7. Figures 5 to 7 are diagrams showing an overview of the index values ​​in the information processing according to Embodiment 1.

[0040] The index value calculation unit 12a calculates the mean squared error (RMS error) between the predicted value corresponding to a specific time point and past actual data. Since a smaller mean squared error indicates better prediction accuracy, the information processing device 10 can evaluate the prediction error for a specific time by calculating the mean squared error and select the optimal set of adjustment parameters (see Figure 5).

[0041] Furthermore, the index value calculation unit 12a calculates the difference between the predicted value for a specific time point and a preset threshold. This allows the index value calculation unit 12a to determine whether or not the predicted value exceeds the threshold (see Figure 5).

[0042] Furthermore, the index value calculation unit 12a calculates an anomaly detection success rate for all anomaly prediction intervals in past data, which is the percentage of anomaly prediction intervals that include a time point corresponding to a predicted value for which anomaly prediction was successful. This allows the information processing device 10 to select the optimal adjustment parameter set by evaluating whether the predicted value corresponding to the time point within the aforementioned anomaly prediction interval exceeds a threshold value (anomaly prediction).

[0043] In the example in Figure 6, there are two anomaly prediction intervals in the historical data, and each of the anomaly prediction intervals (1) and (2) has two time points. Both predicted values ​​corresponding to the time points in anomaly prediction interval (1) are prediction failures, while only one predicted value corresponding to the time point in anomaly prediction interval (2) is a successful prediction. Therefore, it can be said that anomaly prediction interval (2) is an anomaly prediction interval that includes the time point corresponding to the predicted value that was successfully predicted as an anomaly.

[0044] Therefore, since there is one interval in which anomaly prediction was successful, and all anomaly prediction intervals are 2, the index value calculation unit 12a in the example of Figure 6 calculates the anomaly detection success rate, which is the ratio of an anomaly prediction interval containing the time point corresponding to the predicted value in which anomaly prediction was successful, as 1 / 2 = 0.5.

[0045] Furthermore, the index value calculation unit 12a calculates the normal interval prediction success rate, which is the ratio of the number of time points corresponding to the predicted value that was successfully predicted to the total number of time points within the normal interval in past data. This allows the information processing device 10 to evaluate whether the predicted value corresponding to the aforementioned time point within the normal interval does not exceed a threshold (normal prediction), thereby enabling it to select the optimal adjustment parameter set.

[0046] In the example in Figure 7, there are a total of 6 time points within the normal interval in the historical data. Of these, 4 time points correspond to the predicted values ​​that were successfully predicted as normal. Therefore, the index value calculation unit 12a in the example in Figure 6 calculates the normal interval prediction success rate, which is the ratio of the number of time points corresponding to the predicted values ​​that were successfully predicted as normal to the total number of time points within the normal interval, as 4 / 6 = 0.666...

[0047] For example, the index value calculation unit 12a calculates the values ​​of the four aforementioned indices for all candidate adjustment parameter sets for all time points of the time series data, using the data stored in the time series data storage unit 13a, the label information storage unit 13b, and the alarm threshold storage unit 13c, and stores them in the index value storage unit 13d.

[0048] The parameter determination unit 12b selects an adjustment parameter set for each adjustment parameter set based on the calculated index value and the index values ​​of neighboring adjustment parameter sets.

[0049] The parameter determination unit 12b then selects a nearby adjustment parameter set based on the index value of the adjustment parameter set, which is obtained by changing the numerical value of one element in the adjustment parameter set to an adjacent numerical value among the candidates set as the numerical value of that element.

[0050] Here, the effectiveness determination index used by the parameter determination unit 12b when selecting a set of adjustment parameters based on the index values ​​of neighboring adjustment parameter sets, and the effectiveness determination process for calculating it, will be explained with reference to Figure 8. Figure 8 is a diagram showing an overview of the process for selecting a set of adjustment parameters based on neighboring adjustment parameter sets in the information processing according to Embodiment 1.

[0051] In the example shown in Figure 8, the information processing device 10 is configured with 10 candidate values ​​for element p of the adjustment parameter set, ranging from 3 to 30; 10 candidate values ​​for element λ, ranging from 0.05 to 0.5; and 7 candidate values ​​for element n, ranging from 1 to 30. Note that the adjustment parameter q is set to 3 times p, and is therefore omitted in the example shown in Figure 8.

[0052] For example, if we consider the adjustment parameter set to be (p,λ,n)={3,0.1,5}, then changing the value of p (3) to the adjacent value 6 among the candidates gives us (p,λ,n)={6,0.1,5}, which is one example of a neighboring adjustment parameter set. Since there are four cases where the values ​​other than p are changed to adjacent values, the parameter determination unit 12b calculates the five adjustment parameter sets shown in Figure 8 as neighboring adjustment parameter sets for the given adjustment parameter set.

[0053] Furthermore, the parameter determination unit 12b in Embodiment 1 performs the specific processing for calculating neighboring parameter sets based on the distance condition formula in Figure 8. In this case, the parameter determination unit 12b assigns an index from 1 to the number of elements to each element of the adjustment parameter set. The parameter determination unit 12b also calculates the index of each element (for example, p) as index(p1) and calculates the distance between each element as |index(p1)-index(p2)|.

[0054] Subsequently, the parameter determination unit 12b determines whether the five neighboring adjustment parameter sets meet the pre-set conditions based on their respective index values. In the example shown in Figure 8, three of the five neighboring adjustment parameter sets meet the conditions. Therefore, the parameter determination unit 12b calculates the effectiveness determination index, which is the ratio of the number of neighboring adjustment parameter sets that meet the conditions to the total number of neighboring adjustment parameter sets, as 3 / 5 = 0.6.

[0055] For example, the parameter determination unit 12b selects an adjustment parameter set based on the index values ​​of each adjustment parameter set stored in the index value storage unit 13d, and an effectiveness determination index calculated from the index values ​​of neighboring adjustment parameter sets, obtained by changing the numerical value of one element in each adjustment parameter set to an adjacent numerical value among the candidates set as the numerical value of that element.

[0056] [3. Specific Examples of Information Processing] Next, with reference to Figures 9 to 14, a specific example of information processing according to Embodiment 1 will be described. Figure 9 is a diagram showing the overall system flow in the specific example of information processing according to Embodiment 1. Figure 10 is a diagram showing an example of each prerequisite setting in the specific example of information processing according to Embodiment 1. Furthermore, Figures 11 to 14 are diagrams showing an example of the selection result of the adjustment parameter set in the specific example of information processing according to Embodiment 1. Below, the information processing flow of the information processing device 10 will be described, followed by a specific example of the adjustment parameter set selection process.

[0057] (3-1. Information Processing Flow) First, referring to Figure 9, the overall system flow in a specific example of the information processing device 10 will be explained. For example, the information processing device 10 receives time-series data, label information, and alarm thresholds from an external source, and each data is stored in its corresponding memory unit.

[0058] Subsequently, the index value calculation unit 12a calculates, for example, the RMS error, threshold determination, anomaly detection success rate, and normal interval prediction success rate for all pre-set candidate adjustment parameter sets from the data stored in each storage unit, and stores them in the index value storage unit 13d.

[0059] The parameter determination unit 12b first performs process 2: selects the adjustment parameter set with the highest anomaly detection success rate from each index value stored in the index value storage unit 13d. If there are multiple candidates selected, the parameter determination unit 12b then performs process 3: selects the adjustment parameter set with the highest normal interval prediction success rate from among those candidates.

[0060] If multiple candidates are selected by process 3, the parameter determination unit 12b then selects the adjustment parameter set with the highest effectiveness metric from among the candidates by process 4: effectiveness determination process. If there are also multiple selection results, the parameter determination unit 12b finally selects the optimal adjustment parameter set by process 5: selecting the adjustment parameter set with the smallest average value of RMS error in the normal interval from among the candidates. After that, the information processing device 10 outputs the adjustment parameter set determined by the parameter determination unit 12b to the outside.

[0061] Figure 9 shows an example of an information processing device according to Embodiment 1, and since processes 2 to 5 are independent of each other, it is possible to change the order of some of the processes.

[0062] (3-2. Prerequisites for Information Processing) Next, referring to Figure 10, the prerequisite settings for a specific example of information processing according to Embodiment 1 will be explained. In this example, the sampling interval for time-series data is set to 10 seconds, 1.5 days of the time-series data is labeled as the normal interval, and 80 minutes is labeled as the abnormal prediction interval. The alarm threshold is set to th.

[0063] Furthermore, the conditions for the candidate adjustment parameter sets are as follows: element p has 10 possibilities in increments of 3 from 3 to 30; element q is 3 times p (not searched); element λ has 10 possibilities in increments of 0.05 from 0 to 0.5; and element n has 7 possibilities in increments of 5 from 0 to 30. As a result, the index value calculation unit 12a calculates each index value for all 700 candidate adjustment parameter sets for all time points within the evaluation target.

[0064] (3-3. Selection process for the adjustment parameter set) Finally, referring to Figures 11 to 14, the selection results of the adjustment parameter set by each process in a specific example of information processing according to Embodiment 1 will be explained. For example, the parameter determination unit 12b selects the optimal adjustment parameter set by sequentially performing processes 2 to 5 based on each index value stored in the index value storage unit 13d (see Figure 9).

[0065] First, the parameter determination unit 12b performs process 2: selects the adjustment parameter set with the highest anomaly detection success rate, resulting in the result shown in Figure 11. Of the 700 adjustment parameter sets explored, 378 adjustment parameter sets satisfy the conditions of process 2 and are selected. In other words, 378 adjustment parameter sets are selected as the adjustment parameter sets that contain "1", which has the highest anomaly detection success rate. Since there are multiple selection results, the parameter determination unit 12b proceeds to process 3.

[0066] Next, the parameter determination unit 12b performs process 3: selects the adjustment parameter set with the highest success rate for predicting the normal interval, resulting in the result shown in Figure 12. Of the 378 adjustment parameter sets selected in process 2, 262 adjustment parameter sets satisfy the conditions of process 3 and are selected. In other words, similar to process 2 described above, 262 adjustment parameter sets are selected as the adjustment parameter sets that have "1" listed, which is the highest success rate for predicting the normal interval. Since there are multiple selection results, the parameter determination unit 12b proceeds to process 4.

[0067] Next, the parameter determination unit 12b performs process 4: effectiveness determination process to select the adjustment parameter set with the highest effectiveness determination index, resulting in the result shown in Figure 13. Of the 262 adjustment parameter sets selected in process 3, 67 adjustment parameter sets satisfy the conditions of process 4 and are selected. In other words, similar to processes 2 and 3 described above, 67 adjustment parameter sets are selected as the adjustment parameter sets that have the highest effectiveness determination index, "1". Since there are multiple selection results, the parameter determination unit 12b proceeds to process 5.

[0068] Finally, the parameter determination unit 12b performs process 5: selects the adjustment parameter set with the smallest average RMS error in the normal interval, resulting in the result shown in Figure 14. Of the 67 adjustment parameter sets selected in process 4, the adjustment parameter set (p,q,λ,n)={30,90,0.1,30} satisfies the conditions of process 5 and is selected.

[0069] Through the series of processes described above, the parameter determination unit 12b can select a parameter set that satisfies all the conditions of processes 2 to 5. In other words, the parameter determination unit 12b can appropriately perform abnormal and normal predictions, is not affected by outliers, and can select the optimal adjustment parameter set that has a small error between the predicted value and the actual data.

[0070] [4. Processing Procedure] Next, the processing of the information processing device 10 according to Embodiment 1 will be described with reference to Figure 15. Figure 15 is a flowchart of an example of the processing procedure according to Embodiment 1. Note that Figure 15 is an example of the information processing device according to Embodiment 1, and since S103, S105, S107, and S109 are independent of each other, it is possible to change the order of some of them during processing.

[0071] In the example shown in Figure 15, the information processing device 10 receives the input time-series data, labels, and alarm threshold (step S101). If the information processing device 10 has not received the time-series data, labels, and alarm threshold (step S101; No), it waits until it receives the time-series data, labels, and alarm threshold.

[0072] On the other hand, if the information processing device 10 receives time-series data, labels, and alarm thresholds (step S101; Yes), the index value calculation unit 12a calculates each index value for each of the adjustment parameter sets to be searched (step S102). Then, the parameter determination unit 12b selects, for example, the adjustment parameter set with the highest anomaly detection success rate (step S103). After that, the parameter determination unit 12b determines whether there are two or more selected adjustment parameter sets (step S104).

[0073] If the number of selected adjustment parameter sets is not two or more (Step S104; No), the information processing device 10 outputs the selected adjustment parameter sets (Step S110). On the other hand, if there are two or more selected adjustment parameter sets (Step S104; Yes), the parameter determination unit 12b selects the adjustment parameter set with the highest success rate for normal interval prediction from the selection results (Step S105). After that, the parameter determination unit 12b determines whether there are two or more selected adjustment parameter sets (Step S106).

[0074] If the number of selected adjustment parameter sets is not two or more (Step S106; No), the information processing device 10 outputs the selected adjustment parameter sets (Step S110). On the other hand, if there are two or more selected adjustment parameter sets (Step S106; Yes), the parameter determination unit 12b selects the adjustment parameter set with the highest effectiveness determination index from the selection results (Step S107). After that, the parameter determination unit 12b determines whether there are two or more selected adjustment parameter sets (Step S108).

[0075] If the number of selected adjustment parameter sets is not two or more (Step S108; No), the information processing device 10 outputs the selected adjustment parameter sets (Step S110). On the other hand, if there are two or more selected adjustment parameter sets (Step S108; Yes), the parameter determination unit 12b selects the adjustment parameter set with the smallest average value of the RMS error in the normal interval from the selection results (Step S109). After that, the information processing device 10 outputs the selected adjustment parameter sets (Step S110).

[0076] [5. Effects of Embodiment 1] As described above, the information processing device 10 according to this embodiment 1 calculates the value of each indicator for each candidate adjustment parameter set for all time points within the evaluation target for the given time series data. Then, the information processing device 10 selects an appropriate adjustment parameter set for prediction and outputs it externally based on the calculated indicator values ​​and the indicator values ​​of the adjustment parameter set (neighboring adjustment parameter set) obtained by changing the numerical value of one element in each element of the adjustment parameter set to an adjacent numerical value in the candidate set as the numerical value of the element.

[0077] As a result, when the information processing device 10 evaluates the adjustment parameter set, it can avoid determining an inappropriate set of adjustment parameters due to sudden abnormal data by adding the index values ​​of neighboring adjustment parameter sets to the evaluation target.

[0078] Furthermore, the index value calculation unit 12a of the information processing device 10 calculates the following as index values: the mean squared error between the predicted value corresponding to the time point that serves as the basis for calculating the predicted value in past data and the actual data from the past; the difference between the predicted value corresponding to the time point and a preset threshold; the anomaly detection success rate, which is the ratio of anomaly prediction intervals in past data that include a time point corresponding to a predicted value for which anomaly prediction was successful, for all anomaly prediction intervals in past data; and the normal interval prediction success rate, which is the ratio of the number of time points corresponding to a predicted value for which normal prediction was successful, to the total number of time points in the normal interval in past data.

[0079] As a result, the information processing device 10 can appropriately predict abnormalities and normal conditions when selecting a set of adjustment parameters, and can efficiently select the optimal set of adjustment parameters with a small error between the actual data and the predicted values, and output it externally.

[0080] Here, referring to Figure 16, we will explain the prediction results by the autoregressive model 20 using the set of adjustment parameters selected by the above process and the set of adjustment parameters that were not selected. Figure 16 is a diagram showing an example of the prediction results by the information processing according to Embodiment 1.

[0081] In the example shown in Figure 16, the autoregressive model 20 calculates predicted values ​​for the time series data using two sets of adjustment parameters: one selected by the information processing device 10 and another set of adjustment parameters not selected by the information processing device 10 for comparison.

[0082] The information processing device 10 performs the following steps in order for the time series data: calculate the value of each indicator (process 1), select the adjustment parameter set with the highest anomaly detection success rate (process 2), select the adjustment parameter set with the highest normal prediction success rate (process 3), select the adjustment parameter set with the highest effectiveness judgment indicator (process 4), and select the adjustment parameter set with the smallest average RMS error (process 5) (see Figure 9).

[0083] Through the above series of processes, the selected adjustment parameter set is (p,q,λ,n)={30,90,0.1,30}. In addition, in the example in Figure 16, the prediction result of the adjustment parameter set (p,q,λ,n)={30,90,0.5,30}, which was not selected by the information processing device 10, is used as a comparison target.

[0084] In the example in Figure 16, four graphs are shown. The top two graphs show the prediction results for the unselected adjustment parameter sets, while the bottom two graphs show the prediction results for the selected adjustment parameter sets. The two graphs on the left of Figure 16 show the prediction results for each normal interval, and the two graphs on the right show the prediction results for each abnormal prediction interval. In the example in Figure 16, the value of 110 on the vertical axis of each graph is the alarm threshold.

[0085] When using the adjustment parameter set (p,q,λ,n)={30,90,0.5,30} that was not selected by the information processing device 10, the prediction result in the normal interval (Figure 16, upper left) shows that the dashed line representing the predicted value does not exceed the alarm threshold, indicating that the prediction in the normal interval is appropriate. On the other hand, the prediction result in the abnormal prediction interval (Figure 16, upper right) shows that the dashed line representing the predicted value does not exceed the alarm threshold, indicating that abnormal prediction is not possible and therefore the prediction is not appropriate.

[0086] On the other hand, when using the adjustment parameter set (p,q,λ,n)={30,90,0.1,30} selected by the information processing device 10, the prediction result in the normal interval (Figure 16, lower left) shows that the dashed line representing the predicted value does not exceed the alarm threshold, indicating that the prediction in the normal interval is appropriate. Furthermore, the prediction result in the abnormal prediction interval (Figure 16, lower right) shows that the dashed line representing the predicted value exceeds the alarm threshold, indicating that the abnormal prediction was successful and appropriate.

[0087] Therefore, when the autoregressive model 20 used the adjustment parameter set not selected by the information processing device 10, it was unable to make appropriate predictions in the abnormal prediction interval, whereas when the selected adjustment parameter set was used, it was able to make appropriate predictions in both the normal interval and the abnormal prediction interval. Thus, the adjustment parameter set (p,q,λ,n)={30,90,0.1,30} selected by the information processing device 10 can be said to be an adjustment parameter set that enables the autoregressive model 20 to make appropriate predictions compared to the adjustment parameter set under comparison. Therefore, it can be said that by performing processes 1 to 5, which are examples of information processing according to Embodiment 1, the information processing device 10 can select an appropriate adjustment parameter set.

[0088] [Embodiment 2] [1. Overview of Information Processing Methods] First, with reference to Figure 17, an overview of the information processing method performed by the information processing device according to Embodiment 2 will be described. Figure 17 is an explanatory diagram showing an overview of the information processing method according to Embodiment 2. In Figure 17, an example of information processing is described in which time-series data is obtained from the target device 30, an optimal set of adjustment parameters is selected based on the index values ​​calculated from the obtained time-series data, and then the autoregressive model 20 is updated with the selected set of adjustment parameters.

[0089] In the example shown in Figure 17, the information processing device 10 is an information processing device that selects the optimal set of adjustment parameters in response to the input of time-series data from the target device 30 and updates the autoregressive model 20 with the selected set of adjustment parameters, and is implemented by a computer or cloud system.

[0090] The information processing device 10 calculates index values ​​for all candidate adjustment parameter sets with respect to the regression coefficient calculation interval reference position (reference time), which will be described later, using time series data obtained from the target device 30. Based on these index values, it selects the optimal adjustment parameter set and updates the autoregressive model 20 with the selected adjustment parameter set.

[0091] Specifically, first, the information processing device 10 obtains time-series data to be predicted from the target device 30. For example, the information processing device 10 obtains time-series data by receiving observation data transmitted in real time from the target device 30.

[0092] Next, the information processing device 10 calculates indicator values ​​from the acquired time-series data and selects the optimal set of adjustment parameters. For example, in a prediction evaluation interval with the current time as the evaluation reference position, the information processing device 10 calculates prediction data for all candidate sets of adjustment parameters, then calculates the average RMS error described later for each set of adjustment parameters, and selects the set of adjustment parameters with the smallest average RMS error.

[0093] Finally, the information processing device 10 updates the autoregressive model 20 with the selected set of adjustment parameters. For example, the information processing device 10 inputs the selected set of adjustment parameters into the autoregressive model and updates the autoregressive model 20 so that predicted data is calculated using the regression coefficients updated by the set of adjustment parameters.

[0094] In this way, the information processing device 10 calculates index values ​​in response to the input of time-series data and selects the optimal set of adjustment parameters. Subsequently, the information processing device 10 updates the autoregressive model 20 with the selected set of adjustment parameters. As a result, the information processing device 10 enables the autoregressive model 20 to make highly accurate predictions using the optimal set of adjustment parameters when calculating predicted values.

[0095] [2. Configuration of the information processing device 10] Next, with reference to Figure 18, the configuration of the information processing device 10 according to Embodiment 2 will be described. Figure 18 is a diagram showing an example of the configuration of the information processing device 10 according to Embodiment 2. As shown in Figure 18, the information processing device 10 according to Embodiment 2 has a communication unit 11, a control unit 12, and a storage unit 13. Furthermore, the information processing device 10, the autoregressive model 20, and the target device 30 are connected to each other so as to be able to communicate with each other by wired or wireless means.

[0096] The communication unit 11 is implemented, for example, by a NIC (Network Interface Card). The communication unit 11 is connected to the autoregressive model 20 and the target device 30 by wire or wireless connection, and transmits and receives information between the autoregressive model 20 and the target device 30. For example, the input of time-series data from the target device 30 and the updating of the autoregressive model 20 with a selected set of adjustment parameters are performed via the communication unit 11.

[0097] The storage unit 13 is implemented by a storage device such as RAM (Random Access Memory) or a hard disk. The storage unit 13 stores data and programs necessary for various processes performed by the control unit 12, but is particularly closely related to the present invention and includes a time-series data storage unit 13a, an adjustment parameter set storage unit 13b, a regression coefficient calculation interval reference position storage unit 13c, and an index value storage unit 13d.

[0098] The time-series data storage unit 13a stores time-series data obtained by monitoring the target device 30 over time. For example, the time-series data storage unit 13a stores data recorded via the communication unit 11, including measured values ​​such as the temperature of the target device 30 obtained by monitoring the target device 30 over time, along with the time of measurement.

[0099] The adjustment parameter set storage unit 13b stores candidate adjustment parameter sets to be searched. For example, the adjustment parameter set storage unit 13b stores combinations of adjustment parameter sets that are candidates in the adjustment parameter set selection process described later.

[0100] The regression coefficient calculation interval reference position storage unit 13c stores information that allows for the identification of the regression coefficient calculation interval reference position, which serves as the reference for the interval used to calculate the regression coefficients when calculating the prediction data. For example, the regression coefficient calculation interval reference position storage unit 13c stores information that sets the regression coefficient calculation interval reference position to a time point a certain time before the current time of the acquired time series data.

[0101] Here, the information processing device 10 defines the interval from the regression coefficient calculation interval reference position toward the current time as the prediction evaluation interval, and the interval from the regression coefficient calculation interval reference position toward the past as the regression coefficient calculation interval (see Figure 17). In other words, the information processing device 10 evaluates the predicted data of the regression coefficient calculation interval reference position in the prediction evaluation interval. The interval width of the prediction evaluation interval is set in advance, and the interval width of the regression coefficient calculation interval is determined by each element of the adjustment parameter set. Furthermore, the aforementioned candidate adjustment parameter sets and information regarding the regression coefficient calculation interval reference position must be stored in advance before the control unit 12 selects the optimal adjustment parameter set.

[0102] The index value storage unit 13d stores the index values ​​calculated by the index value calculation unit 12a, which will be described later. For example, the index value storage unit 13d stores the index values ​​calculated for all candidate adjustment parameter sets for the reference position of the regression coefficient calculation interval of the acquired time series data.

[0103] The control unit 12 is implemented by a CPU (Central Processing Unit) or MPU (Micro Processing Unit), etc., which executes various programs stored in the memory device inside the information processing device 10 using RAM as the working area. Alternatively, the control unit 12 may be implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). The control unit 12 includes an index value calculation unit 12a and a parameter determination unit 12b, and may also include a prediction model update unit 12c as needed.

[0104] The index value calculation unit 12a calculates a pre-set index value for each of the adjustment parameter sets used to calculate regression coefficients used to obtain predicted values ​​from past data. The index value calculation unit 12a then stores the calculated index values ​​in the index value storage unit 13d.

[0105] For example, the index value calculation unit 12a calculates index values ​​for all candidate adjustment parameter sets relative to the regression coefficient calculation interval reference position of the time series data from each data stored in the time series data storage unit 13a, the adjustment parameter set storage unit 13b, and the regression coefficient calculation interval reference position storage unit 13c, and stores them in the index value storage unit 13d.

[0106] Furthermore, the index value calculation unit 12a may calculate the mean squared error between the predicted value corresponding to the reference position of the regression coefficient calculation interval and past actual data as an index value. The index value calculation unit 12a then stores the calculated RMS error in the index value storage unit 13d.

[0107] For example, the index value calculation unit 12a calculates the RMS error between the predicted value of the prediction evaluation interval and the actual data for all candidate adjustment parameter sets with respect to the reference position of the regression coefficient calculation interval, and stores it in the index value storage unit 13d.

[0108] The parameter determination unit 12b selects an adjustment parameter set for each adjustment parameter set based on the calculated index value and the index values ​​of neighboring adjustment parameter sets. Subsequently, the parameter determination unit 12b notifies the prediction model update unit 12c of the selected adjustment parameter set.

[0109] The parameter determination unit 12b then sets the index value of a nearby adjustment parameter set to a predetermined past regression coefficient calculation interval reference position, calculates the index value for each adjustment parameter set using the data of a predetermined interval based on the said regression coefficient calculation interval reference position, changes the regression coefficient calculation interval reference position, and repeats the process of calculating the index value for each adjustment parameter set using the data of a predetermined interval based on the changed regression coefficient calculation interval reference position a predetermined number of times, and selects an adjustment parameter set based on the calculated index value.

[0110] Here, for example, the regression coefficient calculation interval reference position storage unit 13c stores information such as setting the regression coefficient calculation interval reference position to a time arbitrary time before the current time of the acquired time series data, as well as information such as setting the time when the regression coefficient calculation interval reference position is changed as the regression coefficient calculation interval reference position. As a result, even when the regression coefficient calculation interval reference position is changed, the index value calculation unit 12a calculates the index value based on the changed regression coefficient calculation interval reference position and stores it in the index value storage unit 13d.

[0111] For example, the parameter determination unit 12b selects the adjustment parameter set with the smallest average RMS error, calculated based on the RMS error at the regression coefficient calculation interval reference position and the RMS errors for each adjustment parameter set at the regression coefficient calculation interval reference position (neighboring adjustment parameter sets) when the regression coefficient calculation interval reference position is changed a predetermined number of times to the past, for each candidate adjustment parameter set stored in the index value storage unit 13d. Subsequently, the parameter determination unit 12b notifies the prediction model update unit 12c of the selected adjustment parameter set.

[0112] The prediction model update unit 12c updates the regression coefficients of the future prediction model according to the set of adjustment parameters selected by the parameter determination unit 12b. For example, the prediction model update unit 12c updates the regression coefficients by updating the parameters of the autoregressive model 20 via the communication unit 11 in response to notification of the set of adjustment parameters from the parameter determination unit 12b.

[0113] [3. Specific Examples of Information Processing] Next, with reference to Figures 19 and 20, a specific example of information processing according to Embodiment 2 will be described. Figure 19 is a diagram showing the overall system flow in a specific example of information processing according to Embodiment 2. Figure 20 is a diagram showing an example of a method for calculating the average RMS error in a specific example of information processing according to Embodiment 2. Below, the information processing flow of the information processing device 10 will be described, followed by a specific example of the process for calculating the average RMS error.

[0114] (3-1. Information Processing Flow) First, referring to Figure 19, the overall system flow in a specific example of the information processing device 10 will be explained. For example, the information processing device 10 stores in its storage unit 13 a set of candidate adjustment parameters, information regarding the reference position of the regression coefficient calculation interval, information regarding the interval width of the prediction evaluation interval, and information that identifies the current time, which will be the evaluation reference position, from the acquired time-series data.

[0115] When the information processing device 10 obtains time-series data from the target device 30, it identifies the current time in the time-series data as the evaluation reference position from the information stored in the storage unit 13. Subsequently, the index value calculation unit 12a calculates the average RMS error, as described later, for each of the candidate adjustment parameter sets.

[0116] The parameter determination unit 12b then selects the adjustment parameter set with the smallest average RMS error from the average RMS errors of each adjustment parameter set stored in the index value storage unit 13d, for example, and notifies the prediction model update unit 12c. Subsequently, the prediction model update unit 12c updates the parameters of the autoregressive model 20 according to the notified adjustment parameter set.

[0117] (3-2. Calculation process for the mean RMS error) Next, with reference to Figure 20, a method for calculating the average RMS error in a specific example of information processing according to Embodiment 2 will be described. First, the index value calculation unit 12a calculates the RMS error for each of the M types of adjustment parameter sets stored in the adjustment parameter set storage unit 13b using time-series data of a predetermined interval (regression coefficient calculation interval in Figure 20) based on the regression coefficient calculation interval reference position.

[0118] Next, the index value calculation unit 12a shifts the regression coefficient calculation interval to each past time point an arbitrary S times and calculates the RMS error at the reference position of each regression coefficient calculation interval for each of the M types of adjustment parameter sets. In other words, the index value calculation unit 12a calculates M × S RMS errors using M types of adjustment parameter sets × S regression coefficient calculation intervals.

[0119] Subsequently, the index value calculation unit 12a calculates the average value of the RMS error for each common adjustment parameter set. Since the index value calculation unit 12a has calculated S RMS errors for each of the M types of adjustment parameter sets through the process described above, it can calculate the averaged RMS error for each of the M types of adjustment parameter sets by calculating the average value of these S RMS errors. In other words, the index value calculation unit 12a calculates M average RMS errors. The index value calculation unit 12a then stores the calculated average RMS errors in the index value storage unit 13d.

[0120] Finally, the parameter determination unit 12b selects the adjustment parameter set with the smallest average RMS error from the average RMS errors stored in the index value storage unit 13d through the series of processes performed by the index value calculation unit 12a described above. This allows the parameter determination unit 12b to select the adjustment parameter set that minimizes the error between the actual data and the predicted data in the prediction evaluation interval.

[0121] [4. Processing Procedure] Next, the processing of the information processing device 10 according to Embodiment 2 will be described with reference to Figure 21. Figure 21 is a flowchart showing an example of the processing procedure according to Embodiment 2. In the example shown in Figure 21, the information processing device 10 receives the input time-series data (step S101). If the information processing device 10 has not received time-series data (step S101; No), it waits until it receives time-series data.

[0122] On the other hand, if the information processing device 10 receives time-series data (step S101; Yes), the information processing device 10 determines the current time of the time-series data as the evaluation reference position (step S102). Then, the index value calculation unit 12a calculates, for example, the average RMS error, which is the index value of the adjustment parameter set (step S103). After that, the parameter determination unit 12b selects the adjustment parameter set with the smallest average RMS error (step S104).

[0123] Subsequently, the parameter determination unit 12b selects the set of adjustment parameters that minimizes the mean RMS error (step S104). Finally, the information processing device 10 updates the autoregressive model 20 using the selected set of adjustment parameters (step S105).

[0124] [5. Effects of Embodiment 2] As described above, the information processing device 10 according to this second embodiment calculates index values ​​for all candidate adjustment parameter sets relative to the regression coefficient calculation interval reference position from time-series data obtained from the target device 30. The information processing device 10 then changes the regression coefficient calculation interval reference position a predetermined number of times and selects an adjustment parameter set based on the index values ​​calculated for each adjustment parameter set (neighboring adjustment parameter set) relative to the changed regression coefficient calculation interval reference position.

[0125] As a result, when the information processing device 10 evaluates the adjustment parameter set, it can avoid determining an inappropriate set of adjustment parameters due to sudden abnormal data by adding the index values ​​of neighboring adjustment parameter sets to the evaluation target.

[0126] Furthermore, the index value calculation unit 12a of the information processing device 10 calculates the RMS error between the predicted value corresponding to the reference position of the regression coefficient calculation interval for each adjustment parameter set and the actual past data, as index values. Then, the prediction model update unit 12c of the information processing device 10 updates the regression coefficients by updating the parameters of the autoregressive model 20 according to the adjustment parameter set selected by the parameter determination unit 12b.

[0127] As a result, the information processing device 10 can select the optimal set of tuning parameters that minimizes the error between the actual data and the predicted values, and can update the parameters of the autoregressive model 20 with the optimal set of tuning parameters. Consequently, the autoregressive model 20 can make appropriate predictions using the regression coefficients calculated with the optimal set of tuning parameters.

[0128] Here, referring to Figure 22, we will explain the prediction results by the autoregressive model 20 using the set of adjustment parameters selected by the aforementioned process and the default parameters. Figure 22 is a diagram showing an example of the prediction results obtained by the information processing according to Embodiment 2. In the example in Figure 22, the accuracy of each predicted value is shown by the RMS error between each predicted value and the actual data.

[0129] In the example in Figure 22, the default parameters used are (p=20, q=60, λ=0.1, Δymax=5.0, emax=2.5). The autoregressive model 20 calculates the predicted values ​​for the entire interval of the time series data to be predicted, both when the parameters are sequentially estimated using the adjustment parameter set selected by the information processing device 10, and when the default parameters are used for comparison.

[0130] Then, in the information processing device 10, the index value calculation unit 12a first calculates the mean RMS error from the acquired time series data, and based on this, the parameter determination unit 12b selects an adjustment parameter set. Subsequently, the prediction model update unit 12c of the information processing device 10 updates the parameters of the autoregressive model 20 using the selected adjustment parameter set.

[0131] In the example shown in Figure 22, the information processing device 10 obtains time-series data from the target device 30 at regular intervals and performs the above processing each time. As a result, the parameters of the autoregressive model 20 are updated sequentially in response to the time-series data obtained by the information processing device 10.

[0132] In the example in Figure 22, the prediction result using the comparison default parameters (solid line in Figure 16) shows that the RMS error, which represents the error with the actual data, is large in the 3000min to 4000min range where the actual data measurements are erratic. In other words, when using the comparison default parameters, the autoregressive model 20 calculates predicted values ​​with large errors with the actual data in a certain interval, and therefore cannot be said to be making appropriate predictions.

[0133] On the other hand, the prediction results (dotted line in Figure 16) when the parameters of the autoregressive model 20 are successively updated by the adjustment parameter set selected by the information processing device 10 show that the RMS error is relatively small even in the period from 3000 min to 4000 min, where the measured values ​​of the actual data are erratic. In other words, when the parameters of the autoregressive model 20 are successively updated by the information processing device 10, the autoregressive model 20 calculates predicted values ​​with small errors from the actual data over the entire interval of the target time series data, and can be said to be making appropriate predictions.

[0134] Therefore, when the parameters of the autoregressive model 20 are sequentially estimated using the set of adjustment parameters selected by the information processing device 10, it can be said that the autoregressive model 20 is able to make more appropriate predictions than when the default parameters of the comparison target were used.

[0135] Furthermore, when comparing the two cases described above with time-series data consisting of sine waves of different frequencies, sine waves with gradually increasing noise, and sine waves with changing frequencies, it was confirmed that for all time-series data, the autoregressive model 20 made more appropriate predictions when the parameters of the autoregressive model 20 were sequentially estimated by the information processing device 10. Therefore, it can be said that by performing the series of processing described above, which is an example of information processing according to Embodiment 2, the information processing device 10 can select an appropriate set of adjustment parameters.

[0136] [Hardware configuration] The information processing device 10 according to Embodiments 1 and 2 described above is implemented by a computer 1000 having a configuration such as that shown in Figure 23. Figure 23 is a hardware configuration diagram showing an example of a computer that implements the functions of the information processing device 10. The computer 1000 has a configuration in which a CPU 1100, RAM 1200, ROM 1300, auxiliary storage device 1400, communication interface 1500, and input / output interface 1600 are connected by a bus 1800.

[0137] The CPU 1100 operates based on programs stored in the ROM 1300 or auxiliary storage device 1400, and controls various parts. The ROM 1300 stores boot programs executed by the CPU 1100 when the computer 1000 starts up, as well as programs that depend on the computer 1000's hardware.

[0138] The auxiliary storage device 1400 stores programs executed by the CPU 1100, and data used by such programs. The communication interface 1500 receives data from other devices via a predetermined communication network and sends it to the CPU 1100, and transmits data generated by the CPU 1100 to other devices via the predetermined communication network.

[0139] The CPU 1100 controls output devices such as displays and printers, and input / output devices 1700 such as keyboards and mice, via the input / output interface 1600. The CPU 1100 acquires data from the input / output devices 1700 via the input / output interface 1600. The CPU 1100 also outputs the generated data to the input / output devices 1700 via the input / output interface 1600.

[0140] For example, when the computer 1000 functions as the information processing device 10 according to Embodiment 1 and Embodiment 2, the CPU 1100 of the computer 1000 realizes the functions of the control unit 12 by executing a program loaded on the RAM 1200.

[0141] [others] Of the processes described in Embodiments 1 and 2 above, 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 by known methods. In addition, the processing procedures, specific names, and information including various data and parameters shown in the above document and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the information shown.

[0142] Furthermore, the components of each illustrated device are functionally conceptual and do not necessarily need to be physically configured as shown. In other words, the specific forms of distribution and integration of each device are not limited to those illustrated, and all or part of them can be functionally or physically distributed and integrated in any unit according to various loads, usage conditions, etc.

[0143] The aforementioned components include those that can be easily conceived by those skilled in the art, those that are substantially identical, and those that fall within the so-called equivalent range. Furthermore, Embodiments 1 and 2 described above can be combined as appropriate, as long as the processing content is not contradictory.

[0144] Furthermore, the terms "section," "module," and "unit" mentioned above can be replaced with "means" or "circuit," etc. For example, a control unit can be replaced with a control means or a control circuit.

[0145] Although some embodiments of the present invention have been described in detail above with reference to the drawings, these are illustrative examples, and the present invention can be implemented in various modified and improved forms based on the knowledge of those skilled in the art, starting with the embodiments described in the disclosure section of the invention. [Explanation of Symbols]

[0146] [Embodiment 1] 10 Information Processing Devices 11 Communications Department 12 Control Unit 12a Index Value Calculation Unit 12b Parameter determination unit 13 Storage section 13a Time-series data storage unit 13b Label information storage unit 13c Alarm threshold memory unit 13d Index value storage unit 20. Autoregressive Models

[0147] [Embodiment 2] 10 Information Processing Devices 11 Communications Department 12 Control Unit 12a Index Value Calculation Unit 12b Parameter determination unit 12c Prediction Model Update Unit 13 Storage section 13a Time-series data storage unit 13b Adjustment parameter set storage unit 13c Regression coefficient calculation interval reference position storage unit 13d Index value storage unit 20. Autoregressive Models 30 Target devices

Claims

1. An index value calculation unit calculates multiple pre-set index values ​​for each candidate in the set of adjustment parameters used to calculate regression coefficients used to obtain predicted values ​​from past data, A parameter determination unit selects an adjustment parameter set using the comparison results obtained by comparing the plurality of index values ​​calculated for each candidate of the adjustment parameter set with the index values ​​of neighboring adjustment parameter sets in a predetermined order. An information processing device characterized by comprising:

2. The parameter determination unit selects the adjustment parameter set as the neighboring adjustment parameter set based on the index value of the adjustment parameter set obtained by changing the numerical value of one element in the adjustment parameter set to an adjacent numerical value among the candidates set as the numerical value of that element. The information processing apparatus according to feature 1.

3. The index value calculation unit determines the plurality of index values ​​as follows: The mean squared error between the predicted value corresponding to the reference time point used when calculating the predicted value in past data and the actual past data, The difference between the predicted value corresponding to the aforementioned time point and a preset threshold, For all anomaly prediction intervals in past data, the anomaly detection success rate is the ratio of anomaly prediction intervals that include the time point corresponding to the predicted value for which anomaly prediction was successful, The success rate of the normal interval prediction is calculated as the ratio of the number of time points corresponding to the predicted value that was successfully predicted to the total number of time points within the normal interval in past data. The information processing apparatus according to feature 2.

4. The parameter determination unit sets an index value for a nearby adjustment parameter set to a predetermined past reference time, calculates the index value for each adjustment parameter set using data from a predetermined interval based on that reference time, then changes the reference time and calculates the index value for each adjustment parameter set using data from a predetermined interval based on the changed reference time, repeats this process a predetermined number of times, and selects the adjustment parameter set based on the calculated index values. The information processing apparatus according to feature 1.

5. The index value calculation unit calculates the mean squared error between the predicted value calculated using data for a predetermined interval based on the reference time and the actual past data as the index value. The system further includes a prediction model update unit that updates the regression coefficients of the future prediction model according to the adjustment parameter set selected by the parameter determination unit. The information processing apparatus according to feature 4.

6. An information processing process performed by an information processing device, An index value calculation step for each candidate set of adjustment parameters, which calculates multiple pre-set index values ​​for each regression coefficient used to obtain predicted values ​​from past data, A parameter determination step in which a parameter is selected using the comparison results obtained by comparing the plurality of index values ​​calculated for each candidate of the adjustment parameter set with the index values ​​of neighboring adjustment parameter sets in a predetermined order. An information processing method characterized by including

7. A procedure for calculating index values ​​that calculates multiple pre-set index values ​​for each candidate set of adjustment parameters used to determine regression coefficients used to obtain predicted values ​​from historical data, A parameter determination procedure for selecting an adjustment parameter set, which involves comparing the multiple index values ​​calculated for each candidate of the adjustment parameter set with the index values ​​of neighboring adjustment parameter sets in a predetermined order, and using the comparison results. An information processing program that causes a computer to execute something.

Images