Prediction device, system, method and program

Abstract

A prediction device is provided, comprising: a time series data acquisition unit that acquires time series values ​​obtained from an object; a prediction data generation unit that generates a prediction value following the time series values ​​based on the time series values ​​and a trained model; a regression line calculation unit that calculates a regression line based on the time series values ​​and the prediction value; and a future prediction unit that makes a future prediction regarding the object based on the regression line.

Description

[Technical field]

[0001] The present invention relates to an apparatus for predicting the future of an object based on a signal obtained from the object, such as an apparatus for predicting the lifespan of an object. [Background technology]

[0002] 2. Description of the Related Art There have been known techniques for predicting the future of an object, for example, predicting the lifespan of the object, based on a signal obtained from the object.

[0003] For example, a technique is known in which a regression line (or an approximation line) is calculated for data obtained from an object, and a lifespan is predicted based on the regression line (e.g., Patent Document 1). Another technique is known in which the state of an object is input into a trained model to directly predict a lifespan (e.g., Patent Document 2). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2016-208773 A [Patent Document 2] JP 2018-97723 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, while the former method, which involves linear regression from past data, allows predictions to be made using a clear calculation process and is therefore explanatory, the accuracy of the predictions is not sufficient.

[0006] In addition, in the latter method, which directly predicts lifespan using machine learning technology, it is expected that the prediction accuracy will be improved, but the prediction process becomes a so-called black box, and the prediction results cannot be explained.

[0007] The present invention has been made against the above-mentioned technical background, and an object of the present invention is to provide a highly accurate yet explainable prediction technology for predicting the future of an object based on a signal obtained from the object. [Means for solving the problem]

[0008] The above-mentioned technical problems can be solved by a prediction device, a prediction system, a prediction method, a prediction program, etc. having the following configurations.

[0009] That is, the prediction device of the present invention includes a time series data acquisition unit that acquires time series values ​​obtained from an object, a prediction data generation unit that generates a predicted value following the time series value based on the time series value and a learned model, a regression line calculation unit that calculates a regression line based on the time series value and the predicted value, and a future prediction unit that makes a future prediction regarding the object based on the regression line.

[0010] According to such a configuration, future prediction can be performed based on the predicted value obtained using the machine learning technology and the time series value up to that point. This allows future prediction to be performed with high accuracy. In addition, since a regression line is used, it is possible to increase explainability compared to the case where a predicted value is obtained directly using the machine learning technology. In other words, it is possible to provide a prediction technology that is highly accurate and yet explainable in a technology for predicting the future of an object based on a signal obtained from the object.

[0011] The trained model may include a plurality of trained models that predict different time steps, and the prediction data generation unit may generate a plurality of the prediction values ​​based on the time series values ​​and each of the trained models.

[0012] According to this configuration, since a plurality of trained models for predicting different time steps are used, the prediction accuracy of each time step can be improved, thereby enabling accurate future prediction.

[0013] The future prediction may be a prediction regarding the lifespan of the object.

[0014] With this configuration, the lifespan of the object can be predicted with high accuracy.

[0015] The regression line may be a straight line.

[0016] With this configuration, it is possible to appropriately respond to linear changes in the object.

[0017] The prediction data generation unit may further include a pre-processing unit that extracts features of the time series values, and the prediction data generation unit may generate a prediction value following the time series values ​​based on the features extracted in the pre-processing unit and the trained model.

[0018] According to this configuration, the characteristics of the time series values ​​can be further reflected when generating a predicted value.

[0019] The feature amount may include a moving average of the time series values ​​and / or a difference value between time steps of the time series values.

[0020] With this configuration, the characteristics of the time series values ​​can be further expressed.

[0021] The apparatus may further include a confidence interval calculation unit that calculates a confidence interval based on the regression line.

[0022] With this configuration, it is possible to provide a margin of error for prediction based on the confidence interval.

[0023] The data analysis apparatus may further include a display control unit that displays the time series values, the predicted values, and the regression line.

[0024] According to this configuration, the user can visually confirm the time series values, the forecast values, the regression line, and the relationships between them.

[0025] The method may further include a confidence interval display control unit that displays the time series values, the predicted values, the regression line, and the confidence interval.

[0026] According to this configuration, the user can visually check the time series values, the forecast values, the regression line, the confidence interval, and the relationships between them. Therefore, from the boundaries of the confidence interval, it is possible to perform future predictions regarding the target object in the shortest and longest time periods.

[0027] Viewed from another aspect, the present invention is a prediction system comprising: a time series data acquisition unit that acquires time series values ​​obtained from an object; a prediction data generation unit that generates a predicted value following the time series value based on the time series value and a trained model; a regression line calculation unit that calculates a regression line based on the time series value and the predicted value; and a future prediction unit that makes a future prediction regarding the object based on the regression line.

[0028] From another aspect, the present invention is a prediction method, which causes a computer to execute a time series data acquisition step of acquiring time series values ​​obtained from an object, a prediction data generation step of generating a predicted value following the time series values ​​based on the time series values ​​and a trained model, a regression line calculation step of calculating a regression line based on the time series values ​​and the predicted value, and a future prediction step of making a future prediction regarding the object based on the regression line.

[0029] From another aspect, the present invention is a prediction program, comprising: a time series data acquisition step of acquiring a time series value obtained from an object; a prediction data generation step of generating a prediction value following the time series value based on the time series value and a trained model; a regression line calculation step of calculating a regression line based on the time series value and the prediction value; and a future prediction step of making a future prediction regarding the object based on the regression line. Execute the command. Effect of the Invention

[0030] According to the present invention, it is possible to provide a prediction technology that is highly accurate and explainable in a technology for predicting the future of an object based on a signal obtained from the object. [Brief description of the drawings]

[0031] [Figure 1] FIG. 1 is a schematic diagram of an information processing device. [Diagram 2] FIG. 2 is a functional block diagram. [Diagram 3] FIG. 3 is a flowchart showing a life expectancy prediction operation of an information processing device. [Figure 4] FIG. 4 is a conceptual diagram of the life prediction algorithm. [Diagram 5] FIG. 5 is a diagram showing an example of a screen displayed on the display device by the display control unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0032] Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

[0033] 1. First Embodiment A first embodiment will be described with reference to Figures 1 to 5. In the first embodiment, an example will be described in which the present invention is applied to a device that predicts the lifespan of an object based on a signal from a sensor attached so as to be able to observe the object.

[0034] (1.1 Configuration) 1 is a schematic configuration diagram of an information processing device 1 (or a lifespan prediction device) according to this embodiment. The information processing device 1 may be, for example, a microcomputer, a PC, a tablet terminal, or the like.

[0035] As is clear from the figure, the information processing device 1 includes a control unit 10, a storage unit 11, an input unit 12, a display control unit 15, a communication unit 16, and an I / O unit 17, which are connected via a bus. The information processing device 1 is connected to a display device 19 via the display control unit 15. The information processing device 1 is also connected to a sensor 18 via the I / O unit 17. These may be collectively referred to as the information processing device 1.

[0036] The control unit 10 is composed of a calculation device such as a CPU, and executes computer programs read from the storage unit 11 to realize various operations described below. The storage unit 11 is composed of various storage devices such as ROM / RAM, flash memory, and hard disk, and stores computer programs and data necessary for various operations described below.

[0037] The input unit 12 is configured to process input signals from a predetermined input device such as a button, a touch panel, a mouse, or a keyboard, and provide the processed signals to the control unit 10. The display control unit 15 is configured with a CPU or a GPU, and performs processing related to images / moving images to be displayed on the display device 19. The control unit 10 may be responsible for a part or all of the operations of the display control unit 15.

[0038] The communication unit 16 is configured as a communication unit for realizing wired or wireless communication with an external device. The I / O unit 17 functions as an interface for input and output with the external device, and is configured to be connectable to various devices. In this embodiment, a sensor 18 is connected to the I / O unit 17. This allows a sensor signal to be taken into the information processing device 1. The type of sensor is not particularly limited. Therefore, all kinds of sensors, such as a strain sensor, a temperature sensor, a light sensor, a sound sensor, and an image sensor, are included. Also, a device that includes a sensor element as a part thereof may be considered as the sensor 18.

[0039] 2 is a functional block diagram according to this embodiment. As is clear from the figure, a sensor 18 and a display device 19 are connected to the information processing device 1, and the sensor 18 is attached to the target device 2 to be observed. The sensor 18 can detect at least the state of the target device 2 and its changes over time.

[0040] The target device 2 is not particularly limited and can be any type of device. For example, it may be a mechanical device such as a machine tool, or a component part thereof. It may also be a device that performs not only a mechanical operation but also an electronic, electromagnetic, or chemical operation. Furthermore, it may not be a fixed device, but may be a mobile device such as a vehicle.

[0041] The data acquisition unit 101 of the information processing device 1 acquires a signal obtained by the sensor 18 as a time-series signal, and performs processing to provide the signal to the feature extraction unit 102 and the predicted data generation unit 103. The feature extraction unit 102 extracts features from the sensor signal obtained from the data acquisition unit 101, and provides the features to the predicted data generation unit 103. The predicted data generation unit 103 generates predicted data using a learned model based on current or past sensor values ​​and their features.

[0042] The data acquisition unit 101 and the prediction data generation unit 103 provide the current or past data and the prediction data to the regression line calculation unit 105. The regression line calculation unit 105 performs a calculation to calculate a regression line, and provides the calculation result to the future prediction unit 106, the confidence interval calculation unit 107, and the display control unit 151. The future prediction unit 106 performs a life prediction of the target device 2 based on the provided regression line and a predetermined criterion, and provides the result to the display control unit 151. The confidence interval calculation unit 107 performs a process to calculate a confidence interval based on the provided regression line, and provides the result to the display control unit 151.

[0043] The display control unit 151 performs processing to display, on the display device 19, in addition to a predetermined basic screen, current or past data obtained from the data acquisition unit 101, future prediction data obtained from the prediction data generation unit 103, the regression line provided from the regression line calculation unit 105, the future prediction results provided from the future prediction unit 106, and the confidence interval provided from the confidence interval calculation unit 107.

[0044] The configurations of the functional blocks shown in this embodiment, i.e., the data acquisition unit 101, the feature extraction unit 102, the predicted data generation unit 103, the regression line calculation unit 105, the future prediction unit 106, the confidence interval calculation unit 107, etc., are realized by the control unit 10 or the control unit 10 and the storage unit 11. Also, the configurations shown in this embodiment are merely examples, and it goes without saying that other hardware configurations may be used. For example, they may be configured as a system (or a server-client system) via a network.

[0045] (1.2 Operation) FIG. 3 is a flowchart showing the life prediction operation of the information processing device 1. As shown in FIG.

[0046] As is clear from the figure, when the process starts, the data acquisition unit 101 performs a process of acquiring values ​​(sensor values) detected by the sensor 18 in time series (S1). This makes it possible to acquire sensor values ​​from the past to the present. The data acquisition unit 101 provides the acquired data to the feature extraction unit 102, the predicted data generation unit 103, and the regression line calculation unit 105.

[0047] After acquiring the sensor values, the feature extraction unit 102 extracts (or generates) features from the time-series sensor values ​​as preprocessing (S2). In this embodiment, the feature includes a moving average value of a predetermined time step, a difference between adjacent time steps, etc. Note that other quantities may be used as long as they represent the characteristics of the time-series data. The feature extraction unit 102 provides the extracted feature to the prediction data generation unit 103.

[0048] According to this configuration, the characteristics of the time series values ​​can be further reflected when generating a predicted value.

[0049] After the feature extraction process, the prediction data generation unit 103 performs a process of generating a prediction value based on the data obtained from the data acquisition unit 101 and the data obtained from the feature extraction unit 102 (S3). More specifically, in this embodiment, the prediction data generation unit 103 has N trained models that receive the sensor value obtained from the data acquisition unit 101 and the feature obtained from the feature extraction unit 102 as input, and output prediction values ​​for one step ahead, two steps ahead, ..., and N steps ahead in time steps. The prediction data generation unit 103 receives the sensor value obtained from the data acquisition unit 101 and the feature obtained from the feature extraction unit 102 as input, and outputs prediction values ​​for one step ahead, two steps ahead, ..., and N steps ahead in time steps for these trained models. The prediction data generation unit 103 provides the output prediction value to the regression line calculation unit 105.

[0050] According to this configuration, since a plurality of trained models for predicting different time steps are used, the prediction accuracy of each time step can be improved, thereby enabling accurate future prediction.

[0051] After the process of generating the predicted values, the regression line calculation unit 105 performs a process of calculating a regression line from the current and past time-series sensor values ​​and the predicted values ​​obtained from the data acquisition unit 101 (S5). More specifically, in this embodiment, a simple linear regression is performed on the current or past time-series sensor values ​​and the predicted values. That is, a linear function represented by the following equation is calculated as the regression line.

[0052]

number

[0053] where a_hat is the sum of squared deviations of x, S xx , the sum of the products of deviations of x and y S xy , the average value of y, y_ave, and the average value of x, x_ave, are expressed as follows:

[0054]

number

[0055] Also, b_hat is the sum of squared deviations of x, S xx , the sum of the products of deviations of x and y S xy Using this, it is expressed as follows:

[0056]

number

[0057] Such a configuration using a linear regression line can suitably handle linear changes in the object.

[0058] The regression line calculation unit 105 provides information about the regression line to the future prediction unit 106, the display control unit 151, and the confidence interval calculation unit 107.

[0059] In this embodiment, a regression line is calculated, but an approximation line may be calculated. Also, a regression line of second degree or higher may be calculated.

[0060] After the regression line is calculated, the confidence interval calculation unit 107 calculates the confidence interval (S6). More specifically, in this embodiment, the confidence interval calculation unit 107 calculates a straight line corresponding to the confidence interval based on the following equation.

[0061]

number

[0062] Note that t is the t value, N is the number of samples, α is the reliability coefficient, and V_hat ε represents the unbiased variance with N-2 degrees of freedom for the regression line. The reliability coefficient α is 0.9 in this embodiment.

[0063] The confidence interval calculation unit 107 provides the display control unit 151 with a straight line corresponding to the confidence interval, which is the calculation result.

[0064] With this configuration, it is possible to provide a margin of error for prediction based on the confidence interval.

[0065] After the confidence interval calculation process, the future prediction unit 106 performs a future prediction process, which in this embodiment is a lifespan prediction process for the target device 2 (S7). More specifically, the lifespan is determined based on the relationship between the regression line and a predetermined threshold (or standard). For example, in this embodiment, the time when the regression line becomes equal to or smaller than the predetermined threshold is set as the predicted lifespan. The future prediction unit 106 provides the predicted lifespan to the display control unit 151.

[0066] After the future prediction process, the display control unit 151 performs a process of displaying the predicted life obtained from the future prediction unit 106, the regression line obtained from the regression line calculation unit 105, and the confidence interval obtained from the confidence interval calculation unit 107 on the display device 19 (S8) (see also FIG. 5 for an example). After this display process, the process ends.

[0067] According to this configuration, the user can visually check the time series values, the forecast values, the regression line, the confidence interval, and the relationships between them. Therefore, for example, it is possible to perform future predictions for the object in the shortest and longest time periods from the boundaries of the confidence interval.

[0068] FIG. 4 is a conceptual diagram of a life prediction algorithm according to this embodiment. Note that in FIG. 4, description of feature quantities is omitted for ease of understanding. As is clear from FIG. 4, the sensor data from the present to the past (M+1) steps obtained from the sensor 18 is directly provided to the life prediction model. In addition, the sensor data is provided as input to a one-step future prediction trained model, a two-step future prediction trained model, . . ., N-step future prediction trained model, and a one-step future sensor prediction value, a two-step future sensor prediction value, . . ., N-step future sensor prediction value are generated, respectively. These generated prediction values ​​are also provided to the life prediction model.

[0069] In the life prediction model, a predicted life is generated from the relationship between a regression line generated from the current (M+1) step (sensor value 0 steps in the past) and past sensor values ​​and the predicted value for N steps, and a predetermined threshold value.

[0070] 5 is a diagram showing an example of a screen displayed on the display device 19 by the display control unit 151. The screen includes a two-dimensional xy graph, where x represents time (steps) and y represents the sensor value. Furthermore, x0 represents the current sensor value (or the sensor value 0 steps past), with past sensor values ​​displayed to the left of that and predicted values ​​displayed to the right of that. Note that in the diagram, the number of past sensor values ​​and the number of predicted values ​​are merely examples.

[0071] The graph also displays two lines: a regression line (shown as a solid line in the figure) obtained based on the current or past sensor values ​​and predicted values, and a straight line (shown as a dashed line in the figure) corresponding to the confidence interval boundary.

[0072] In this embodiment, the threshold value is y=0. Therefore, in the example shown in the figure, x1 is the predicted life. 11 indicates the shortest predicted life expectancy with confidence interval, and x 12 By displaying these values, the user can be shown the predicted lifespan, the minimum predicted lifespan, and the maximum predicted lifespan.

[0073] As described above, according to the configuration of this embodiment, future prediction can be performed based on the predicted value obtained using the machine learning technology and the time series value up to that point. This allows future prediction to be performed with high accuracy. In addition, since a regression line is used, it is possible to increase explainability compared to the case where a predicted value is obtained directly using the machine learning technology. In other words, it is possible to provide a prediction technology that is highly accurate and yet explainable in a technology for predicting the future of an object based on a signal obtained from the object.

[0074] (2. Modifications) In the above embodiment, the prediction data generating unit 103 is configured to output a plurality of predicted values ​​using a plurality of trained models, but the present invention is not limited to such a configuration. Therefore, a number of trained models less than the number of predicted values ​​may be prepared. For example, a single trained model may be used to simultaneously generate predicted values ​​for a plurality of different time steps.

[0075] In the above embodiment, the future prediction unit 106 is described as performing a life prediction, but the present invention is not limited to such a configuration. Therefore, a threshold value may not be set and a future performance prediction may simply be performed.

[0076] In the above embodiment, the regression line is calculated based on a plurality of predicted values, but the present invention is not limited to such a configuration. Therefore, a single predicted value may be generated, and a regression line may be calculated from a current or past value and the single predicted value.

[0077] Although the embodiments of the present invention have been described above, the above embodiments merely show some of the application examples of the present invention, and the technical scope of the present invention is not limited to the specific configurations of the above embodiments. Moreover, the above embodiments can be appropriately combined within a range where no contradiction occurs. [Industrial Applicability]

[0078] The present invention can be used in various industries that utilize machine learning technology. [Explanation of symbols]

[0079] 1. Information processing device 10 Control section 11 Storage section 12 Input section 15 Display control section 16 Communications Department 17 I / O section 18 Sensors 19 Display device

Claims

1. a time series data acquisition unit that acquires a time series value obtained from the object; A prediction data generation unit that generates a prediction value following the time series value based on the time series value and a trained model; a regression line calculation unit that calculates a regression line based on the time series value and the predicted value; A future prediction unit that performs a future prediction regarding the object based on the regression line; A prediction device comprising:

2. The trained model includes a plurality of trained models that predict different time steps, The prediction device according to claim 1 , wherein the prediction data generation unit generates a plurality of the prediction values ​​based on the time-series values ​​and each of the trained models.

3. The prediction device according to claim 1 , wherein the future prediction is a prediction regarding a lifespan of the object.

4. The prediction device according to claim 1 , wherein the regression line is a straight line.

5. The prediction data generating unit further includes a preprocessing unit that extracts features of the time-series values. The prediction device according to claim 1 , wherein the prediction data generation unit generates a prediction value following the time-series value based on the feature amount extracted in the preprocessing unit and the trained model.

6. The prediction device according to claim 5 , wherein the feature amount includes a moving average of the time series values ​​and / or a difference value between time steps of the time series values.

7. The prediction device according to claim 1 , further comprising a confidence interval calculation unit that calculates a confidence interval based on the regression line.

8. The prediction device according to claim 1 , further comprising a display control unit that displays the time-series values, the predicted values, and the regression line.

9. The prediction device according to claim 7 , further comprising a confidence interval display control unit that displays the time series values, the predicted values, the regression line, and the confidence intervals.

10. A time series data acquisition unit that acquires a time series value obtained from the object; A prediction data generation unit that generates a prediction value following the time series value based on the time series value and a trained model; a regression line calculation unit that calculates a regression line based on the time series value and the predicted value; A future prediction unit that performs a future prediction regarding the object based on the regression line; A prediction system comprising:

11. On the computer, A time series data acquisition step of acquiring a time series value obtained from the object; A prediction data generation step of generating a prediction value following the time series value based on the time series value and a trained model; a regression line calculation step of calculating a regression line based on the time series values ​​and the predicted values; A future prediction step of performing a future prediction regarding the object based on the regression line; A prediction method that executes the following:

12. On the computer, A time series data acquisition step of acquiring a time series value obtained from the object; A prediction data generation step of generating a prediction value following the time series value based on the time series value and a trained model; a regression line calculation step of calculating a regression line based on the time series values ​​and the predicted values; A future prediction step of performing a future prediction regarding the object based on the regression line; A prediction program that executes the following: