Information processing apparatus, information processing system, and information processing method

Abstract

To improve the accuracy of a model generated in sparse modeling and prevent a decrease in an amount of information obtained from input data, in an information processing apparatus.SOLUTION: A model calculation unit calculates a correlation coefficient between a plurality of pieces of data, performs smoothing processing on discontinuous first data that is discontinuous between data points among the plurality of pieces of data, determines whether the correlation coefficient between the first data and second data among the plurality of pieces of data is equal to or greater than a threshold value, and performs the smoothing processing on the second data whose correlation coefficient is equal to or greater than the threshold as a result of the determination.SELECTED DRAWING: Figure 2

Description

[Technical field]

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

[0002] A sparse modeling technique has been proposed that extracts time differential equations related to physical phenomena using time series data.

[0003] A related technique is, for example, Patent Document 1. In Patent Document 1, an appropriate model of a physical phenomenon is generated by correcting generation probabilities and hyperparameters of machine learning based on a loss function, and extracting basis functions based on the generation probabilities. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2022-75213 Summary of the Invention [Problem to be solved by the invention]

[0005] In Patent Document 1, the temperature of an electronic device is predicted using data acquired from sensors provided in the electronic device, such as a temperature sensor, a flow rate sensor, a current sensor, and a voltage sensor.

[0006] However, the time series data actually acquired from the sensor may contain noise. In addition, the time resolution of the time series data may be low. In addition, the time series data may be flag data (1 / 0 data) of the device. For this reason, the time series data may be discontinuous between data points. When the time series data is discontinuous between data points, the data may not be differentiated appropriately during the calculation process for generating the model.

[0007] As a result, it may not be possible to obtain an appropriate model, and in some cases, it may not be possible to perform differentiation at all. Hereinafter, data that gives rise to such problems will be called non-continuous data (discontinuous data).

[0008] To address the above issues, there is a method of smoothing the data, which makes it possible to perform differential processing on the data, but this leads to the following two further issues:

[0009] The first issue is that when generating a model using two or more parameters among the parameters (such as temperature, pressure, flow rate, and current) consisting of data obtained from each sensor, the change in numerical value due to smoothing processing affects the relationship between the data of the smoothed parameters and the non-smoothed parameters. In sparse modeling, where the relationship between data is important, a highly accurate model cannot be obtained if the correlation between data is shifted.

[0010] The second issue is that smoothing reduces the amount of information contained in the input data. In other words, it becomes impossible to extract the phenomena contained in the data. Therefore, it is better to have fewer parameters for smoothing. This is particularly important in sparse modeling, which requires less data than deep learning.

[0011] An object of the present invention is to improve the accuracy of a model generated in sparse modeling in an information processing device, and to prevent a decrease in the amount of information obtained from input data. [Means for solving the problem]

[0012] An information processing device of one embodiment of the present invention is an information processing device having a model calculation unit that performs sparse modeling, which performs sparse estimation to generate a governing equation for a phenomenon based on data of a plurality of parameters related to the phenomenon obtained from a plurality of sensors, to generate a predetermined model, wherein the model calculation unit calculates correlation coefficients between each of the plurality of data, performs a smoothing process on discontinuous first data that is discontinuous between data points among the plurality of data, determines whether the correlation coefficient between the first data and second data among the plurality of data is greater than or equal to a threshold, and performs the smoothing process on the second data whose correlation coefficient is greater than or equal to the threshold as a result of the determination. Effect of the Invention

[0013] According to one aspect of the present invention, in an information processing device, it is possible to improve the accuracy of a model generated in sparse modeling and prevent a decrease in the amount of information obtained from input data. [Brief description of the drawings]

[0014] [Figure 1] 5 is a flowchart showing data processing in a model calculation unit of the information processing device according to the first embodiment of the present invention. [Diagram 2] This is a subroutine in the flowchart shown in FIG. [Diagram 3] 1 is a diagram illustrating a configuration of an information processing system according to a first embodiment of the present invention. [Figure 4A] FIG. 13 is a diagram illustrating an example of visualization of input data. [Figure 4B] FIG. 13 is a diagram illustrating an example of visualization of input data. [Figure 4C] FIG. 13 is a diagram illustrating an example of visualization of input data. [Diagram 5] FIG. 13 is a diagram illustrating an example of visualization of correlation coefficients. [Figure 6] 13A and 13B are diagrams illustrating an example of visualization of data before and after smoothing processing. [Figure 7A] FIG. 13 is a diagram illustrating an example of discontinuous data. [Figure 7B]FIG. 13 is a diagram illustrating an example of discontinuous data. [Figure 8A] FIG. 13 is a diagram illustrating an example of continuous data. [Figure 8B] FIG. 13 is a diagram illustrating an example of continuous data. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. EXAMPLES

[0016] First, the configuration of the information processing system according to the first embodiment will be described with reference to Fig. 3. In Fig. 3, arrows indicate the flow of data.

[0017] The data to be processed is acquired from a sensor 1 (assuming that the number of sensors is N, sensor_1 to sensor_N) installed in an apparatus or the like. The acquired data is stored in a cloud storage or an information storage device 2. The same number of parameters (X1 to X N ) is obtained. Data is sent from the information storage device 2 to the information processing device 3 as required.

[0018] The information processing device 3 has a model calculation unit 10 and a simulation system 11. A model generated by the model calculation unit 10 is implemented in the simulation system 11. In addition, input and instructions to the information processing device 3 can be given to the system via an input device 4. A visualization device 5 visualizes and outputs the processing results of the information processing device 3. Here, the visualization device 5 is composed of, for example, a display device.

[0019] Thus, the information processing system of the first embodiment includes a plurality of sensors 1 (sensor_1 to sensor_N), a cloud storage or information storage device 2 for storing data, an information processing device 3, an input device 4, and a visualization device 5.

[0020] The model calculation unit 10 uses data (parameters (X1 to XN )) to perform sparse modeling to generate the model.

[0021] The visualization device 5 visualizes and displays at least one of the data acquired from the sensor 1 (see FIGS. 4A to 4C), the correlation coefficient between multiple pieces of data (see FIG. 5), and the data before and after smoothing processing (see FIG. 6).

[0022] In addition to the above data, the visualization device 5 visualizes and displays the generative model generated by the model calculation unit 10, the results of the simulation in the simulation system 11, input information from the input device 4, etc. Here, the input device 4 performs predetermined inputs and instructions to the information processing device 3.

[0023] Next, the flow of data processing in the model calculation unit 10 of the information processing device 3 will be described with reference to FIG.

[0024] In step S10, parameters (data) to be processed are selected from among multiple parameters acquired from the sensor. In step S20, data of the selected parameters is acquired from cloud storage or the information storage device 2. In step S30, when multiple parameters are selected, the correlation coefficient corr between the parameters is calculated.

[0025] In step S40, the time series data of each selected parameter and the calculated correlation coefficient are output to the visualization device 5. In step S50, the correlation coefficient is used as a reference to determine whether there is a problem with the selected parameters. If it is necessary to reselect parameters, the process returns to step S10.

[0026] In step S60, parameters that include discontinuous data are extracted from the selected parameters. This may be determined subjectively by an operator, automatically using the difference between data points, or the discontinuity may be discovered when a numerical error is returned during modeling in step S100.

[0027] Here, non-continuous data refers to discontinuous data that is discontinuous between data points. Discontinuous data is, for example, discrete data or noise data that includes high-frequency noise. Discrete data is, for example, flag data, binary data, or a step function. Discontinuous data is, for example, time-series data that is discontinuous between data points in the time axis direction.

[0028] An example of discontinuous data is shown in Figures 7A and 7B. Also, an example of continuous data is shown in Figures 8A and 8B. Figure 7A shows flag data, and Figure 7B shows noise data including high frequency noise.

[0029] It can be seen that the continuous data shown in Figures 8A and 8B is time series data that is continuous between data points in the time axis direction, whereas the discontinuous data shown in Figures 7A and 7B is time series data that is discontinuous between data points in the time axis direction.

[0030] Step S70 is a subroutine for determining whether or not to perform smoothing processing on the selected parameters, and will be described in detail with reference to FIG.

[0031] In step S80, it is confirmed whether all the selected parameters have been processed in step S70. If there are parameters that have not been confirmed, the process returns to step S70. When confirmation or processing of all the selected parameters has been completed, the time series data of the parameters that have been smoothed in step S90 is output to the visualization device 5.

[0032] In step S100, sparse modeling is performed using the processed time series data. In step S110, the generated model is corrected. In step S120, the model is output to the simulation system 11.

[0033] Next, the process of subroutine step S70 in the flowchart of FIG. 1 will be described with reference to FIG.

[0034] In step S130, parameters to be processed in the subroutine (herein referred to as processing parameters) are selected from the parameters selected in step S10.

[0035] In step S140, it is checked whether the processing parameters are parameters including the non-continuous data extracted in step S60. If yes, in step S170, a data smoothing process is performed. The smoothing process removes high frequency data from the time series data using, for example, a low pass filter. Then, in step S180, the data is recorded.

[0036] On the other hand, if the processing parameters do not include discontinuous data in step S140, that is, if the answer is no, then in step S150 it is determined whether the parameters selected in step S10 include discontinuous data other than the processing parameters. If the answer is yes, then in step S160 the magnitude of the correlation coefficient |corr| between the processing parameters and the parameters including discontinuous data extracted in step S60 is compared with an arbitrarily set correlation coefficient threshold C.

[0037] If |corr|≧C, proceed to step S170 and perform smoothing processing of the processing parameters. If not, smoothing processing is not performed. For example, when it is assumed that there is a correlation, it is preferable to set the threshold value C of the correlation coefficient to 0.2 or more.

[0038] Next, with reference to Fig. 4A to Fig. 4C, Fig. 5 and Fig. 6, an example of visualization of the results of data processing in the present embodiment 1 will be described using an example of heater heating. Here, Fig. 4A to Fig. 4C are diagrams showing a visualization example of data acquired from a sensor. Fig. 5 is a diagram showing a visualization example of correlation coefficients between input parameters. Fig. 6 is a diagram showing a visualization example of data before and after smoothing processing.

[0039] Fig. 4A is data obtained from the sensor, showing the heater heat generation amount (X1). Fig. 4B is data obtained from the sensor, showing the room temperature (X2). Fig. 4C is data obtained from the sensor, showing the heater output control signal (X3).

[0040] Assume that a time-varying model of X1 is created by sparse modeling using the data shown in Figures 4A to 4C. First, select these parameters, calculate the correlation coefficient, and check whether non-continuous data is included (steps S10 to S60 in Figure 1).

[0041] Among Figs. 4A to 4C, high frequency vibrations are observed especially in X3, and discontinuity between data can be confirmed, so X3 is set as the parameter to be smoothed. Some vibrations (noise) are also observed in X2, and in reality, this may also be subject to smoothing processing, but for the sake of explanation, only X3 is considered here. An example of visualizing these correlation coefficients is shown in Fig. 5.

[0042] The results shown in Figures 4A to 4C and Figure 5 are visualized in step S40 in Figure 1. From the obtained correlation coefficient, it is determined in step S70 whether or not smoothing processing is performed. The correlation coefficients for X3 are 0.36 for X1 and -0.13 for X2, and if the correlation coefficient threshold C is set to 0.2, X1 satisfies |corr|>0.2 in step S160. In other words, X1 is a parameter that requires data smoothing processing.

[0043] As a result, in step S70, X1 and X3 are smoothed, while X2 is not processed, and modeling is performed in step S100 using the data after the smoothing process.

[0044] Figure 6 visualizes X3 before and after the smoothing process, which is visualized in step S90 in Figure 1. Here, it can be seen that the smoothing process using the moving average reduces discontinuities between data points and suppresses amplitude.

[0045] As described above, in the present embodiment 1, a white-box low-dimensional model is generated in a data-driven manner using sparse modeling. At this time, the amount of data required for sparse modeling is relatively small.

[0046] Therefore, since a model can be constructed sequentially using data obtained from the sensors, in addition to modeling components such as the heater heating described in this embodiment, it can also be applied to applications such as building cyber-physical systems, optimizing operations, and real-time control.

[0047] As mentioned above, sparse modeling is characterized by its ability to model with a small amount of data. On the other hand, if the small amount of data contains noise or flag data that is difficult to obtain physical phenomena, the accuracy of the generative model decreases. Therefore, this data requires smoothing processing.

[0048] If smoothing is performed only on data that contains noise, consistency between parameters cannot be achieved, and in this case too, a desirable model cannot be constructed. Smoothing is also necessary for data that is correlated with the smoothed data.

[0049] If all data is smoothed, data that does not need smoothing may also be processed, and there is a concern that the phenomena and factors that should be modeled may not be visible. Therefore, it is necessary to extract learning parameters so that the amount of data to be smoothed is kept to a minimum.

[0050] Therefore, in the first embodiment, parameters including flag data and high frequency noise are extracted from the learning parameters, and parameters with correlation coefficients of a threshold value (e.g., 0.2) or more are extracted. Then, the parameters including flag data and high frequency noise and the parameters with correlation coefficients exceeding the threshold value are smoothed by the same method, and the other parameters are not smoothed.

[0051] In this way, in the first embodiment, parameters of data including flag data and high-frequency noise are extracted from a plurality of learning parameters in the measurement data used for sparse estimation (sparse regression, sparse modeling). Then, among the learning parameters, parameters having a correlation coefficient with parameters consisting of flag data and noise data that is equal to or greater than a threshold value are also extracted, and the time series data of the extracted parameters is subjected to smoothing processing using a low-pass filter, function fitting, or the like.

[0052] In this way, the information processing device 3 of the first embodiment (see FIG. 3) The system includes a model calculation unit 3 that performs sparse modeling, which performs sparse estimation to generate a governing equation for a phenomenon based on data of a plurality of parameters relating to the phenomenon acquired from a plurality of sensors 1, to generate a model.

[0053] The model calculation unit 3 calculates correlation coefficients between the plurality of data, and performs a smoothing process on discontinuous first data (for example, X3 in FIG. 5) that is discontinuous between data points among the plurality of data.

[0054] Then, it is determined whether the correlation coefficient between the first data (e.g., X3 in FIG. 5) and the second data (e.g., X1 and X2 in FIG. 5) among the multiple data is equal to or greater than a threshold, and as a result of the determination, a smoothing process is performed on the second data (e.g., X1 in FIG. 5) whose correlation coefficient is equal to or greater than the threshold.

[0055] As a result of the determination, the model calculation unit 3 does not perform the smoothing process on the second data (for example, X2 in FIG. 5) whose correlation coefficient is less than the threshold value.

[0056] It is preferable that the model calculation unit 3 performs the smoothing process on discrete data or noise data including high frequency noise as the discontinuous first data and second data.

[0057] For example, the model calculation unit 3 performs the smoothing process on flag data, binary data, or a step function as the discrete data.

[0058] For example, the model calculation unit 3 performs the smoothing process by removing high frequency noise from the noise data using a low pass filter. Alternatively, the model calculation unit 3 may perform the smoothing process by using function fitting.

[0059] For example, the discontinuous first data is time series data that is discontinuous between the data points in the time axis direction. When a numerical error is returned during sparse modeling, the model calculation unit 3 changes the threshold value of the correlation coefficient, determines whether the correlation coefficient between the first data and a second data among the plurality of data is equal to or greater than the threshold value, and performs the smoothing process on the second data whose correlation coefficient is equal to or greater than the threshold value as a result of the determination.

[0060] Moreover, the model calculation unit 3 performs sparse modeling using the first data on which the smoothing process has been performed and the second data on which the smoothing process has been performed to generate a model.

[0061] The system further includes a simulation system 11 that uses the model generated by the model calculation unit 3 .

[0062] According to the above-mentioned first embodiment, it is possible to improve the accuracy of a model generated in sparse modeling and prevent a decrease in the amount of information obtained from input data. EXAMPLES

[0063] The configuration of an information processing system according to a second embodiment will be described with reference to FIG.

[0064] In the second embodiment, data obtained from a sensor 1 shown in FIG. 3 is stored in a cloud storage or an information storage device 2, and is processed by a model calculation unit 10 as a group of multiple time-series data.

[0065] That is, when sparse modeling is performed in step S100 of Fig. 1, modeling can be performed using time-series data in various past situations. At this time, the processes from steps S30 to S90 are performed for each group of classified data, for example, classified by date and time.

[0066] According to the second embodiment, since a plurality of situations can be taken into consideration, it is expected that a sparser model or a model with high versatility can be obtained.

[0067] As described above, in the information processing system of the second embodiment, the model calculation unit 3 of the information processing device 3 performs sparse modeling using past data stored in the cloud storage or the information storage device 2 to generate a model.

[0068] Here, the information processing device 3 shown in FIG. 3 is composed of, for example, an input device, an output device, a memory, a storage device, and a computer having a CPU.

[0069] The model calculation unit 10 and the simulation system 11 shown in FIG. 3 realize their "functions" by, for example, executing a program using a processor (such as a CPU).

[0070] For example, the model calculation unit 10 realizes a model calculation function by executing a program using a processor (such as a CPU), and the simulation system 11 realizes a simulation function by executing a program using a processor (such as a CPU).

[0071] The visualization device 5 shown in FIG. 13 is configured, for example, by a display device.

[0072] According to the above embodiment, it is possible to reduce the deviation in correlation between data due to smoothing processing for input data used in sparse modeling, and improve the accuracy of the generated model. Furthermore, by minimizing the number of parameters to be smoothed, it is possible to maximize the information obtained from the data. This makes it possible to obtain a governing equation that can appropriately reproduce a phenomenon. [Explanation of symbols]

[0073] 1 Sensor 2. Cloud storage or information storage device 3. Information processing equipment 4 Input Devices 5 Visualization device 10 Model calculation section 11 Simulation System

Claims

1. An information processing device having a model calculation unit that performs sparse modeling to perform sparse estimation to generate a governing equation of a phenomenon based on data of a plurality of parameters related to the phenomenon obtained from a plurality of sensors, thereby generating a predetermined model, The model calculation unit is Calculating correlation coefficients between the plurality of data, performing a smoothing process on discontinuous first data among the plurality of data points; determining whether the correlation coefficient between the first data and a second data among the plurality of data is equal to or greater than a threshold value; The information processing apparatus according to claim 1, wherein the smoothing process is performed on the second data whose correlation coefficient is equal to or greater than the threshold as a result of the determination.

2. The model calculation unit is The information processing apparatus according to claim 1 , wherein the smoothing process is not performed on the second data for which the correlation coefficient is less than the threshold as a result of the determination.

3. The model calculation unit is 2 . The information processing apparatus according to claim 1 , wherein the smoothing process is performed on discrete data or noise data including high frequency noise as the discontinuous first data.

4. The model calculation unit is 4. The information processing apparatus according to claim 3, wherein the smoothing process is performed on flag data, binary data or a step function as the discrete data.

5. The model calculation unit is 4. The information processing apparatus according to claim 3, wherein the smoothing process is performed by removing high frequency noise from the noise data using a low pass filter.

6. The model calculation unit is 5. The information processing apparatus according to claim 4, wherein the smoothing process is performed using function fitting.

7. The discontinuous first data is 2. The information processing apparatus according to claim 1, wherein the data is time-series data in which the data points are discontinuous in the time axis direction.

8. The model calculation unit is modifying the correlation coefficient threshold if a numerical error is returned when performing the sparse modeling; determining whether the correlation coefficient between the first data and a second data among the plurality of data is equal to or greater than the threshold value; 2 . The information processing apparatus according to claim 1 , wherein the smoothing process is performed on the second data whose correlation coefficient is equal to or greater than the threshold as a result of the determination.

9. The model calculation unit is The information processing apparatus according to claim 1 , further comprising: performing the sparse modeling using the first data on which the smoothing process has been performed and the second data on which the smoothing process has been performed to generate the model.

10. 2. The information processing apparatus according to claim 1, further comprising a simulation system that uses the model generated by the model calculation unit.

11. The information processing device according to claim 1 ; A plurality of said sensors; A storage device for storing the data; having The model calculation unit is An information processing system, comprising: a storage device that stores the past data and generates the model;

12. A visualization device for displaying a result of the calculation in the information processing device, The visualization device includes: A display device is provided. The information processing system according to claim 11, characterized in that at least one of the data acquired from the sensor, the correlation coefficient between a plurality of the data, and the data before and after the smoothing process is visualized and displayed.

13. Further comprising an input device, The input device is 12. The information processing system according to claim 11, wherein predetermined inputs and instructions are given to said information processing device.

14. 1. An information processing method comprising: a model calculation step of performing sparse modeling to perform sparse estimation to generate a governing equation for a phenomenon based on data of a plurality of parameters related to the phenomenon obtained from a plurality of sensors, thereby generating a predetermined model, The model calculation step is performed by a model calculation unit. Calculating correlation coefficients between the plurality of data, performing a smoothing process on discontinuous first data among the plurality of data points; determining whether the correlation coefficient between the first data and a second data among the plurality of data is equal to or greater than a threshold value; an information processing method, characterized in that the smoothing process is performed on the second data whose correlation coefficient is equal to or greater than the threshold as a result of the determination.

15. The model calculation step is performed by the model calculation unit. The information processing method according to claim 14 , wherein the smoothing process is not performed on the second data for which the correlation coefficient is less than the threshold as a result of the determination.

Images