Simulation device, simulation method, and simulation program

Abstract

To provide a simulation device that is able to prevent deterioration of the accuracy of an arithmetic result without prolonging a time required to obtain the arithmetic result even when a sampling time for each partial arithmetic model is different.SOLUTION: A simulation device includes: a prediction unit that simulates an arithmetic model including a learning model obtained by degenerating a first partial model and a second partial model at a different sampling time and that predicts an output value at each of an (n) number of times clocked at predetermined time intervals after a certain time, by using a learning model in which a correspondence relationship between an input value of the first partial model at the certain time and an output value of the first partial model at each of the (n) number of times clocked at the predetermined time intervals after the certain time is learned in advance; and an acquisition unit that acquires a predicted number (n+1) of output values and extracts, as an input value of the second partial model, an output value at a time corresponding to a sampling time of the second partial model.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present invention relates to a simulation device, a simulation method, and a simulation program, and more particularly to a simulation device, a simulation method, and a simulation program for performing a simulation by coupled analysis of a computation model including a plurality of partial computation models having different sampling times for executing computation processing. [Background technology]

[0002] Conventionally, in the process of designing and developing products in various fields, computer simulations have been used to confirm and evaluate the performance of the products. In product simulations, a computer processes a calculation model that is expressed as a mathematical formula that imitates the mechanical, physical, and electrical configurations of an actual product, and the performance of the product is confirmed and evaluated based on the calculation results.

[0003] A computation model of a large-scale system or the like is composed of multiple partial computation models. Simulation of such a system is performed by coupled analysis of multiple partial computation models. If the sampling time differs for each computation process of multiple partial computation models, the execution time of the computation process executed for each partial computation model will differ, resulting in a difference in the acquisition time of the computation results for each partial computation model, which causes an error in the simulation by coupled analysis. Even if the error in one computation process of a computation model is small, the error accumulates due to the computation process being repeatedly executed, and cannot be ignored, which may cause a deterioration in the accuracy of the computation results of the simulation.

[0004] Therefore, for example, a technique disclosed in Patent Document 1 has been proposed. The technique disclosed in Patent Document 1 relates to a simulator system including a control device and a model calculation device that simulates the behavior of a control target model by giving a control command output from the control device to the control target model. The simulator system disclosed in Patent Document 1 resolves the decrease in simulation accuracy caused by the difference between the sampling time of the control device and the sampling time of the model calculation device by making the sampling period of the model calculation device variable. That is, according to the technique disclosed in Patent Document 1, the sampling period of the control device is fixed, and by shortening the sampling period of the model calculation device, the sampling time of the control device and the sampling time of the model calculation device can be made to coincide with each other, thereby suppressing the decrease in simulation accuracy of the simulator system. However, the technology disclosed in Patent Document 1 has a drawback in that shortening the sampling period of the model calculation device increases the number of calculations performed by the model calculation device, thereby lengthening the time required to obtain the calculation results of the simulation of the simulator system. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2014-63391 A Summary of the Invention [Problem to be solved by the invention]

[0006] Therefore, the present invention aims to provide a simulation device, a simulation method, and a simulation program that can suppress a decrease in accuracy of the simulation calculation results without lengthening the time required to obtain the simulation calculation results, even when a calculation model representing a target system to be simulated includes multiple partial calculation models and the sampling times of the calculation processes for each partial calculation model are different. [Means for solving the problem]

[0007] That is, the simulation device according to the first aspect is a simulation device that performs coupled analysis of a computation model consisting of a plurality of partial computation models including a learning model obtained by degenerating a first partial model and a second partial model having a sampling time different from the sampling time of the first partial model, and simulates a target system represented by the computation model, and performs coupled analysis of an input value A(t) of the first partial model at a certain time t, an output value B(t) of the first partial model at the time t, and a first partial model for n (n is an integer of 1 or more) times (t+a×dt) (a includes an integer from 1 to n) that are intervals of a predetermined time dt after the time t. The present invention is characterized in that it comprises a prediction unit that predicts the output value B(t) at time t and the output values ​​B(t+a×dt) for n times (t+a×dt) after time t and spaced every predetermined time dt, based on the input value A(t) at time t, using a learning model that has previously learned the correspondence between the output value B(t) and the output value B(t+a×dt) of the second partial model, and an acquisition unit that acquires the n output values ​​B(t+a×dt) predicted by the prediction unit and extracts, from the n output values ​​B(t+a×dt), the output value B(t+a×dt) at the time (t+a×dt) corresponding to the sampling time of the second partial model as an input value of the second partial model.

[0008] A simulation method according to a second aspect is a simulation method for simulating a target system represented by a computation model by coupling analysis of a computation model consisting of a plurality of partial computation models including a learning model obtained by degenerating a first partial model and a second partial model having a sampling time different from the sampling time of the first partial model, in which a computer performs coupling analysis of an input value A(t) of the first partial model at a certain time t, an output value B(t) of the first partial model at the time t, and n (n is an integer equal to or greater than 1) times (t+a×dt) (a includes an integer from 1 to n) of the first partial model at predetermined intervals dt after the time t. The method is characterized by carrying out a prediction step of predicting the output value B(t) at time t and the output values ​​B(t+a×dt) for n times (t+a×dt) after time t, which are intervals of a predetermined time dt, based on the input value A(t) at time t, using a learning model that has previously learned the correspondence between the input value A(t) and the output value B(t) of the second partial model; and an acquisition step of acquiring the n output values ​​B(t+a×dt) predicted in the prediction step, and extracting the output value B(t+a×dt) at the time (t+a×dt) corresponding to the sampling time of the second partial model from the n output values ​​B(t+a×dt) as an input value for the second partial model.

[0009] A simulation program according to a third aspect is a simulation program for simulating a target system represented by a computation model by coupling analysis of a computation model consisting of a plurality of partial computation models including a learning model obtained by degenerating a first partial model and a second partial model having a sampling time different from the sampling time of the first partial model, the simulation program being configured to cause a computer to input an input value A(t) of the first partial model at a certain time t, an output value B(t) of the first partial model at the time t, and n (n is an integer equal to or greater than 1) time (t+a×dt) (a includes an integer from 1 to n) values ​​of the input value A(t) of the first partial model at a certain time t, an output value B(t) of the first partial model at the time t, and n (n is an integer equal to or greater than 1) values ​​of the output value B(t) of the first partial model at a certain time dt after the time t. The system is characterized by having a prediction function that predicts the output value B(t) at time t and the output values ​​B(t+a×dt) for n times (t+a×dt) after time t, which are intervals of a predetermined time dt, based on the input value A(t) at time t, using a learning model that has previously learned the correspondence between the output value B(t) of the first partial model and the input value A(t), and an acquisition function that acquires the n output values ​​B(t+a×dt) predicted by the prediction function and extracts, from the n output values ​​B(t+a×dt), the output value B(t+a×dt) at the time (t+a×dt) corresponding to the sampling time of the second partial model as an input value for the second partial model. Effect of the Invention

[0010] The simulation device etc. according to the present invention is a simulation device that simulates a target system represented by a computation model by performing coupled analysis on a computation model consisting of a plurality of partial computation models including a learning model obtained by degenerating a first partial model and a second partial model having a sampling time different from the sampling time of the first partial model, and that uses a learning model that has previously learned a correspondence relationship between an input value A(t) of the first partial model at a certain time t, an output value B(t) of the first partial model at time t, and an output value B(t) of the first partial model for each of n (n is an integer of 1 or more) times (t+a×dt) (a includes an integer from 1 to n) after time t and intervals of a predetermined time dt, and calculates an output value B(t) at time t and an output value B(t+a×dt) of the first partial model for each of n (n is an integer of 1 or more) times (t+a×dt) (a includes an integer from 1 to n) after time t and intervals of a predetermined time dt. The simulation system is characterized by comprising a prediction unit that predicts an output value B(t+a×dt) for each of n times (t+a×dt) that are after time t and are spaced every predetermined time dt, and an acquisition unit that acquires the n output values ​​B(t+a×dt) predicted by the prediction unit and extracts, from the n output values ​​B(t+a×dt), the output value B(t+a×dt) at the time (t+a×dt) that corresponds to the sampling time of the second partial model as an input value for the second partial model. Therefore, even if the computation model representing the target system to be simulated includes a plurality of partial computation models and the sampling times of the computation processes for each partial computation model are different, it is possible to suppress a decrease in accuracy of the computation results of the simulation without lengthening the time until the computation results of the simulation are obtained. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 is a block diagram for explaining an overview of a simulation device according to this embodiment. [Diagram 2] FIG. 2 is a block diagram for explaining an example of a hardware configuration of the simulation device according to this embodiment. [Diagram 3] FIG. 3 is a block diagram for explaining an example of the functional configuration of the simulation device according to the present embodiment. [Figure 4]FIG. 4 is a graph for explaining the operation of the ROM in the simulation device according to this embodiment. [Diagram 5] FIG. 5 is a timing chart for explaining sampling times of the calculation process for each partial calculation model of the simulation device according to this embodiment. [Figure 6] FIG. 6 is a timing chart for explaining the sampling time of the calculation process for each partial calculation model of the simulation device according to this embodiment. [Figure 7] FIG. 7 is a timing chart for explaining sampling times of the calculation process for each partial calculation model of the simulation device according to this embodiment. [Figure 8] FIG. 8 is a block diagram for explaining an example of the configuration of a vehicle system that is a target system according to this embodiment. [Figure 9] FIG. 9 is a block diagram for explaining an example of the configuration of a computation model according to this embodiment. [Figure 10] FIG. 10 is a flowchart of the calculation process for simulating a vehicle system in the simulation device of the first example according to this embodiment. [Figure 11] FIG. 11 is a flowchart of the calculation process for simulating a vehicle system in the simulation device of the second example according to this embodiment. [Figure 12] FIG. 12 is a flowchart of the simulation program according to this embodiment. [Figure 13] FIG. 13 is a block diagram for explaining an outline of a simulation device according to the prior art. [Figure 14] FIG. 14 is a timing chart for explaining sampling times of calculation processing for each partial calculation model in a simulation device according to the prior art. [Figure 15] FIG. 15 is a flowchart of the calculation process for simulating a vehicle system in a simulator according to the prior art. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] An embodiment of a simulation device 10 according to the present disclosure will now be described with reference to FIGS.

[0013] (Overview of the simulation device 10 according to the present disclosure) With reference to Fig. 1, an overview of the simulation device 10 will be described while comparing it with a simulation device 10A according to the prior art (see Fig. 13). Fig. 1 is a block diagram for explaining the overview of the simulation device 10. Fig. 13 is a block diagram for explaining the overview of the simulation device 10A according to the prior art, and Fig. 14 is a timing chart for explaining the sampling time of the calculation process for each partial calculation model of the simulation device 10A.

[0014] The simulation devices 10 and 10A are computers that execute arithmetic processing of a calculation model that represents a target system to be simulated. In order to easily explain the outline of the simulation device 10, the calculation model that is the target of the calculation processing of the simulation device 10 is assumed to include three partial calculation models, a model A, a ROM, and a second partial calculation model. The calculation model that is the target of the calculation processing of the simulation device 10A according to the conventional technology is assumed to include three partial calculation models, a model A, a first partial calculation model, and a second partial calculation model.

[0015] In the simulation device 10, the input side of the ROM is connected to the output side of model A, and the input side of the second partial model is connected to the output side of the ROM. In the simulation device 10A according to the conventional technology, the input side of the first partial model is connected to the output side of model A, and the input side of the second partial model is connected to the output side of the first partial model. The ROM is a learning model obtained by degenerating the first partial calculation model of the simulation device 10A according to the conventional technique. Furthermore, model A is not limited to a partial calculation model, but may be a device with a function of outputting an input value of a ROM, such as a signal generator that outputs a signal based on a simple predetermined algorithm, or a buffer circuit that receives and holds a signal from the outside and outputs the signal based on a simple predetermined algorithm, and does not have to perform calculation processing.

[0016] A partial computation model is a constituent unit of a computation model obtained by dividing a computation model, for example, by function or by control object. A computational model is a model that expresses the characteristics and properties of the target system being simulated, or the characteristics and properties common to multiple target systems of the same type, mathematically or in machine language so that the computer can easily perform computations on them. Specifically, a computational model refers to a computational procedure, algorithm, etc., that expresses the behavior and characteristics of a target system using variables and numerical values, and is defined by the computational procedure, algorithm, etc., for generating an expected output from a specific input. For example, a computational model is a mathematical expression of the physical characteristics of the behavior of a target system of a simulation over time.

[0017] Between the ROM and the second partial model, there is an acquisition unit 27 described later. The acquisition unit 27 acquires an output value B of the ROM, extracts an output value B corresponding to a sampling time of the second partial model, and acquires an input value B′(t now )

[0018] 13, the computation model of the simulation device 10A according to the conventional technology includes a partial computation model, a model A, a first partial model, and a second partial model, which affect each other, and therefore the computation models are subjected to coupled analysis. Model A performs computation processing with a step size A, the first partial model performs computation processing with a step size B, and the second partial model performs computation processing with a step size C. At least, the step size B of the first partial model is different from the step size C of the second partial model. The step size is the time required for the calculation processing in the simulation divided into several steps, and is also called a sampling period in this disclosure. The sampling time is the time at which the calculation processing of the simulation is performed every sampling period.

[0019] In the computation model of the simulation device 10A according to the prior art, the input value of the first partial model is expressed by a function A(t) with the simulation time t as a variable, and the output value of the first partial model is expressed by a function B(t) with the simulation time t as a variable. The output value B(t) of the first partial model becomes the input value of the second partial model. The simulation time is the time elapsed in the simulation being executed.

[0020] FIG. 14 shows that the step size B of the first partial model and the step size C of the second partial model are different in the simulation device 10A, and shows a state in which the execution time of the calculation process of the first partial model, i.e., the sampling times X(t=i) and X(t=i+1), do not match the sampling times T'(t=j-1), T'(t=j), and T'(t=j+1) of the calculation process of the second partial model. From this state, in order to match the sampling time X(t=j) of the calculation process of the first partial model with the sampling time T'(t=j) of the calculation process of the second partial model at the time t=j, the simulation device 10A makes the step size B of the first calculation model changeable. In order to match the sampling time X(t=i+1) of the calculation process of the first partial model with T'(t=j), the step size (sampling period) of the first partial model is shortened. As a result, the number of calculations required for the calculation processing of the first partial model increases, which causes a problem that the time required to complete the calculation processing of the simulation of the calculation model in the simulation device 10A according to the conventional technology becomes longer.

[0021] Therefore, in the simulation device 10 according to this embodiment, as shown in FIG. 1, a ROM (Reduced Order Model) is placed in place of the first partial model, and an acquisition unit 27 described below is interposed between the ROM and the second partial model. The simulation device 10 according to this embodiment performs coupled analysis of a computation model consisting of multiple partial computation models including a learning model, which is a reduced-dimensional model obtained by degenerating the first partial model, and a second partial model having a sampling time different from the sampling time of the first partial model, to simulate a target system represented by the computation model.

[0022] ROM is a technique for reducing the dimensions of a model (degeneration, reduction) while maintaining the essential behavior of the target system, significantly reducing analysis time and data volume, and is also called a reduced model or reduced-dimensional model. Dimension refers to the number of features and properties contained in the behavior characteristics of variables such as time changes of the target system. Specifically, a ROM refers to a learning model that learns the physical, mechanical, and electrical characteristics and properties of a computational model representing a target system (in this case, the first partial model, which is a partial computational model) through machine learning. For example, in vehicle development, the computational model representing the target system is a detailed physical model, so by replacing a part or all of it with a degenerate ROM, the computational model becomes lighter, making it possible to perform high-speed simulation of the target system. In addition, the calculation model includes multiple partial calculation models, which are coupled and affect each other, making the calculation model complex. However, by replacing a part or all of the calculation model with a degenerated ROM, the target system can be made lighter and faster to simulate. In addition, the step size of the ROM is the same as the step size B of the first partial model, since the ROM is a learning model obtained by degenerating the first partial model.

[0023] As shown in Figure 1, the ROM receives the current input value A(tnow ) and obtain the corresponding input value A(t now ) based on the output value B(t now ) and the current time t at every predetermined time dt now The acquisition unit 27 outputs n output values ​​B(t+a×dt) (a includes an integer from 1 to n) that are after n×predetermined time dt (n is an integer equal to or greater than 1) and that correspond to the sampling time of the second partial model from among the n output values ​​B(t+a×dt). now ) and use it as an input value for the second partial model. As a result, even if the step size B of the first partial model and the step size C of the second partial model are different, by replacing the first partial model with a ROM, the second partial model can obtain the output value B of the first partial model corresponding to the sampling time of the second partial model as an input value, and it is possible to suppress a decrease in accuracy of the simulation calculation results caused by a difference in sampling time between the first partial model and the second partial model. Furthermore, since the ROM calculates n output values ​​B(t+a×dt) during one sampling period, the time until the calculation results of the simulation are obtained is not extended.

[0024] The current time may refer to a point on the time axis of the simulation time of the computation process under consideration. n is an integer of 1 or more that is specified by the user or in the simulation program described later. For example, if n=5, the ROM stores the input value A(t now ), it predicts and outputs the output values ​​B(t), B(t+dt), B(t+2dt), B(t+3dt), B(t+4dt), and B(t+5dt) of model B. The predetermined time dt is the unit of division pieces used to divide the ROM step size B, and is set to a value equal to or less than the step size B divided by n, and is specified by the user or in the simulation program.

[0025] (Hardware configuration of simulation device 10) An example of the hardware configuration of the simulation device 10 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a block diagram for explaining an example of the hardware configuration of the simulation device 10.

[0026] The simulation device 10 includes a communication interface 10a, a non-volatile memory 10b, a random access memory (RAM) 10C, a recording unit 10d, a central processing unit (CPU) 10e, and an input / output interface 10f.

[0027] The communication interface 10a has a function of transmitting and receiving data to and from other devices via the communication network 11. The recording unit 10d can be used as a recording device, and records a simulation program (to be described later), an OS (Operating System) required for the operation of the simulation device 10, various applications, and various data used by the applications.

[0028] The simulation device 10 stores the simulation program in a recording unit 10d, and loads the simulation program into a main memory constituted by a RAM 10c etc. The CPU 10e accesses the main memory into which the simulation program has been loaded, and executes the simulation program. The input / output interface 10f transmits and receives data to and from devices external to the simulation device 10. The external devices are an input device 10g and an output device 10h that input and output data to and from the simulation device 10. The input device 10g is a keyboard, a mouse, etc., and the output device 10h is an external monitor, a printer, an external speaker, etc.

[0029] Next, the functional configuration of the simulation device 10 will be described with reference to Fig. 3. Fig. 3 is a block diagram for explaining an example of the functional configuration of the simulation device 10 according to this embodiment. The simulation device 10 has functional units, such as a prediction unit 23 and an acquisition unit 27, in a CPU 10e by executing a simulation program described later.

[0030] The prediction unit 23 predicts the output value B(t) at a certain time t and the output value B(t+a×dt) for each of n times (t+a×dt) (a includes an integer from 1 to n) after the time t, where n is an integer equal to or greater than 1, and is spaced every predetermined time dt, based on the input value A(t) at that time t, using a learning model that has previously learned the correspondence between the input value A(t) of the first partial model at that time t, the output value B(t) of the first partial model at that time t, and the output value B(t+a×dt) for each of n times (t+a×dt) after the time t, where n is an integer equal to or greater than 1, and is spaced every predetermined time dt.

[0031] The learning model is a reduced order model (ROM) that is a degenerate version of the first partial model. The learning model may be a reduced-dimensional model, and may be generated through machine learning using the calculation results of a simulation obtained by previously executing a simulation on the first partial model as learning data.

[0032] The learning model may be a recurrent neural network (hereinafter, abbreviated as RNN) model. As a type of learning model, a deep learning model of a neural network suitable for learning complex patterns is preferable. Among deep learning models, an RNN model is a learning model in which a neural network is extended to handle time series data, and is preferable for the learning model of this embodiment. In addition, a long short-term memory (hereinafter, referred to as LSTM) network, which is a type of RNN model, is suitable for learning time series data and predicting time series data, and is preferable for the learning model of this embodiment.

[0033] The learning model may be generated through machine learning using gradient boosting, using the calculation results of the simulation of the first partial model as learning data. Boosting is a type of ensemble learning, in which multiple weak learners, such as decision trees, are prepared and connected in series. The multiple weak learners are trained in sequence, and weights are added so that each learner takes priority in correctly classifying data that was misclassified by the previous learner. Gradient boosting is a type of boosting, and is a method that uses gradient descent to minimize an objective function that summarizes the difference between the output and the predicted value of each data. A learner refers to a learning model, and a weak learner is a learner that has low accuracy when used alone.

[0034] The learning model may be generated through machine learning using a random forest, using the calculation results of a simulation of the first partial model as learning data. Random forest is a type of ensemble learning similar to boosting, and is a method for determining the final prediction value by majority vote of multiple decision trees created by bagging, and is a method for using multiple decision trees linked in parallel. Bagging is a technique in which multiple learning devices are trained using randomly selected subsets from a training data set, and these prediction results are combined. A subset refers to a collection of elements that form part of a set, and is created by selecting some (or all) elements from the original set. In this embodiment, a subset of training data using part of the calculation results of the simulation of the first partial model is used.

[0035] The output value B(t+a×dt) (where a is an integer ranging from 1 to n) of the ROM will be described with reference to Fig. 4. Fig. 4 is a graph for explaining the function of the ROM of the simulation device 10. B(t) at time t of the simulation time is a value predicted by the ROM regarding the output value of the first partial model when A(t) is given to the first partial model as an input value. The ROM predicts the output value B(t) of the first partial model at times (t+dt), (t+2dt), (t+3dt), (t+4dt), . . . , (t+n×dt) after time t, defined by time (t+a×dt) (where a is an integer from 1 to n), where n is an integer equal to or greater than 1. "Now" in Fig. 4 refers to the sampling time of the second partial model. The acquisition unit 27 described later extracts B(t) at the time (t+a×dt) (a includes an integer from 1 to n) corresponding to this "now" and sets it as the input value of the second partial model as B'(t). For example, in Fig. 4, the time (t+a×dt) (a includes an integer from 1 to n) corresponding to the "now" is the time (t+3dt), and the acquisition unit 27 sets B(t+3dt) as the input value B'(t) of the second partial model.

[0036] The acquisition unit 27 acquires the n output values ​​B(t+a×dt) predicted by the prediction unit 23, and extracts from the n output values ​​B(t+a×dt) the output value B(t+a×dt) at the time (t+a×dt) corresponding to the sampling time T'(t=j) (see Figure 5) of the second partial model as an input value of the second partial model.

[0037] The acquiring unit 27 will be described with reference to Fig. 5. Fig. 5 is a timing chart for explaining the sampling time of the calculation process for each partial calculation model of the simulation device 10. As shown in Fig. 5, the step size B of the ROM is different from the step size C of the second partial model in the simulation device 10. Fig. 5 shows a state in which the execution time points of the ROM calculation process, i.e., the sampling times X(t=i) and X(t=i+1), do not match with the sampling times T'(t=j-1), T'(t=j), and T'(t=j+1) of the calculation process of the second partial calculation model.

[0038] The prediction unit 23 uses a ROM to predict, based on an input value A(t=i) at a sampling time X(t=i), an output value B(t=i) for each time (i+a×dt) (a includes an integer from 1 to n), and B(i+a×dt). Then, the acquisition unit 27 extracts the output value B(i+a×dt) at time (i+a×dt) corresponding to the sampling time T'(t=j) of the second partial model as B'(t=j) from the output values ​​B(i+a×dt) predicted by the prediction unit 23 for each time (i+a×dt), and sets it as the input value of the second partial model.

[0039] Next, the "correspondence" of the time (i+a×dt) corresponding to the sampling time T' (t=j) of the second partial model in the operation of the acquisition unit 27 will be described with reference to Fig. 6 and Fig. 7. Fig. 6 and Fig. 7 are timing charts for explaining the sampling time of the calculation process for each partial calculation model of the simulation device 10.

[0040] As shown in Fig. 6, the acquisition unit 27 may extract the output value B(t+a×dt) at the time (t+a×dt) closest to the sampling time of the second partial model from among the acquired n output values ​​B(t+a×dt) as the input value of the second partial model. Note that Fig. 6 shows an example where n=5.

[0041] As shown in FIG. 6, the prediction unit 23 uses a ROM to predict, based on the input value A(t=i) at the sampling time X(t=i), the output value B(t=i) at sampling time X(i+dt), the output value B(i+2dt) at sampling time X(i+2dt), the output value B(i+3dt) at sampling time X(i+3dt), the output value B(i+4dt) at sampling time X(i+4dt), and the output value B(i+5dt) at sampling time X(i+5dt). Then, the acquiring unit 27 extracts the output value B(i+a×dt) at the time (i+a×dt) closest to the sampling time T′(t=j) of the second partial model as B′(t=j) from the output value B(i+a×dt) predicted by the predicting unit 23 for each time (i+a×dt), and sets it as the input value of the second partial model. In the example shown in FIG. 6, the time (i+3×dt) is closest to the time T′(t=j), and the acquiring unit 27 extracts the output value B(i+3dt) at the time (i+3dt) as B′(t=j), and sets it as the input value of the second partial model.

[0042] As shown in Fig. 7, the acquisition unit 27 may extract output values ​​B at the two closest times (t+a×dt) sandwiching the sampling time of the second partial model from among the acquired n output values ​​B(t+a×dt), and may use the two extracted output values ​​B to interpolate the output value B at the sampling time T'(t=j) of the second partial model, and extract the interpolated output value B as the input value of the second partial model. This is effective when there is no output value B(t+a×dt) at the same time as the sampling time T'(t=j) of the second partial model among the n output values ​​B(t+a×dt) acquired by the acquisition unit 27. Note that Fig. 7 shows an example where n=4.

[0043] As shown in FIG. 7, the prediction unit 23 uses a ROM to predict, based on the input value A(t=i) at the sampling time X(t=i), the output value B(t=i) at sampling time X(i+dt), the output value B(i+2dt) at sampling time X(i+2dt), the output value B(i+3dt) at sampling time X(i+3dt), and the output value B(i+4dt) at sampling time X(i+4dt). Then, the acquisition unit 27 calculates the output value B at sampling time T'(t=j) of the second partial model by interpolation using the output value B(i+a×dt) and the output value B(i+3dt) at sampling time X(i+2dt) and sampling time X(i+3dt), which are the two times (i+a×dt) closest to the sampling time T'(t=j) of the second partial model, from the output values ​​B(i+a×dt) predicted by the prediction unit 23 for each time (i+a×dt), and extracts the interpolated output value as B'(t=j) and sets it as the input value of the second partial model.

[0044] The interpolation method includes linear interpolation, Lagrange interpolation, spline interpolation, and Newtonian interpolation. Linear interpolation is a method of connecting two points with a straight line and performing interpolation using a point on the line at the corresponding sampling time. For example, the acquisition unit 27 interpolates the output value B at the sampling time T'(t=j) using a point specified by a point corresponding to the sampling time T'(t=j) on a line passing through the output value B(i+2dt) and the output value B(i+3dt), extracts the interpolated output value B as the output value B'(t=j), and sets it as an input value of the second partial model.

[0045] Lagrange interpolation is an interpolation method that uses the polynomial Lagrange interpolation formula, and uses a linear function for interpolation between two points. For example, the acquisition unit 27 derives a linear function that passes through the output value B(i+2dt) and the output value B(i+3dt), interpolates the output value B at the sampling time T'(t=j) using a point specified by a point on the linear function that corresponds to the sampling time T'(t=j), extracts the interpolated output value B as the output value B'(t=j), and sets it as an input value of the second partial model.

[0046] Spline interpolation is a method of connecting two points with a spline curve and performing interpolation using a point on the spline curve at the corresponding sampling time. For example, the acquisition unit 27 derives a spline curve that passes through the output value B(i+2dt) and the output value B(i+3dt), interpolates the output value B at the sampling time T'(t=j) using a point specified by a point on the spline curve that corresponds to the sampling time T'(t=j), extracts the interpolated output value B as the output value B'(t=j), and sets it as an input value of the second partial model.

[0047] Newtonian interpolation is an interpolation method using Newtonian interpolation formula. For example, the acquisition unit 27 derives a Newtonian interpolation formula that passes through the output value B(i+2dt) and the output value B(i+3dt), interpolates the output value B at the sampling time T'(t=j) by a point specified by a point corresponding to the sampling time T'(t=j) on the Newtonian interpolation formula, extracts the interpolated output value B as the output value B'(t=j), and sets it as an input value of the second partial model.

[0048] (Regarding vehicle system 30, which is the target system for the simulation) A vehicle system 30 which is a target system of a simulation executed by the simulation device 10 according to this embodiment will be described with reference to Fig. 8. Fig. 8 is a block diagram for explaining an example of the configuration of the vehicle system 30 which is the target system according to this embodiment.

[0049] As shown in FIG. 8, the vehicle system 30 includes an engine 31, an engine controller 32, a shaft 33, a TM (transmission) 34, a transmission controller 35, a shaft 36, a differential gear (DIF) 37, a shaft 38, a shaft 39, a tire 40, and a tire 41.

[0050] The engine 31 is a power source that provides power to drive the vehicle.

[0051] The engine controller 32 is a control device having a function of controlling the engine 31. The engine controller 32 executes torque control of the engine 31, for example.

[0052] The shaft 33 is an axis that connects the engine 31 and the TM (transmission) 34. The shaft 33 receives engine torque from the engine 31 and a transmission load from the TM 34, and rotates in the direction of the combined force.

[0053] The TM 34 is made up of gears and shafts, and has the function of converting the torque, rotation speed, and rotation direction of the power received from the engine 31 via the shaft 33 and transmitting it to the vehicle side (differential 37 side).

[0054] The transmission controller 35 is a control device having a function of controlling the torque and rotation speed of the power of the TM 34. The transmission controller 35 controls, for example, a conversion coefficient that converts the torque from the engine 31.

[0055] The shaft 33 is an axis that connects the TM 34 and the differential 37. The differential 37 receives an output shaft torque from the TM 34 and a differential load from the differential 37, and rotates in the direction of the combined force.

[0056] The differential 37 is a mechanism that controls the rotation speed of the left and right tires 40 and 41 . The differential 37 is formally called a differential gear. The differential 37 converts the torque transmitted from the TM 34 via the shaft 33 into the rotational speed of each of the left and right tires 40 and 41 and transmits it.

[0057] The shaft 38 is an axis that connects the differential 37 and the tires 40. The shaft 38 receives the torque from the differential 37 and the load from the tires 40, and rotates in the direction of the combined force.

[0058] The tire 40 rotates as the shaft 38 rotates.

[0059] The shaft 39 is an axis that connects the differential 37 and the tires 41. The shaft 39 receives the torque from the differential 37 and the load from the tires 41, and rotates in the direction of the combined force.

[0060] The tire 41 rotates in conjunction with the rotation of the shaft 39 .

[0061] The simulation device 10 executes a simulation of the operation of a vehicle system 30 as shown in FIG.

[0062] (Regarding the computation model when the vehicle system 30 is the target system) A computation model in the case where the vehicle system 30 is the target system for simulation will be described with reference to Fig. 9. Fig. 9 is a block diagram for explaining an example of the configuration of the computation model according to this embodiment.

[0063] The target system of the simulation device 10 is a vehicle system 30, and the partial calculation models that constitute the calculation model used for the calculation processing of the simulation of the vehicle system 30 include an engine plant model that is included in the vehicle system 30 and is assigned the calculation processing of the simulation of the operation of the engine 31 that is the control target of the engine control, a transmission plant model that is included in the vehicle system 30 and is assigned the calculation processing of the simulation of the operation of the TM 34 that is the control target of the transmission control, and a vehicle motion plant model that is included in the vehicle system 30 and is assigned the calculation processing of the simulation of the motion of the vehicle connected to the TM 34.

[0064] The engine plant model performs calculations with a step size A, the transmission plant model performs calculations with a step size B, and the vehicle motion plant model performs calculations with a step size C. At least step size B is different from step size C, and step size B is larger than step size C. The sampling time of the transmission plant model and the sampling time of the vehicle motion plant model are different.

[0065] The engine plant model has as its input the feedback value of the engine angular velocity [rad / s], which is the output value of the transmission plant model, and has as its output the engine torque [Nm].

[0066] The transmission plant model takes the feedback values ​​of engine torque [Nm], which is the output value of the engine plant model, and output shaft angular velocity [rad / s], which is the output value of the vehicle motion plant model, as input values, and takes the output shaft torque [Nm] and engine angular velocity [rad / s] as output values.

[0067] The vehicle motion plant model has the output shaft torque [Nm], which is the output value of the transmission plant model, as its input value, and the output shaft angular velocity [rad / s] as its output value.

[0068] (First and second embodiments of the simulation device 10) 10 and 11, a first embodiment and a second embodiment in which the target system of the simulation device 10 is a vehicle system 30 will be described in comparison with a simulation device 10A according to the prior art. Fig. 10 is a flowchart of the calculation process for simulating the vehicle system 30 in the simulation device 10 of the first embodiment, and Fig. 11 is a flowchart of the calculation process for simulating the vehicle system 30 in the simulation device 10 of the second embodiment. Fig. 15 is a flowchart of the calculation process for simulating the vehicle system 30 in the simulation device 10A according to the conventional technology. The transmission plant model corresponds to the above-mentioned first partial model, and the vehicle motion plant model corresponds to the above-mentioned second partial model (see FIG. 13).

[0069] (Simulation device 10A according to the prior art) In the engine plant model of the simulation device 10A according to the conventional technology, a simulation of the operation of the engine 31 included in the vehicle system 30 and a simulation of engine control of the engine 31 by the engine controller 32 are performed, and an engine control step S15 and an engine step S16 are provided and executed, as shown in Fig. 15. In the engine control step S15, calculation processing for the simulation of the engine control is executed, and in the engine step S16, calculation processing for the simulation of the operation of the engine 31 is executed.

[0070] Furthermore, in the transmission plant model of simulation device 10A according to the conventional technology, a simulation of the operation of a TM (transmission) 34 included in a vehicle system 30 and a simulation of transmission control of the TM 34 by a transmission controller 35 are performed, and a transmission control step S17 and a transmission step S18 are provided and executed, as shown in Fig. 15. In the transmission control step S17, calculation processing for the simulation of the transmission control is executed, and in the transmission step S18, calculation processing for the simulation of the operation of the TM 34 is executed.

[0071] In addition, in the vehicle motion plant model of the simulation device 10 according to the conventional technology, a simulation of the motion of a vehicle connected to the TM 34 is performed, and a vehicle step S19 is provided and executed as shown in Fig. 15. In the vehicle step S19, calculation processing of the simulation of the motion of a vehicle connected to the TM 34 is executed. The motion of the vehicle connected to the TM 34 refers to the motion of the shaft 36, the differential 37, the shaft 38, the shaft 39, the tire 40, and the tire 41.

[0072] . 15, the delay element 20 connected to the input side of the engine step S16 is an element that delays the input value to the engine step S16 by one step, and is described by a transfer function of a Z function. The delay element 20 represents a delay when the feedback value of the engine angular velocity [rad / s], which is the calculation result of the transmission step S18, is temporarily stored in a buffer memory (not shown) for one step.

[0073] 15, the delay element 21 connected to the input side of the transmission step S18 is an element that delays the input value to the transmission step S18 by one step, and is described by a transfer function of a Z function. The delay element 21 represents a delay when the feedback value of the output shaft angular velocity [rad / s], which is the calculation result of the vehicle step S19, is temporarily stored in a buffer memory (not shown) for one step.

[0074] (Simulation device 10 according to the first embodiment) A simulation device 10 according to a first example of the present embodiment will be described with reference to Figures 1 and 10. In describing the simulation device 10 according to the first example, differences from the simulation device 10A according to the conventional technology will be described, and points in common with the simulation device 10A according to the conventional technology will be denoted by the same reference numerals in Figure 10 as those used in the description of the simulation device 10A according to the conventional technology, and description thereof will be omitted.

[0075] As shown in FIG. 10, the flowchart of the computation process for simulating the vehicle system 30 of the simulation device 10 according to the first embodiment is such that the transmission step S18 of the simulation device 10A according to the conventional technology is replaced with a transmission ROM step S25, and an acquisition step S27 is further connected to the input side of the vehicle step S19. In the simulation device 10 according to the first embodiment, the engine plant model corresponds to model A shown in FIG. 1, the transmission ROM corresponds to the ROM shown in FIG. 1, and the vehicle motion plant model corresponds to the second partial model shown in FIG.

[0076] In the transmission ROM step S25, the prediction unit 23 predicts the output value B(t) at time t and the output value B(t+a×dt) for each of n times (t+a×dt) after time t and each of n times (t+a×dt) after time t and each of n times (t+a×dt) after time t and each of n times dt (n is an integer of 1 or more) (a includes an integer from 1 to n) based on the input value A(t) including the calculation result of the engine plant model at a certain time t and a feedback value of the calculation result of the vehicle motion plant model, using a learning model that has previously learned the correspondence between the input value A(t) of the transmission plant model including the calculation result of the engine plant model at a certain time t and a feedback value of the calculation result of the vehicle motion plant model, and the output value B(t) of the transmission plant model at time t and each of n times (t+a×dt) after time t and each of n times (t+a×dt) after time t and each of n times dt (n is an integer of 1 or more).

[0077] The learning model of the simulation device 10 according to the first embodiment is a transmission ROM. The transmission ROM is a learning model that is learned through machine learning using the computation results of a simulation performed in advance on a transmission plant model as learning data.

[0078] In the acquisition step S27, the acquisition unit 27 acquires the n output values ​​B(t+a×dt) predicted by the prediction unit 23, and extracts from the n output values ​​B(t+a×dt) the output value B at the time (t+a×dt) corresponding to the sampling time T'(t=j) of the vehicle motion plant model as the input value B'(t) of the vehicle motion plant model.

[0079] In the simulation device 10 according to the first embodiment, by replacing the partial model corresponding to the transmission step S18 with a transmission ROM and connecting the acquisition unit 27 to the input side of the vehicle motion plant model, even if a deviation occurs between the sampling time of the transmission plant model and the sampling time of the vehicle motion plant model, it is possible to suppress a decrease in accuracy of the simulation calculation results caused by the deviation. In other words, the simulation device 10 according to the first embodiment can achieve the effect without changing the overall configuration of the transmission plant model.

[0080] (Simulation device 10 according to the second embodiment) A simulation device 10 according to a second example of this embodiment will be described with reference to Figures 1 and 11. In describing the simulation device 10 according to the second example, differences from the simulation device 10A according to the conventional technology will be described, and points in common with the simulation device 10A according to the conventional technology will be denoted by the same reference numerals in Figure 11 as those used in the description of the simulation device 10A according to the conventional technology, and description thereof will be omitted.

[0081] In the simulation device 10 according to the second embodiment, the transmission model included in the simulation device 10A according to the conventional technique is replaced with a transmission ROM, and an acquisition unit 27 is connected to the output side of the vehicle motion plant model. The transmission model is interposed between the transmission plant model and the vehicle motion plant model, obtains the calculation result of the transmission plant model as an input value A(t), and transmits the calculation result to the vehicle motion plant model as an output value B(t). The transmission model corresponds to the first partial model of the simulation apparatus 10A (see FIG. 13). The transmission ROM is a learning model that is trained through machine learning using the results of calculations of a simulation performed in advance on the transmission model as learning data.

[0082] In the simulation device 10 according to the second embodiment, the engine plant model and the transmission plant model correspond to model A shown in FIG. 1, the transmission ROM corresponds to the ROM shown in FIG. 1, and the vehicle motion plant model corresponds to the second partial model shown in FIG. 1. The sampling time of the transmission model and the sampling time of the vehicle motion plant model are different.

[0083] As shown in FIG. 11, a flowchart of the computation process for simulating the vehicle system 30 of the simulation device 10 according to the second embodiment is such that a transmission ROM step S26 is connected to the output side of a transmission step S18 of the simulation device 10A according to the conventional technology, and an acquisition step S27 is further connected to the input side of a vehicle step S19.

[0084] In the transmission ROM step S26, the prediction unit 23 predicts the output value B(t+a×dt) for each of n (n is an integer of 1 or more) times (t+a×dt) (a includes an integer from 1 to n) after time t and intervals of a predetermined time dt (n is an integer of 1 or more), based on the input value A(t) at a certain time t, using a learning model that has previously learned the correspondence between the input value A(t) of the transmission model at a certain time t and the output value B(t+a×dt) for each of n (n is an integer of 1 or more) times (t+a×dt) (a includes an integer from 1 to n) after time t and intervals of a predetermined time dt, and transmits the n output values ​​B(t+a×dt) to the vehicle motion plant model.

[0085] In the acquisition step S27, the acquisition unit 27 acquires the n output values ​​B transmitted by the prediction unit 23, and extracts from the n output values ​​B the output value B at the time (t+a×dt) corresponding to the sampling time T' (t=j) of the vehicle motion plant model as the input value B'(t) of the vehicle motion plant model.

[0086] In the simulation device 10 according to the second embodiment, by connecting a transmission ROM obtained by degenerating a transmission model to the output side of a partial model corresponding to the transmission step S18 and connecting an acquisition unit 27 to the input side of a vehicle motion plant model, even if a deviation occurs between the sampling time of the transmission plant model and the sampling time of the vehicle motion plant model, it is possible to suppress a decrease in accuracy of the simulation calculation results caused by the deviation. In other words, the simulation device 10 according to the second embodiment can achieve the effect without making any changes to the existing transmission plant model.

[0087] (Simulation program and simulation method) Next, a simulation program according to an embodiment of the present disclosure will be described together with a simulation method with reference to Fig. 12. Fig. 12 is a flowchart of the simulation program according to the present embodiment. The simulation method is executed by the CPU 10e of the simulation device 10 based on the simulation program.

[0088] The simulation program includes a prediction step S23 and an acquisition step S27. The simulation program causes the CPU 10e of the simulation device 10 to perform various functions such as a prediction function and an acquisition function. Note that the various functions overlap with the description of the various functional units of the simulation device 10 described above, and therefore detailed description thereof will be omitted.

[0089] The prediction function uses a learning model that has previously learned the correspondence between the input value A(t) of the first partial model at a certain time t, the output value B(t) of the first partial model at time t, and the output value B(t+a×dt) of the first partial model for each of n times (t+a×dt) (a includes an integer from 1 to n) after time t and separated by a predetermined time dt, based on the input value A(t) at a certain time t (S23: prediction step).

[0090] The acquisition function acquires the n output values ​​B(t+a×dt) predicted in the prediction step, and extracts from the n output values ​​B(t+a×dt) the output value B(t+a×dt) at the time (t+a×dt) corresponding to the sampling time of the second partial model as an input value of the second partial model (S27: acquisition step).

[0091] According to the simulation device 10 of the present embodiment described above, even if the step size B of the first partial model is different from the step size C of the second partial model and a deviation occurs between the sampling time of the first partial model and the sampling time of the second partial model, the first partial model is replaced with a learning model that is a contraction of the first partial model, thereby suppressing a decrease in accuracy of the simulation calculation result caused by the deviation. Furthermore, since the learning model calculates n output values ​​B(t+a×dt) during one sampling period, the time required to obtain the simulation calculation result is not prolonged.

[0092] The present disclosure is not limited to the simulation device 10, the simulation method, and the simulation program according to the above-described embodiments, and may be implemented by various other modified examples or application examples without departing from the gist of the invention described in the claims. [Explanation of symbols]

[0093] 10 Simulation device 10A Simulation Device (Conventional Technology) 10a Communication Interface 10b Non-volatile memory 10c RAM (Random Access Memory) 10d storage section 10e CPU(Central processing unit) 10f Input / Output Interface 10g Input Device 10h output device 11. Communications Networks 20 Delay element 21 Delay element 23 Prediction Department 27 Acquisition Department 30 Vehicle Systems 31 Engine 32 Engine Controller 33 Shaft 34 Transmission (TM) 35 Transmission Controller 36 Shaft 37 Def 38 Shaft 39 Shaft 40 Tires 41 Tires S15 Engine control step S16 Engine Step S17 Transmission control step S18 Transmission Step S19 Vehicle Step S23 Prediction step S25 Transmission ROM Step S26 Transmission ROM step S27 Acquisition Step

Claims

1. A simulation device that performs coupled analysis of a computation model consisting of a plurality of partial computation models including a learning model obtained by degenerating a first partial model and a second partial model having a sampling time different from a sampling time of the first partial model, and simulates a target system represented by the computation model, a prediction unit that predicts the output value B(t) at time t and the output value B(t+a×dt) for each of n times (t+a×dt) after the time t and ... an acquisition unit that acquires the n output values ​​B(t+a×dt) predicted by the prediction unit, and extracts, from the n output values ​​B(t+a×dt), the output value B(t+a×dt) at the time (t+a×dt) corresponding to the sampling time of the second partial model as an input value of the second partial model; A simulation device comprising:

2. The simulation device according to claim 1, characterized in that the acquisition unit extracts, from the acquired n output values ​​B(t+a×dt), the output value B(t+a×dt) at the time (t+a×dt) closest to the sampling time of the second partial model as an input value of the second partial model.

3. The simulation device described in claim 1, characterized in that the acquisition unit extracts the output value B at the two times (t + a × dt) closest to the sampling time of the second partial model from the acquired n output values ​​B (t + a × dt), interpolates the output value B at the sampling time of the second partial model using the extracted two output values ​​B, and extracts the interpolated output value B as an input value of the second partial model.

4. The simulation device according to claim 1, characterized in that the learning model is a reduced-dimensional model and is generated through machine learning using the calculation results of the simulation obtained by previously running a simulation targeting the first partial model as learning data.

5. 5. The simulation device according to claim 4, wherein the learning model is a recurrent neural network model.

6. The simulation device according to claim 4 , wherein the learning model is generated through machine learning using gradient boosting with the learning data.

7. the target system is a vehicle system; A partial calculation model constituting a calculation model used for calculation processing of the simulation of the vehicle system includes: an engine plant model to which a computation process for simulating the operation of an engine that is included in the vehicle system and is a control target of the engine control is assigned; a transmission plant model to which a calculation process for simulating the operation of a transmission that is included in the vehicle system and is a control target of the transmission control is assigned; and a vehicle motion plant model that is included in the vehicle system and is assigned a computation process for simulating the motion of the vehicle connected to the transmission; Including, a sampling time of the transmission plant model and a sampling time of the vehicle motion plant model are different; the transmission plant model corresponds to the first partial model, the vehicle motion plant model corresponds to the second partial model, the prediction unit predicts the output value B(t) at the time t and the output value B(t+a×dt) for each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) after the time t and each of n times (t+a×dt) the acquisition unit acquires the n output values ​​B(t+a×dt) predicted by the prediction unit, and extracts, from the n output values ​​B(t+a×dt), the output value B at the time (t+a×dt) corresponding to the sampling time of the vehicle motion plant model as an input value of the vehicle motion plant model.

2. The simulation device according to claim 1.

8. the target system is a vehicle system; A partial calculation model constituting a calculation model used for calculation processing of the simulation of the vehicle system includes: an engine plant model to which a calculation process for simulating the operation of an engine that is included in the vehicle system and is a control target of the engine control is assigned; a transmission plant model to which a calculation process for simulating the operation of a transmission that is included in the vehicle system and is a control target of the transmission control is assigned; and a vehicle motion plant model that is included in the vehicle system and is assigned a computation process for simulating the motion of the vehicle connected to the transmission; a transmission model that is interposed between the transmission plant model and the vehicle motion plant model, that acquires a calculation result of the transmission plant model as the input value A(t) and transmits the calculation result to the vehicle motion plant model as the output value B(t); Including, a sampling time of the transmission model is different from a sampling time of the vehicle motion plant model; the transmission model corresponds to the first partial model; the vehicle motion plant model corresponds to the second partial model, the prediction unit predicts the output value B(t+a×dt) for each of n (n is an integer of 1 or more) times (t+a×dt) (a includes an integer from 1 to n) after the time t and intervals of a predetermined time dt (n is an integer of 1 or more), based on the input value A(t) at the time t, using the learning model that has previously learned a correspondence relationship between the input value A(t) of the transmission model at a certain time t and an output value B(t+a×dt) of the transmission model for each of n (n is an integer of 1 or more) times (t+a×dt) (a includes an integer from 1 to n) after the time t and intervals of a predetermined time dt, and transmits the n output values ​​B(t+a×dt) to the vehicle motion plant model; the acquisition unit acquires the n output values ​​B transmitted by the prediction unit, and extracts, from the n output values ​​B, the output value B at the time (t+a×dt) corresponding to the sampling time of the vehicle motion plant model as an input value of the vehicle motion plant model.

2. The simulation device according to claim 1.

9. A simulation method for simulating a target system represented by a computation model by performing coupled analysis on a computation model consisting of a plurality of partial computation models including a learning model obtained by degenerating a first partial model and a second partial model having a sampling time different from a sampling time of the first partial model, the simulation method comprising: The computer a prediction step of predicting, based on the input value A(t) at a certain time t, the output value B(t) of the first partial model at the time t and the output value B(t+a×dt) of the first partial model for each of n (n is an integer of 1 or more) times (t+a×dt) (a includes an integer from 1 to n) after the time t and each of n times (t+a×dt ... an acquisition step of acquiring the n output values ​​B(t+a×dt) predicted in the prediction step, and extracting, from the n output values ​​B(t+a×dt), the output value B(t+a×dt) at the time (t+a×dt) corresponding to the sampling time of the second partial model as an input value of the second partial model; A simulation method comprising the steps of:

10. A simulation program for simulating a target system represented by a computation model by performing coupled analysis on a computation model including a learning model obtained by degenerating a first partial model and a second partial model having a sampling time different from a sampling time of the first partial model, the computation model being composed of a plurality of partial computation models, On the computer, a prediction function for predicting, based on the input value A(t) at a certain time t, the output value B(t) of the first partial model at the time t and the output value B(t+a×dt) of the first partial model for each of n times (t+a×dt) (a includes an integer from 1 to n) after the time t and each of n times (t+a×dt) that are separated by a predetermined time dt (n is an integer equal to or greater than 1), using the learning model that has previously learned a correspondence relationship between the input value A(t) of the first partial model at the time t, the output value B(t) of the first partial model at the time t, and each of n times (t+a×dt) (a includes an integer from 1 to n) that are separated by a predetermined time dt after the time t; an acquisition function for acquiring the n output values ​​B(t+a×dt) predicted by the prediction function, and extracting, from the n output values ​​B(t+a×dt), the output value B(t+a×dt) at the time (t+a×dt) corresponding to the sampling time of the second partial model as an input value of the second partial model; A simulation program characterized by:

Images