Simulation equipment, simulation method

Abstract

To obtain an appropriate predictive model using sparse identification methods, and to perform simulations using that model. [Solution] A simulation device that generates a differential equation model for calculating an objective variable based on training data and predicts the objective variable using the differential equation model, comprising: a training data generation unit that determines an objective variable and a plurality of first state variables from time series data used as training data, and uses the value obtained by integrating or accumulating the data that will be the objective variable at each time step within the time series range as a second state variable, and generates training data including the first state variables and the second state variables; a model generation unit that generates a differential equation model by sparse identification based on the training data; a simulation unit that performs a simulation to predict the objective variable using the differential equation model; and an output unit that outputs the differential equation model and / or the simulation results.

Description

[Technical Field]

[0001] This invention relates to a simulation device and a simulation method for generating a model using data obtained from sensors and the like. [Background technology]

[0002] Patent Document 1 describes a device for generating predictive models of crop production performance, which involves selecting combinations of variables based on cumulative or mean values ​​of past weather data and soil component variables through regression analysis, and generating a predictive model.

[0003] Non-patent document 1 proposes a sparse identification method for nonlinear dynamic systems as a modeling technique using data. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2022-136058 [Non-patent literature]

[0005] [Non-Patent Document 1] SL Brunton, JL Proctor, JN Kutz, Discovering governing equations from data by sparse identification of nonlinear dynamical systems, PNAS, Vol.113 (2016), pp.3932-3937. [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] When generating a predictive model using the sparse identification method described in Non-Patent Literature 1, if the target variable data in the training data exhibits discontinuous behavior, there is a problem in that an appropriate model cannot be generated because the differentiation of discontinuous data increases noise or generates very large values ​​(spikes).

[0007] This invention has been made in view of the above problems, and aims to provide a suitable predictive model using a sparse identification method even when the data of the target variable exhibits discontinuous behavior, and to provide a technology equipped with such a model. [Means for solving the problem]

[0008] To achieve the above objective, the simulation device of the present invention is a simulation device that generates a differential equation model for calculating an objective variable based on training data and predicts the objective variable using the differential equation model, and is configured to include: a training data generation unit that determines the objective variable and a plurality of first state variables from data in a time series range used as training data, and uses the value obtained by integrating or accumulating the data that becomes the objective variable at each time step within the time series range as a second state variable, and generates training data including the first state variable and the second state variable; a model generation unit that generates the differential equation model by sparse identification based on the training data; a simulation unit that performs a simulation to predict the objective variable using the differential equation model; and an output unit that outputs the differential equation model and / or the simulation results. [Effects of the Invention]

[0009] According to the present invention, it is possible to obtain an appropriate predictive model using a sparse identification method and to perform simulations using that model. [Brief explanation of the drawing]

[0010] [Figure 1A]It is a diagram showing an example of the configuration and data flow of a simulation device according to the first embodiment of the present invention. [Figure 1B] It is a diagram showing an example of the outline of a computer. [Figure 2] It is a flowchart showing an example of the flow from model generation to simulation according to the first embodiment of the present invention. [Figure 3] It is a flowchart showing an example of the generation processing procedure of training data according to the first embodiment of the present invention. [Figure 4A] It is a diagram showing an example of discontinuous measurement data before integration processing and values after integration processing according to the first embodiment of the present invention. [Figure 4B] It is a diagram showing an example of the integrated value of power consumption calculated by the training data generation unit 4 from equation (1) in step S160. [Figure 5A] It is a diagram showing an example of the result of simulation using a conventional method. [Figure 5B] It is a diagram showing an example of the result of simulation according to the first embodiment of the present invention. [Figure 6] It is a diagram showing an example of the configuration and data flow of a simulation device according to the second embodiment of the present invention. [Figure 7] It is a flowchart showing an example of the flow from model generation to simulation according to the second embodiment of the present invention. Embodiments for Carrying Out the Invention

[0011] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The embodiments are examples for explaining the present invention, and for the sake of clarity of explanation, appropriate omissions and simplifications have been made. The present invention can also be implemented in various other forms. Unless otherwise particularly limited, each component may be singular or plural. In the drawings, the positions, sizes, shapes, ranges, etc. of the components shown may not represent the actual positions, sizes, shapes, ranges, etc. in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the positions, sizes, shapes, ranges, etc. disclosed in the drawings.

[0012] Examples of various types of information may be described using terms such as "table," "list," and "queue," but these types of information may also be represented by other data structures. For example, various types of information such as "XX table," "XX list," and "XX queue" may be referred to as "XX information." When describing identification information, terms such as "identification information," "identifier," "name," "ID," and "number" are used, and these terms are interchangeable.

[0013] When there are multiple components with the same or similar function, they may be described using the same symbol but with different subscripts. Furthermore, when it is not necessary to distinguish between these multiple components, the subscripts may be omitted in the description.

[0014] In the examples, the processes performed by executing a program may be described. Here, the computer executes the program using a processor (e.g., CPU, GPU) and performs the processing defined in the program using memory resources (e.g., memory) and interface devices (e.g., communication ports). Therefore, the main entity performing the processing by executing the program may be the processor. Similarly, the main entity performing the processing by executing the program may be a controller, device, system, computer, or node having a processor. The main entity performing the processing by executing the program may be an arithmetic unit, and may include dedicated circuits that perform specific processing. Here, dedicated circuits include, for example, FPGAs (Field Programmable Gate Arrays), ASICs (Application Specific Integrated Circuits), CPLDs (Complex Programmable Logic Devices), etc.

[0015] The program may be installed on the computer from the program source. The program source may be, for example, a program distribution server or a storage medium readable by the computer. If the program source is a program distribution server, the program distribution server includes a processor and storage resources for storing the program to be distributed, and the processor of the program distribution server may distribute the program to other computers. In addition, in the embodiment, two or more programs may be implemented as one program, or one program may be implemented as two or more programs.

[0016] The following examples will be described with reference to the drawings. [Examples]

[0017] A simulation apparatus according to an embodiment of the present invention will be described below with reference to Figures 1A to 5.

[0018] Figure 1A is a diagram showing the system and information flow comprising a data acquisition unit that acquires time-series measurement data from sensors attached to the device, a storage unit that stores the acquired measurement data, an input unit that inputs conditions for generating a prediction model and simulation, a training data generation unit that generates training data for generating a prediction model, a model generation unit for generating a prediction model, a simulation unit that executes a simulation using the generated prediction model, and an output unit 7 that outputs the simulation results. Figure 2 is a flowchart showing the flow of prediction model generation and simulation. Figure 3 is a flowchart showing the training data generation processing procedure. Figure 4 is a diagram showing examples of discontinuous measurement data before integration and values ​​after integration to illustrate an embodiment. Figure 5 is a diagram showing the simulation results in this embodiment.

[0019] [Configuration and Features of the Example] The configuration and features of this embodiment will be explained using Figures 1A to 3.

[0020] Figure 1A shows the configuration of a system including the simulation device 100 according to this embodiment. The arrows indicate the flow of information. The simulation device 100 is composed of a training data generation unit 4, a model generation unit 5, a simulation unit 6, and an output unit 7 that outputs the generated prediction model and simulation results. The training data generation unit 4 receives measurement data as needed from a storage unit 2, which stores measurement data acquired by a data acquisition unit 1 from sensors installed on equipment located outside the simulation device 100, and performs processing related to training data generation.

[0021] In the model generation unit 5, a predictive model for the target variable is generated using sparse identification with the generated training data. The predictive model is expressed as a differential equation in polynomial form. In this case, the model generation unit 5 may generate multiple predictive models for multiple target variables. The generated predictive models are incorporated into the simulator of the simulation unit 6, and the simulation is executed. Information necessary for the above processing, such as the conditions for generating the training data and the conditions for the simulation, is input from the input unit 3.

[0022] The simulation device 100 shown in Figure 1A can be realized by a general-purpose computer 1600, which includes, for example, a CPU 1601, memory 1602, an external storage device 1603 such as an HDD (Hard Disk Drive), a reader 1607 for reading and writing information to a portable storage medium 1608 such as a CD (Compact Disk) or USB memory, an input device 1606 for receiving various types of information such as a keyboard and mouse, an output device 1605 such as a display for outputting various types of information that are input and used for processing, a communication device 1604 such as a NIC (Network Interface Card) for connecting to a communication network, and an internal communication line (called a system bus) 1609 such as a system bus that connects these.

[0023] Furthermore, various data stored in the simulation device 100 or used for processing (for example, a database such as the source code DB 109a) can be realized by the CPU 1601 reading and using it from memory 1602 or external storage device 1603. In addition, each part of the simulation device 100 (for example, the source code analysis unit 107, the source code comment generation unit 108, the design document preprocessing unit 111, and the design document terminology DB generation unit 112) can be realized by the CPU 1601 loading a predetermined program stored in the external storage device 1603 into memory 1602 and executing it.

[0024] The aforementioned programs and data may be stored (downloaded) from the storage medium 1608 via the reading device 1607, or from the network via the communication device 1604, into the external storage device 1603, and then loaded onto the memory 1602 and executed by the CPU 1601. Alternatively, they may be loaded directly onto the memory 1602 via the reading device 1607 from the storage medium 1608, or from the network via the communication device 1604, and then executed by the CPU 1601.

[0025] In the following example, the simulation device 100 is described as being composed of a single computer. However, all or part of these functions may be distributed across one or more computers, such as a cloud, and similar functions may be achieved by communicating with each other via a network.

[0026] In Figure 1B, the simulation device 100 is shown as an example where it has an input device 1606, which is hardware that functions as an input unit 3. However, it is not necessary for the simulation device 100 to have this input device; as shown in Figure 1A, it may be located outside the simulation device 100. Similarly, the output device 1605, which is hardware that functions as an output unit 7, may also be located outside the simulation device 100.

[0027] Figure 2 is a flowchart illustrating the flow from the generation of training data to the output of simulation results in the training data generation unit 4, model generation unit 5, simulation unit 6, and output unit 7 described in Figure 1A.

[0028] In step S10, the simulation time step Δt is sent from the input unit 3. s This is input to the simulation device 100. Simulation time step Δt s This refers, for example, to the time interval of measurement data acquired as the target of simulation during a predetermined simulation time from the current time to a predetermined time. Hereafter, this time step will be referred to as the simulation time step.

[0029] In step S20, the time step Δt in the prediction model generated by the model generation unit 5 is input from the input unit 3. m This is input to the simulation device 100. Prediction model time step Δt m Δt is the time interval of the training data used as the predictive model in the simulation. Here, the time step of the predictive model is Δt. m The time unit is the simulation time step Δt s It is desirable that the time unit be equivalent to that of the time unit. This allows for obtaining a prediction model with the time resolution required for the simulation. Furthermore, when dealing with multiple prediction models, by aligning the time steps of the prediction models during model generation, the time step Δt for each prediction model can be obtained. m Because the discrepancy is eliminated, adjustments during simulation setup become easier. In the following, this time step will be referred to as the prediction model time step.

[0030] In step S30, the target variable to be modeled and the state variables used for sparse identification are input to the simulation device 100 from the input unit 3. If multiple target variables are input here, multiple prediction models can be generated.

[0031] Step S40 is the process related to the generation of training data in the training data generation unit 4. Details are explained in Figure 3.

[0032] From step S50 onwards, the processing takes place in the model generation unit 5. First, the model generation unit 5 reads the training data generated by the training data generation unit 4 in step S40.

[0033] In step S60, the model generation unit 5 adjusts the hyperparameters for sparseness identification. At this time, for example, adjustments can be made using known optimization methods. Alternatively, they can be adjusted to appropriate values ​​to obtain a predictive model with the desired sparsity.

[0034] In step S70, the model generation unit 5 expresses the derivative of the second state variable (details will be described later using Figure 3), which is the integral of the target variable, in polynomial form by sparse identification. For sparse identification, the method described in Non-Patent Literature 1 may be used.

[0035] In step S80, the model generation unit 5 outputs the validation results of the generated prediction model from the output unit 7, for example, using the training data. Test data may be used instead of training data for validation. The model generation unit 5 compares the predicted value from the generated prediction model with the correct value of the target variable calculated from the training data (or test data) and evaluates the error between the predicted value and the correct value.

[0036] In step S90, the model generation unit 5 determines whether there is a problem with the generated prediction model based on the evaluation results in step S80. If the model generation unit 5 determines that there is a problem with the generated prediction model, such as the prediction results not achieving the expected prediction accuracy (step S90; No), it returns to step 60. On the other hand, if the model generation unit 5 determines that there is no problem with the generated prediction model (step S90; Yes), it proceeds to step S100. This completes the processing in the model generation unit 5.

[0037] From step S100, it is the processing in the simulation unit 6. In step S100, the simulation unit 6 incorporates the generated prediction model, which has been determined to be problem-free, into the simulation unit 6.

[0038] In step S110, the simulation unit 6 executes the simulation, and in step S120, the output unit 7 outputs the simulation result.

[0039] FIG. 3 is a flowchart showing the flow of the processing procedure regarding the generation of training data in step S40 described in FIG. 2.

[0040] In step S130, the training data generation unit 4 extracts measurement data for generating training data from the storage unit 2. At this time, the start point and the end point of the time range to be used as the training data are specified. By this specification, data within a time range from a certain time to another time in the training data can be extracted as the measurement data. When creating a plurality of prediction models, it is desirable that the time ranges of the training data for each prediction model be the same.

[0041] In step S140, the training data generation unit 4 processes the measurement data extracted in step S130 so as to be the prediction model time step Δt set in step S20 for generating the prediction model m Specifically, when the data extracted is finer than the prediction model time step Δt m the training data generation unit 4 coarsens it by averaging or the like, or extracts data so as to be at intervals of the prediction model time step Δt m When the data extracted is coarser than the prediction model time step Δt m the training data generation unit 4 interpolates it using an appropriate function. That is, from the viewpoint of reducing the prediction error of the generated prediction model, it is desirable that the time resolution of the data measured by the sensor or the like in the data acquisition unit 1 be finer.

[0042] In step S150, the training data generation unit 4 uses the data processed in step S140 to generate the target variable x input in step S30. p Define a (for example, power consumption) and state variables (for example, temperature and humidity). While there are often multiple state variables, here we define the first state variable x. m It should be written as follows.

[0043] In step S160, the training data generation unit 4 generates the oldest data within the time range specified in step S130, i.e., the time that marks the start of the time range, and generates the data from time step Δt. m Each dependent variable x p The second state variable X is obtained by integrating (for example, power consumption) over time, or by integrating or accumulating the value of the power consumption. p (For example, power consumption × Δt) m The second state variable X is calculated as (= power consumption Wh). p It is calculated using the following formula.

[0044]

number

[0045] Alternatively, it can be calculated using the following formula.

[0046]

number

[0047] In step S170, the training data generation unit 4 generates the prediction model time step Δt m Each time point and the first state variable x m And the second state variable X p The matrix consisting of and is output to a data file as training data. An example of the output data file is shown below.

[0048]

number

[0049] In this example, the prediction model time step Δt from the above starting point. m , dependent variable x p , the first state variable x m , the second state variable X p However, they are represented as matrix components from left to right.

[0050] As a prior art related to this embodiment, Patent Document 1 describes a predictive model generation device in which a start date and end date for performing calculations such as accumulation and averaging on each variable of meteorological data and soil components, which are explanatory variables, are set, and the calculation period for accumulation and averaging is performed, and regression analysis of production performance, which is the dependent variable, is performed using these calculated values. In Patent Document 1, accumulation processing is performed on the explanatory variables, but this is not described for the dependent variable. In conventional regression analysis and machine learning, if the accumulated value or mean of the dependent variable is used as an explanatory variable, it is difficult to provide it as an input value when using the generated predictive model.

[0051] On the other hand, in sparse identification, by using regularization such as Lasso regression, it is possible to generate a predictive model that does not include a second state variable, even if it includes explanatory variables that cannot be considered when using the predictive model, by adjusting the regularization parameters. For example, in step S70, when the model generation unit 5 performs sparse identification, the regularization parameter of the generated predictive model, the differential equation model, is increased so as not to include a second state variable, which is the integral value of the target variable. This makes it possible to treat the integral or cumulative value of the target variable to be predicted as a state variable, as in this embodiment.

[0052] [Examples of application] Next, using Figures 4 and 5, we will explain the power consumption prediction results using measurement data from an air conditioner (room air conditioner) as an example of a simulation applying this embodiment.

[0053] Figure 4A shows the power consumption x of the air conditioner, which is the dependent variable. pThis is an example of a measurement. In air conditioners, power consumption drops significantly when the thermostat is turned off (compressor stops). Therefore, the time series graph shows discontinuous behavior as shown in the figure. This is treated as an example of discontinuous data.

[0054] Figure 4B shows the integral value of power consumption calculated by the training data generation unit 4 from equation (1) in step S160. This is the second state variable X p A predictive model for power consumption is generated by sparse identification. The first state variable x m Factors used included indoor and outdoor temperature, indoor humidity, number of occupants, and solar radiation.

[0055] Figures 5A and 5B compare the predicted values ​​obtained from simulations using the generated prediction model (using training data as the simulation condition) with the ground truth values ​​of the power consumption training data. As will be explained in detail below, it can be seen that the discontinuous behavior of power consumption is reproduced even in the predicted values ​​using the method of this embodiment. Figure 5A is a figure showing the comparison of the above predicted values ​​and ground truth values ​​using the conventional method, and Figure 5B is a figure showing the comparison of the above predicted values ​​and ground truth values ​​using the method of this embodiment.

[0056] As shown in Figure 5A, the conventional method treats the target variable (xp) as a state variable (xm), generates a predictive model (dxp / dt=f(xm)), and plots the value obtained by integrating dxp / dt over time (=xp). In this example, when the room air conditioner's thermostat is turned off, the power consumption drops significantly, and the target variable, power consumption, exhibits discontinuous behavior. Differential processing of discontinuous data leads to increased noise and the generation of very large values ​​(spikes), making it impossible to generate an appropriate predictive model.

[0057] On the other hand, as shown in Figure 5B, in the method of this embodiment, as explained above, the integral and cumulative values ​​of power consumption, which are the target variable to be predicted, are also treated as state variables. Therefore, compared to the conventional method shown in Figure 5A, it is possible to generate a prediction model with higher accuracy.

[0058] While the above explanation uses air conditioners as a specific example, the technology demonstrated in this example is not limited to specific applications and can be applied to a variety of analytical targets (plant equipment, industrial equipment, medical equipment, etc.). [Examples]

[0059] A simulation apparatus according to a second embodiment of the present invention will be described below with reference to Figures 6 and 7. Hereinafter, the same reference numerals are used for components identical to those in Embodiment 1, and their descriptions will be omitted. The simulation apparatus 200 in Embodiment 2 has a function to modify the coefficients of the generated prediction model.

[0060] Figure 6 shows a configuration for acquiring measurement data in real time and performing simulations, compared to Example 1. The configuration adds a coefficient correction unit 8 to the simulation device 100 shown in Figure 1, and the training data generated from the training data generation unit 4 is transmitted to the coefficient correction unit 8.

[0061] Figure 7 is a flowchart illustrating the flow from the generation of training data to the output of simulation results in the training data generation unit 4, model generation unit 5, simulation unit 6, output unit 7, and coefficient correction unit 8 described in Figure 6. The differences from the processing in Example 1 shown in Figure 2 will be explained.

[0062] First, in step S200, the input unit 3 inputs to the simulation device 100 the period for updating the coefficients of the generated prediction model (number of simulation steps) and the period for the training data to be generated in step S210 (described later), although the start and end points change during the simulation. The process from input to step S100 until the prediction model is incorporated into the simulation unit is the same as in Figure 2.

[0063] For example, the model generation unit 5 receives the time step Δt of the prediction model input in step S20 from the input unit 3 as the period for updating the coefficients of the polynomial generated in step S70. mThe number of steps that determines when the coefficients mentioned above will be updated is entered. For example, if the model generation unit 5 receives input from the input unit 3 as the period for the training data to be generated in S210, and in Example 1 the period from 0:00:00 to 22:00:00 on a certain day is set as the predetermined simulation time, then in Example 2 the period from 0:00:00 to 24:00:00 on the same day is entered as the predetermined simulation time. In other words, in Example 2 the period is set to be two hours further in the future than in Example 1.

[0064] Once input is completed in steps S10, S20, and S200, the processing from step S30 to step S100 is performed, as in Example 1.

[0065] Then, in step S210, the training data generation unit 4 performs the training data generation processing procedure shown in Figure 3 again. However, when extracting measurement data in step S130, the period entered in step S200 is used. It is important to note that the period set here depends on the characteristics of the data and the regression method used in the coefficient correction unit 8. For example, when performing a real-time simulation, the end point of the period is set to the latest time acquired by the data acquisition unit 1 (for example, 24:00:00 on a certain day), and the start point is determined from the time obtained from the period value of the training data generated in S210 (for example, the predetermined simulation time) and the end point time (for example, 24:00:00 on a certain day) (for example, 0:00:00 on a certain day). This allows the predictive model to be updated to reflect the latest state of the modeling target.

[0066] From step S220 onwards, the processing is carried out by the coefficient correction unit 8. In step S220, the coefficient correction unit 8 reads the training data generated in step S210.

[0067] In step S230, the coefficient correction unit 8 calculates coefficients to update the differential equation of the generated prediction model. Since the differential equation has already been obtained as a polynomial in step S70, the coefficient correction unit 8 calculates the coefficients using regularization methods such as Ridge regression, extended Kalman filters, or ensemble Kalman filters.

[0068] In step S240, the coefficient correction unit 8 updates the coefficients of the differential equation incorporated into the simulator using the coefficients obtained in step S230. This completes the processing procedure in the coefficient correction unit 8.

[0069] In step S250, the simulation unit 6 executes the simulation for the period (number of steps) entered in step S200. After that, in step S260, the simulation unit 6 determines whether or not to terminate the simulation.

[0070] If the simulation is completed within the scheduled time (Step S260; Yes), proceed to Step S120. If the simulation is to continue (Step S260; No), return to Step S210 and generate training data again to correct the model coefficients. By repeating the above steps, the predictive model incorporated into the simulator will continue to be updated until the end of the simulation.

[0071] This embodiment enables simulation while updating the predictive model in real time using measurement data, and can be applied to cyber-physical systems and real-time control of equipment where 1D simulation is expected to be used.

[0072] As described above for each embodiment, in the simulation device 100 which generates a differential equation model for calculating the target variable based on training data and predicts the target variable using the differential equation model, the target variable (e.g., power consumption) and a plurality of first state variables (e.g., temperature and humidity) are determined from the time series range data used as training data (e.g., training data from time series measurement data in which the start and end points of the time range are specified, step S130), and the data that becomes the target variable is subjected to a time step within the time series range. The system comprises: a training data generation unit (e.g., training data generation unit 4) that generates training data including the first and second state variables, with the value obtained by integrating or accumulating each time being used as a second state variable (e.g., power consumption); a model generation unit (e.g., model generation unit 5) that generates the differential equation model by sparse identification based on the training data; a simulation unit (e.g., simulation unit 6) that performs a simulation to predict the target variable using the differential equation model; and an output unit (e.g., output unit 7) that outputs the differential equation model and / or the simulation results. With this configuration, even when the variable to be predicted has discontinuous data, an appropriate differential equation model can be obtained, enabling highly accurate simulations.

[0073] Furthermore, as explained using S10, S20, etc. in Figure 2, the training data generation unit, before generating the training data, takes a first time step (for example, the simulation time step Δt) that indicates the time interval of the data acquired as the target of the simulation. s ) and a second time step (for example, the time step Δt of the prediction model) that indicates the time interval of the training data used as the prediction model in the above simulation. m ) and are input, and in the above input, the second time step and the first time step are set to the same time unit, and the training data is generated based on the second time step. This makes it possible to obtain a model with the time resolution required for the simulation.

[0074] Furthermore, as explained in S130 of Figure 3, the training data generation unit generates and incorporates multiple differential equation models when performing the simulation, setting the second time step to be the same for all of the differential equation models and generating the training data. This eliminates the time step discrepancies between prediction models, making it easier to adjust the simulation settings.

[0075] Furthermore, as explained using S200, S210-S250, etc. in Figure 7, the system includes a coefficient correction unit (for example, coefficient correction unit 8) for calculating and correcting the coefficients of the polynomial differential equation model generated by the model generation unit. The system repeats the generation of training data in the training data generation unit, the calculation and correction of the coefficients of the differential equation model in the coefficient correction unit, and the execution of simulations in the simulation unit. This allows the system to perform simulations while updating the prediction model in real time.

[0076] Furthermore, as explained in Mathematics 1-3, the model generation unit increases the regularization parameter of the differential equation model in the sparse identification process so that the second state variable is not included. This makes it possible to generate a predictive model that does not include the second state variable, even when explanatory variables that cannot be considered when using the predictive model are included, while adjusting the regularization parameter.

[0077] The present invention is not limited to the embodiments described above, and in the implementation stage, the components can be modified and implemented without departing from the gist of the invention, or the multiple components disclosed in the embodiments can be appropriately combined. [Explanation of Symbols]

[0078] 1...Data acquisition unit 2...Storage section 3...Input section 4…Training data generation unit 5…Model generation unit 6…Simulation Department 7…Output section 8...Coefficient correction section

Claims

1. A simulation device that generates a differential equation model for calculating a target variable based on training data, and predicts the target variable using the differential equation model, A training data generation unit that determines the objective variable and a plurality of first state variables from the time series range data used as training data, and uses the value obtained by integrating or summing the data that will be the objective variable at each time step within the time series range as a second state variable, and generates the training data including the first state variable and the second state variable, A model generation unit generates the differential equation model by sparse identification based on the aforementioned training data, A simulation unit that performs a simulation to predict the objective variable using the differential equation model, An output unit that outputs the differential equation model and / or the simulation results, A simulation device characterized by having the following features.

2. A simulation apparatus according to claim 1, The time-series data used as the training data is operational data of the equipment, including discontinuous data. A simulation device characterized by the following features.

3. A simulation apparatus according to claim 1, The aforementioned training data generation unit, Before generating the training data, a first time step indicating the time interval of the data acquired as the target of the simulation and a second time step indicating the time interval of the learning data to be used as a predictive model in the simulation are input. In the input, the second time step and the first time step are given the same time unit, and the training data is generated based on the second time step. A simulation device characterized by the following features.

4. A simulation apparatus according to claim 3, The aforementioned training data generation unit, When generating and incorporating multiple differential equation models during the aforementioned simulation, the second time step is set to be the same for all of the multiple differential equation models, and the training data is generated. A simulation device characterized by the following features.

5. A simulation apparatus according to claim 1, The model generation unit has a coefficient correction unit for calculating and correcting the coefficients of the polynomial differential equation model generated by the model generation unit, The generation of the training data in the training data generation unit, The coefficient correction unit calculates and corrects the coefficients of the differential equation model, The execution of the simulation in the aforementioned simulation unit, Repeat A simulation device characterized by the following features.

6. A simulation apparatus according to claim 1, The model generation unit increases the regularization parameter of the differential equation model so that the second state variable is not included in the sparse identification. A simulation device characterized by the following features.

7. A simulation method performed by a simulation device that generates a differential equation model for calculating an objective variable based on training data and predicts the objective variable using the differential equation model, The aforementioned simulation device is From the time series data used as training data, the objective variable and a plurality of first state variables are determined, and the value obtained by integrating or summing the data that will be the objective variable at each time step within the time series range is used as the second state variable, and the training data including the first state variable and the second state variable is generated. Based on the aforementioned training data, the differential equation model is generated by sparse identification. Using the differential equation model, a simulation is performed to predict the target variable. Outputs the differential equation model and / or the simulation results. A simulation method characterized by the following:

8. A simulation method according to claim 7, The time-series data used as the training data is operational data of the equipment, including discontinuous data. A simulation method characterized by the following:

9. A simulation method according to claim 7, Before generating the training data, a first time step indicating the time interval of the data acquired as the target of the simulation and a second time step indicating the time interval of the learning data to be used as a predictive model in the simulation are input. In the input, the second time step and the first time step are given the same time unit, and the training data is generated based on the second time step. A simulation method characterized by the following:

10. A simulation method according to claim 9, When generating and incorporating multiple differential equation models during the aforementioned simulation, the second time step is set to be the same for all of the multiple differential equation models, and the training data is generated. A simulation method characterized by the following:

11. A simulation method according to claim 7, The coefficients of the generated polynomial form of the differential equation model are calculated and corrected. The generation of the aforementioned training data, Calculation and correction of the coefficients of the differential equation model, Running the simulation, Repeat A simulation method characterized by the following:

12. A simulation method according to claim 7, In the sparse identification described above, the regularization parameter of the differential equation model is increased so that the second state variable is not included. A simulation method characterized by the following:

Images