Simulation execution apparatus, simulation execution method, and program

Abstract

It is desired to provide a technique for accurately estimating model parameters by data assimilation.SOLUTION: And a data assimilation unit configured to obtain a data assimilation result by executing sequential data assimilation on the basis of observation data obtained by observing a state of an observation target, the simulation result, and a variance of system noise in the simulation unit, in which the data assimilation unit makes a variance of system noise at an end of the sequential data assimilation smaller than a variance of system noise at a start of the sequential data assimilation.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a simulation execution device, a simulation execution method, and a program. [Background technology]

[0002] In various fields, there are techniques for modeling and simulating the laws operating in a target system in order to predict future events or verify behavior in hypothetical situations.

[0003] For example, Non-Patent Document 1 proposes a model that can simulate traffic flow on actual expressways with high accuracy and low computational cost. However, the simulation proposed in Non-Patent Document 1 is a method for calculating the time evolution of traffic conditions based on a model formula, and includes multiple model parameters such as traffic capacity and maximum traffic density per unit section. If these model parameters deviate from the true values, there is a concern that the reproducibility of actual phenomena will be reduced. Therefore, it is necessary to identify multiple model parameters with high accuracy.

[0004] In recent years, as a solution to these problems, there have been proposed parameter estimation devices, such as those disclosed in paragraphs 0004 to 0012 and 0015 to 0031 of Patent Document 1, which propose a model parameter estimation method using a data assimilation technique. Data assimilation is a technique for correcting a simulation model or simulation results using observation data obtained by observations such as sensors to increase the reproducibility of actual phenomena. In particular, when the simulation model has nonlinearity, a technique for sequentially assimilating data based on a set of many simulation results (for example, a technique called an ensemble Kalman filter or particle filter) may be used.

[0005] In general, data assimilation deals with models that include observation noise, which arises mainly from observation errors, and system noise, which arises from modeling errors in simulations, etc. In this case, the variance of the system noise is involved in the estimation of model parameters, and if the system noise is not set appropriately, there is a risk that the estimation accuracy and efficiency of the model parameters will decrease.

[0006] As a technology for suppressing such declines in estimation accuracy and estimation efficiency of model parameters, there is a parameter estimation device disclosed in paragraphs 0008 to 0020 of Patent Document 2, which proposes a model parameter estimation method for dynamically setting system noise in a data assimilation technique. This technology aims to improve the accuracy and stability of model parameter estimation by determining the variance-covariance matrix of the system noise so as to maximize the likelihood function of the distribution of simulation predicted values. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2020-201146 [Patent Document 2] Japanese Patent Application Laid-Open No. 2015-52491 [Non-patent literature]

[0008] [Non-Patent Document 1] Toru Takahashi, et al., "Development of a Dynamic Hybrid Traffic Flow Simulation Model and Its Verification Using Actual Expressway Data," Transactions of the Japan Society for Simulation Technology, 2021, Vol. 13, No. 1, pp. 37-47. [Non-patent document 2] Yasuyoshi Kato, "Road Traffic Simulation Using Cellular Automata Methods" (Special Issue: Intelligent Transport Systems (ITS) and AI). Japanese Society for Artificial Intelligence, Vol. 15, No. 2, 2000, pp. 242-250. Summary of the Invention [Problem to be solved by the invention]

[0009] In model parameter estimation using data assimilation, the estimation of model parameters tends to be stable when the variance of the system noise is small. However, when the variance of the system noise is small, the tracking of the model parameters to the true values ​​is not performed well, and the model parameters cannot reach the true values ​​that are far from the initial values, which may result in a decrease in the estimation accuracy of the model parameters.

[0010] Conversely, if the variance of the system noise is large, the model parameters will track well to the true values ​​and will reach true values ​​farther from their initial values ​​more quickly. However, if the variance of the system noise is large, the estimation of the model parameters will become unstable. Therefore, if the variance of the system noise becomes large at the end of the data assimilation process, the estimation of the model parameters will end up being unstable, and the accuracy of the estimation of the model parameters may decrease.

[0011] Therefore, it is desirable to provide a technology that can accurately estimate model parameters through data assimilation. [Means for solving the problem]

[0012] In order to solve the above-mentioned problems, according to one aspect of the present invention, there is provided a simulation execution device comprising: a simulation unit that performs a simulation using a model in which model parameters are set, to obtain a simulation result; and a data assimilation unit that performs sequential data assimilation based on observation data obtained by observing the state of an object to be observed, the simulation result, and a system noise variance in the simulation unit, to obtain a data assimilation result, wherein the data assimilation unit makes the system noise variance at the end of the sequential data assimilation smaller than the system noise variance at the start of the sequential data assimilation.

[0013] The simulation unit may perform a first simulation using a model in which initial values ​​of the model parameters are set, and the data assimilation unit may perform a first sequential data assimilation based on the result of the first simulation, observation data corresponding to the result of the first simulation, and an initial value of the variance of the system noise.

[0014] The simulation unit may execute a second simulation, which is a simulation using a model in which the most recently estimated model parameters are set, and the most recently estimated simulation results; the data assimilation unit executes a second sequential data assimilation, which is sequential data assimilation based on the results of the second simulation, observation data corresponding to the results of the second simulation, and a system noise variance that is smaller than the system noise variance used in the previous data assimilation; and the simulation execution device may include a processing control unit that executes reprocessing, in which the second simulation and the second sequential data assimilation are executed in sequence, one or more times.

[0015] The model parameters may include a plurality of parameters, a variance of the system noise may exist for each of the plurality of parameters, and the data assimilation unit may independently control the variance of the system noise corresponding to each of the plurality of parameters.

[0016] The data assimilation unit may perform the sequential data assimilation based on an ensemble Kalman filter or a particle filter.

[0017] In addition, according to another aspect of the present invention to solve the above-mentioned problems, there is provided a computer-executed simulation execution method, which includes: performing a simulation using a model in which model parameters are set to obtain a simulation result; performing sequential data assimilation based on observation data obtained by observing the state of an object to be observed, the simulation result, and a system noise variance in a simulation unit that executes the simulation to obtain a data assimilation result; and making the system noise variance at the end of the sequential data assimilation smaller than the system noise variance at the start of the sequential data assimilation.

[0018] Furthermore, in order to solve the above-mentioned problems, according to another aspect of the present invention, there is provided a program that causes a computer to function as a simulation unit that performs a simulation using a model in which model parameters are set to obtain a simulation result, and a data assimilation unit that performs sequential data assimilation based on observation data obtained by observing the state of an object to be observed, the simulation result, and the variance of system noise in the simulation unit to obtain a data assimilation result, wherein the data assimilation unit makes the variance of system noise at the end of the sequential data assimilation smaller than the variance of system noise at the start of the sequential data assimilation. [Effects of the Invention]

[0019] As described above, the present invention provides a technique that enables accurate estimation of model parameters through data assimilation. [Brief explanation of the drawings]

[0020] [Figure 1] 1 is a diagram illustrating an example of a functional configuration of a simulation execution device 1 according to an embodiment of the present invention. [Figure 2] 3 is a flowchart showing an example of the operation of the simulation execution device 1 according to the embodiment of the present invention. [Figure 3]10 is a flowchart showing details of a simulation process executed by a simulation unit 120. [Figure 4] 10 is a flowchart showing details of the data assimilation process executed by the data assimilation unit 130. [Figure 5] FIG. 10 is a diagram showing the results of comparing the estimated values ​​of model parameters while changing the initial value of the variance of the system noise and whether or not to perform system noise adjustment processing each time data is assimilated. [Figure 6] 1 is a diagram showing a hardware configuration of an information processing device 900 as an example of a simulation execution device 1 according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0021] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant explanations will be omitted.

[0022] (1. Details of the embodiment) The following describes details of the embodiments of the present invention.

[0023] (1-1. Configuration of the simulation execution device) First, an example of the configuration of a simulation execution device 1 according to an embodiment of the present invention will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the functional configuration of the simulation execution device 1 according to an embodiment of the present invention. Note that the operation of the simulation execution device 1 estimates parameters of a model used in a simulation (hereinafter simply referred to as "model parameters") and simulation results, and outputs the estimated model parameters and estimated simulation results.

[0024] As shown in FIG. 1, a simulation execution device 1 according to an embodiment of the present invention includes an observation data acquisition unit 110, a simulation unit 120, a data assimilation unit 130, and a process control unit 140.

[0025] (Observation data acquisition unit 110) The observation data acquisition unit 110 is a system that acquires observation data by observing the state of an observation target. The observation data acquisition unit 110 includes an observation data storage unit 111 and an observation unit 112.

[0026] The observation data storage unit 111 stores observation data obtained by observing the state of an observation target.

[0027] The observation unit 112 acquires observation data by observing the state of the observation target. Here, the observation data is time-series data. Therefore, the observation data may change over time. More specifically, the observation unit 112 has a sensor, and the observation data may be data (e.g., summed data or averaged data) obtained by continuously obtaining data in time series by the sensor every predetermined time (e.g., one minute).

[0028] The observation data is associated with a time. For example, the time associated with the observation data may increase by a predetermined value (for example, 1) from the oldest to the newest observation data obtained by the sensor.

[0029] The observation data may be obtained by a sensor external to the simulation execution device 1. In this case, the observation data obtained by the sensor external to the simulation execution device 1 may be input to the observation data acquisition unit 110. Here, various types of sensors may be envisioned as the sensor that acquires the observation data.

[0030] For example, if the observation target is traffic flow on a road, the sensor may be a vehicle detection unit (free flow antenna or vehicle detector) that detects vehicles on the road, a driving history acquisition unit (probe data collector) that acquires driving history data of vehicles in motion, etc. The observation data acquisition unit 110 may store the observation data observed and acquired by these sensors in the observation data storage unit 111.

[0031] (Simulation section 120) The simulation unit 120 is a system that performs a simulation using a model in which the laws of the observed object are modeled, and obtains a simulation result. The simulation unit 120 includes a model parameter storage unit 121, an initial condition acquisition unit 122, a simulation execution unit 123, and an output result storage unit 124.

[0032] The model parameter storage unit 121 stores model parameters, which are parameters of a model used in a simulation. The model parameters are set in the model.

[0033] The initial condition acquisition unit 122 acquires initial conditions (hereinafter simply referred to as "initial conditions") to be used in executing a simulation. For example, the initial condition acquisition unit 122 may read initial conditions prepared in advance, or may acquire initial conditions input by a user operation when a simulation is executed by the simulation execution unit 123.

[0034] At a stage where estimation of model parameters and simulation results has never been performed, the simulation execution unit 123 sets initial values ​​of model parameters in the model, and executes initial simulations multiple times based on initial conditions using the model in which the initial values ​​of the model parameters have been set. As a result, the simulation execution unit 123 obtains multiple initial simulation results corresponding to the initial values ​​of the model parameters and the initial conditions.

[0035] The number of times that a simulation is executed corresponding to the initial values ​​and initial conditions of the model parameters may be predetermined as the number of parallel executions of the simulation (hereinafter also referred to as the "ensemble number"). For example, the ensemble number may be 100 times. In the following description, execution of the ensemble number of simulations is also referred to as "one trial" of the simulation.

[0036] On the other hand, when the model parameters and simulation results have already been estimated at least once, the simulation execution unit 123 executes a second simulation several times in ensembles based on the last estimated simulation results using a model in which the last estimated model parameters are set, thereby obtaining the second simulation results corresponding to the last estimated model parameters and the last estimated simulation results, the number of times equal to the number of ensembles.

[0037] It should be noted that "immediately before" as used in this specification includes at least the time immediately preceding the time that the simulation execution device 1 is currently targeting for simulation or data assimilation (hereinafter simply referred to as the "target time"), and may also include times two or more prior to the target time.

[0038] Furthermore, the simulation execution unit 123 obtains model parameters for the ensemble number based on the simulation results of the ensemble number and the model. The simulation execution unit 123 associates a time with each of the simulation results and the model parameters. For example, the times associated with each of the simulation results and the model parameters may increase by a predetermined value from the oldest to the newest in the order in which the simulations and model parameters were obtained, similar to the times associated with the observation data.

[0039] In the embodiment of the present invention, it is mainly assumed that the operation termination condition (hereinafter also referred to as the "termination condition") of the simulation execution device 1 is determined in advance. Therefore, the amount of calculation per trial of the simulation executed by the simulation execution unit 123 does not need to be determined. However, the amount of calculation per trial of the simulation executed by the simulation execution unit 123 may be determined in advance, separately from the termination condition. The amount of calculation per trial of the simulation may be determined in any manner.

[0040] For example, the amount of calculation per trial of a simulation may be determined by the calculation time (e.g., 5 minutes) or the number of calculation steps per unit time (e.g., 10 steps). Alternatively, the amount of calculation per trial of a simulation may be determined according to at least one of the amount of observation data and the number of times data assimilation is performed, which will be described later. For example, the amount of calculation per trial of a simulation may be the amount obtained by dividing the amount of observation data by the number of times data assimilation is performed.

[0041] The type of model used in the simulation does not need to be limited. In other words, any model capable of outputting simulation results for initial conditions can be adopted as the model used in the simulation. For example, when the observation target is road traffic flow, a cellular automaton method or a cell transmission model, as described in Non-Patent Document 2, can be used as a representative traffic flow simulation model.

[0042] The output result storage unit 124 stores the simulation results of the ensemble numbers obtained by the simulation execution unit 123 and the model parameters of the ensemble numbers.

[0043] (Data Assimilation Unit 130) The data assimilation unit 130 is a system that performs data assimilation based on observation data and simulation results to obtain data assimilation results. The data assimilation results include simulation results of the ensemble number after data assimilation and model parameters of the ensemble number after data assimilation. In this specification, the data assimilation unit 130 includes a data assimilation setting acquisition unit 131, a system noise adjustment unit 132, a data assimilation execution unit 133, and a data assimilation result storage unit 134.

[0044] The data assimilation setting acquisition unit 131 acquires setting conditions (hereinafter simply referred to as "setting conditions") used for performing data assimilation. For example, the setting conditions may include the variance of observation noise occurring in the observation data, the initial value of the variance of system noise in the simulation unit 120, the termination condition, the calculation time per simulation trial, and the number of ensembles. As described above, system noise may arise from modeling errors in the simulation. Furthermore, in this specification, the variance of observation noise may be a fixed value, but the variance of system noise changes. Therefore, the setting conditions include the initial value of the variance of system noise.

[0045] For example, the data assimilation setting acquisition unit 131 may read setting conditions prepared in advance, or may acquire setting conditions input by user operation when data assimilation is performed by the data assimilation execution unit 133.

[0046] The system noise adjustment unit 132 controls the variance of the system noise. More specifically, the system noise adjustment unit 132 makes the variance of the system noise at the end of data assimilation smaller than the variance of the system noise at the start of data assimilation. This allows for accurate estimation of model parameters through data assimilation.

[0047] More specifically, when data assimilation has never been performed, the system noise adjustment unit 132 determines the initial value of the system noise variance included in the set conditions as the system noise variance to be used in the next data assimilation performed by the data assimilation execution unit 133.

[0048] On the other hand, when data assimilation has already been performed once or more, the system noise adjustment unit 132 determines, in accordance with noise adjustment conditions that specify how small the system noise variance should be, a system noise variance that is smaller than the system noise variance used in the immediately preceding data assimilation as the system noise variance to be used in the next data assimilation performed by the data assimilation execution unit 133.

[0049] The noise adjustment conditions may be defined in any manner. Specific examples of noise adjustment conditions will be described later. The noise adjustment conditions may be stored in advance by the system noise adjustment unit 132, or may be included in the setting conditions acquired by the data assimilation setting acquisition unit 131.

[0050] At a stage where data assimilation has never been performed, the data assimilation execution unit 133 performs initial data assimilation using the observation noise variance included in the setting conditions acquired by the data assimilation setting acquisition unit 131 and the system noise variance determined by the system noise adjustment unit 132, based on the initial simulation result stored in the output result storage unit 124, the model parameters corresponding to the initial simulation result, and the observation data associated with the same time as the time associated with the initial simulation result, thereby obtaining a data assimilation result.

[0051] The data assimilation that is first performed by the data assimilation unit 130 corresponds to the data assimilation at the start.

[0052] On the other hand, when data assimilation has already been performed once or more, the data assimilation execution unit 133 uses the observation noise variance included in the setting conditions acquired by the data assimilation setting acquisition unit 131 and the system noise variance determined by the system noise adjustment unit 132 to perform data assimilation again based on the re-simulation results stored in the output result storage unit 124, the model parameters corresponding to the re-simulation results, and the observation data associated with the same time as the time associated with the re-simulation results, thereby obtaining data assimilation results.

[0053] The final data assimilation performed by the data assimilation unit 130 corresponds to the data assimilation at the end.

[0054] The model parameters included in the data assimilation results are treated as estimated model parameters, and the simulation results included in the data assimilation results are treated as estimated simulation results.

[0055] Here, as an example of a data assimilation method, any method for performing sequential data assimilation on time-series data may be adopted. For example, a method called an ensemble Kalman filter or a particle filter may be adopted as the sequential data assimilation method. In the embodiment of the present invention, a case where an ensemble Kalman filter is adopted as the sequential data assimilation method will be mainly described.

[0056] The data assimilation results storage unit 134 stores the data assimilation results. The data assimilation results stored by the data assimilation results storage unit 134 may include a combination of estimated model parameters and estimated simulation results.

[0057] (Processing control unit 140) The process control unit 140 causes the simulation unit 120 to perform the first simulation, and causes the data assimilation unit 130 to perform the first data assimilation.

[0058] Furthermore, the process control unit 140 causes the simulation unit 120 and the data assimilation unit 130 to execute reprocessing, which sequentially executes a second simulation and a second data assimilation, one or more times until a predetermined termination condition is satisfied. Note that the number of times data assimilation is executed by the data assimilation unit 130 in this manner may be predetermined as the number of data assimilation executions.

[0059] The initial condition acquisition unit 122, the simulation execution unit 123, the data assimilation setting acquisition unit 131, the system noise adjustment unit 132, the data assimilation execution unit 133, and the process control unit 140 include a computing device such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit), and their functions can be realized by the computing device expanding a program stored in a ROM (Read Only Memory) into a RAM and executing it. In this case, a computer-readable recording medium on which the program is recorded can also be provided.

[0060] Alternatively, the initial condition acquisition unit 122, the simulation execution unit 123, the data assimilation setting acquisition unit 131, the system noise adjustment unit 132, the data assimilation execution unit 133, and the process control unit 140 may be configured by dedicated hardware or by a combination of multiple hardware components. Data necessary for calculations by the calculation device is stored as appropriate in a storage unit (not shown).

[0061] The observation data storage unit 111, the model parameter storage unit 121, the output result storage unit 124, and the data assimilation result storage unit 134 are realized by a storage unit (not shown), which may be configured by a memory such as a RAM (Random Access Memory), a hard disk drive, or a flash memory.

[0062] The above describes an example of the configuration of the simulation execution device 1 according to an embodiment of the present invention.

[0063] (1-2. Operation of the simulation execution device) Next, an example of the operation of the simulation execution device 1 according to the embodiment of the present invention will be described with reference to Figures 2 to 4. First, the overall flow of the operation of the simulation execution device 1 according to the embodiment of the present invention will be described with reference to Figure 2.

[0064] 2 is a flowchart showing an example of the operation of the simulation execution device 1 according to an embodiment of the present invention. Model parameters can be mainly estimated by the operation of the simulation execution device 1. The operation of the simulation execution device 1 can be broadly divided into three steps: "observation data acquisition processing," "simulation processing," and "data assimilation processing." The three steps of "simulation processing" and "data assimilation processing" are repeatedly executed until a predetermined termination condition is met.

[0065] First, the observation data acquisition unit 110 acquires observation data by observing the state of an observation target (S201).

[0066] Next, the simulation unit 120 executes a simulation process using the model in which the model parameters have been set, and obtains a simulation result (S202). At this time, a predetermined number of ensemble simulations are executed, and a simulation result for the number of ensembles is obtained in one step S202. Furthermore, the simulation unit 120 obtains model parameters for the number of ensembles based on the simulation result for the number of ensembles and the model.

[0067] Next, the data assimilation unit 130 performs a data assimilation process based on the observation data acquired by the observation data acquisition unit 110, the simulation results of the ensemble numbers acquired by the simulation unit 120, and the model parameters of the ensemble numbers acquired by the simulation unit 120, to obtain data assimilation results (S203). The model parameters included in the data assimilation results are regarded as estimated model parameters, and the simulation results included in the data assimilation results are regarded as estimated simulation results.

[0068] Next, if the termination condition is not satisfied ("NO" in S204), the process control unit 140 shifts the operation to a re-simulation process (S202) using the model in which the estimated model parameters have been set and the estimated simulation results.

[0069] On the other hand, if the termination condition is met (YES in S204), the process control unit 140 outputs the estimated model parameters and the estimated simulation results, and ends the operation.

[0070] In the following explanation, the "observation data acquisition process," "simulation process," and "data assimilation process" will be described in detail.

[0071] (Observation data acquisition process) The observation data acquisition unit 110 stores the observation data obtained by the observation unit 112 in the observation data storage unit 111. The observation data may be data obtained by continuously obtaining data in time series by a sensor and collating the data into statistics every predetermined time (for example, one minute).

[0072] As described above, the observation data may be obtained by a sensor external to the simulation execution device 1. In this case, the observation data obtained by the sensor external to the simulation execution device 1 may be input to the observation data acquisition unit 110 and acquired by the observation data acquisition unit 110.

[0073] (Simulation processing) The detailed operation of the simulation process executed by the simulation unit 120 will be described with reference to Fig. 3. Fig. 3 is a flowchart showing the details of the simulation process executed by the simulation unit 120.

[0074] First, the initial condition acquisition unit 122 acquires the initial conditions of the simulation (S301). As described above, the initial condition acquisition unit 122 may read the initial conditions prepared in advance, or may acquire the initial conditions input by the user when the simulation is executed by the simulation execution unit 123.

[0075] Next, the simulation execution unit 123 acquires the initial values ​​of the model parameters stored in the model parameter storage unit 121 (S302). The simulation execution unit 123 executes a simulation of the ensemble number based on the initial conditions using the model in which the initial values ​​are set as the model parameters. As a result, the simulation execution unit 123 obtains a simulation result of the ensemble number corresponding to the initial values ​​and the initial conditions (S303).

[0076] 3, it is assumed that S303 is executed for the first time, but it is also assumed that S303 has already been executed one or more times.

[0077] In this case, the simulation execution unit 123 acquires the most recent estimated simulation results instead of the initial conditions in S301, and acquires the most recent estimated model parameters instead of the initial values ​​of the model parameters in S302. Then, in S303, the simulation execution unit 123 executes a simulation of the ensemble number based on the most recent estimated simulation results using a model in which the most recent estimated model parameters are set. As a result, the simulation execution unit 123 obtains simulation results of the ensemble number corresponding to the estimated model parameters and the estimated simulation results.

[0078] Furthermore, the simulation execution unit 123 obtains model parameters for the ensemble numbers based on the simulation results of the ensemble numbers and the model. When the simulation by the simulation execution unit 123 is completed, the simulation unit 120 stores the output results from the model (i.e., the simulation results of the ensemble numbers and the model parameters of the ensemble numbers) obtained by the simulation by the simulation execution unit 123 in the output result storage unit 124 (S304).

[0079] (Data assimilation processing) The detailed operation of the data assimilation process executed by the data assimilation unit 130 will be described with reference to Fig. 4. Fig. 4 is a flowchart showing the details of the data assimilation process executed by the data assimilation unit 130.

[0080] First, the data assimilation setting acquisition unit 131 acquires the setting conditions used for executing data assimilation (S401). The setting conditions may include the variance of the observation noise, the initial value of the variance of the system noise, the termination condition, the calculation time per trial of the simulation, the number of ensembles, etc.

[0081] The termination condition may be determined by the number of times data assimilation is performed (for example, 50 times).

[0082] Alternatively, the termination condition may be determined based on the total simulation time (e.g., 3 hours). When the termination condition is determined based on the total simulation time, the data assimilation setting acquisition unit 131 may calculate the data assimilation execution time based on the total simulation time and the calculation time per trial of the simulation, and may use the calculated data assimilation execution time as the termination condition.

[0083] Alternatively, the cutoff condition may be determined by the rate of change of the model parameters due to data assimilation (for example, the rate of change is 1% or less).

[0084] 4 shows an example in which the setting conditions are acquired (S401) each time the data assimilation process (S203) is executed. However, the setting conditions do not have to be acquired (S401) each time the data assimilation process (S203) is executed. For example, the data assimilation setting acquisition unit 131 may acquire the setting conditions before the simulation process (S202) is executed (e.g., in acquiring the observation data shown in S201).

[0085] Next, the system noise adjustment unit 132 determines the variance of the system noise and performs the system noise adjustment process based on the determined variance of the system noise (S402). As described above, the system noise adjustment unit 132 may set the variance of the system noise at the end of the data assimilation to be smaller than the variance of the system noise at the start of the data assimilation.

[0086] More specifically, when data assimilation has never been performed (S403), the system noise adjustment unit 132 determines the initial value of the system noise variance as the new system noise variance. On the other hand, when data assimilation has already been performed at least once, the process control unit 140 determines, as the new system noise variance, a system noise variance that is smaller than the system noise variance used in the immediately preceding data assimilation (S403), based on noise adjustment conditions that specify how small the system noise variance should be.

[0087] The noise adjustment conditions may be specified in any manner. For example, the noise adjustment conditions may be specified so that the variance of the system noise at the start of data assimilation is a value obtained by multiplying the initial value of the model parameter by a predetermined ratio, and the variance of the system noise in the second or subsequent data assimilation is a value obtained by multiplying the variance of the system noise used in the immediately preceding data assimilation by a constant smaller than 1.

[0088] For example, the predetermined rate by which the initial value of the model parameter is multiplied may be 5%, and the constant by which the variance of the system noise is multiplied may be 0.9. In this case, the variance of the system noise in the first data assimilation may be 5% of the initial value of the model parameter, the variance of the system noise in the second data assimilation may be 5% × 0.9 = 4.5% of the initial value of the model parameter, and the variance of the system noise in the third data assimilation may be 4.5% × 0.9 = 4.05% of the initial value of the model parameter.

[0089] Alternatively, a monotonically decreasing function F(x) may be prepared, with the number of data assimilation runs being x, and the variance of the system noise may be set to the value of F(x). The number of data assimilation runs x in the first data assimilation run may be 1, and the number of data assimilation runs x in the second data assimilation run may be 2. Alternatively, the monotonically decreasing function F(x)=A / x (where A is a constant) may be used.

[0090] It is also conceivable that the model parameters include a plurality of parameters. In such a case, the system noise variance may be set in common to the plurality of parameters. In this case, one or both of the initial value of the system noise variance and the noise adjustment condition may be set in common to the plurality of parameters. In this case, the system noise adjustment unit 132 may control the system noise variance set in common to each of the plurality of parameters.

[0091] Alternatively, the system noise variance may be set separately for each of the multiple parameters. In this case, one or both of the initial value of the system noise variance and the noise adjustment condition may be set separately for each of the multiple parameters. In this case, the system noise adjustment unit 132 may independently control the system noise variance corresponding to each of the multiple parameters.

[0092] Furthermore, repeated execution of the system noise adjustment process (S402) may result in the variance of the system noise becoming too small, making it difficult for the model parameters to track the true values. In such cases, a lower limit may be set for the variance of the system noise. In this case, once the variance of the system noise reaches the lower limit, the variance of the system noise may be fixed at the lower limit. Such a lower limit may be included in the noise adjustment conditions.

[0093] The data assimilation unit 133 can obtain the ensemble number included in the set conditions and use it as the ensemble number to be used for data assimilation. The data assimilation unit 133 then performs data assimilation based on the simulation results of the ensemble number obtained in S202, the model parameters of the ensemble number, and the observation data corresponding to the same time as the simulation results (S403). In this way, the data assimilation unit 133 obtains the data assimilation results.

[0094] When performing data assimilation, the data assimilation execution unit 133 takes into consideration the variance of the observation noise included in the set conditions and the variance of the system noise determined by the system noise adjustment unit 132 .

[0095] For example, the data assimilation execution unit 133 acquires the variance of the observation noise included in the set conditions, and randomly determines the observation noise for each ensemble so that the variance of the acquired observation noise is the same as the variance of the observation noise to be determined, and so that the observation noise follows a normal distribution.

[0096] Similarly, the data assimilation execution unit 133 acquires the variance of the system noise determined by the system noise adjustment unit 132, and randomly determines the system noise for each ensemble so that the acquired variance of the system noise is the same as the variance of the system noise to be determined, and so that the system noise follows a normal distribution.

[0097] The data assimilation execution unit 133 then uses the determined observation noise and system noise for data assimilation. Subsequently, the data assimilation unit 130 stores the data assimilation results obtained by the data assimilation execution unit 133 in the data assimilation result storage unit 134 (S404).

[0098] The details of the embodiment of the present invention have been described above.

[0099] (1-3. Effects) As described above, according to an embodiment of the present invention, a simulation is performed using a model in which the laws of the observed object are modeled, and data assimilation is performed using the simulation results, observation data acquired by observing the state of the observed object, and the variance of the system noise. Here, the variance of the system noise is adjusted so that the variance of the system noise at the end of the data assimilation is smaller than the variance of the system noise at the start of the data assimilation. This allows for accurate and stable estimation of model parameters.

[0100] In the following explanation, as a simple example, we will show the effect that can be obtained when the model parameters of the simulation are estimated using a model that models frictionless simple harmonic motion. As the model formula for the simulation, Equation (1) can be adopted, which is obtained by differentiating the equation of motion of frictionless simple harmonic motion and ignoring the infinitesimal terms of third order and above.

[0101] x t =2x t-1 -(1+ω 2 Δt 2 )x t-2 ···(1)

[0102] where x t is the position of the object at time t (simulation result), Δt is the time step size of the simulation (predetermined time), and ω is the angular frequency of vibration (model parameter). In other words, by inputting the initial conditions x0 and x1 into the model, the position of the object at times t=2 and later can be obtained.

[0103] The angular frequency ω is a model parameter, and it is assumed that this is obtained by parameter estimation. In this experiment, ω = ω t The simulation results obtained by running the simulation with the value of ω are added with observation noise, and the data obtained is used as the observation data. t When a value of ω different from the above is given as the initial value of the model parameters and data assimilation is performed, the true value ω t This method is a verification method known as a twin experiment.

[0104] In practice, the variance of the system noise is estimated using modeling errors, etc. However, in order to clarify the effects achieved by the embodiment of the present invention, two types of variance of the system noise were set in this experiment. In this experiment, the variance of the observed noise was set to 20% of the amplitude of the simple harmonic motion.

[0105] FIG. 5 shows the results of comparing the estimated values ​​of the model parameters while changing the initial value of the variance of the system noise and whether or not to perform system noise adjustment processing each time data assimilation is performed.

[0106] Referring to Figure 5, the graph shows the progress of the estimated values ​​of the model parameters for each of the following cases: (a) the system noise variance is kept constant from the start to the end of data assimilation, and the initial value of the system noise variance is set to 1% of the initial value of the model parameter; (b) the system noise variance is kept constant from the start to the end of data assimilation, and the initial value of the system noise variance is set to 0.01% of the initial value of the model parameter; and (c) a system noise adjustment process is performed each time data assimilation is performed, where the system noise variance used in data assimilation is set to 0.7 times the system noise variance in the immediately preceding data assimilation, and the initial value of the system noise variance is set to 1% of the initial value of the model parameter.

[0107] The estimated value is the average value of the model parameters for the ensemble number. In this experiment, the ensemble number was set to 100. Note that if the initial value of the system noise variance is small and system noise adjustment processing is performed, the variance of the system noise will become too small, and it is clear that tracking to the true value will not be performed well, so this was not verified.

[0108] First, as in the example shown in Figure 5(a), when the initial value of the system noise variance is large and system noise adjustment processing is not performed, the transition of the model parameters is unstable from the beginning, and the model parameter estimates obtained by data assimilation fluctuate significantly even after they reach their true values. The reason for this is thought to be that the variance of the system noise is large, resulting in large variations in the correction of the simulation results by data assimilation.

[0109] Next, as shown in the example in Figure 5(b), when the initial value of the system noise variance is small and system noise adjustment processing is not performed, the estimated values ​​of the model parameters eventually reach their true values, although many data assimilation runs are required. However, the number of data assimilation runs required for the estimated values ​​of the model parameters to reach their true values ​​is large. The reason for this is thought to be that the variance of the system noise is small and the amount of correction to the simulation results by data assimilation is also small, so the model parameters are corrected stably. However, the amount of correction is small from the start to the end of data assimilation, so it takes a long time to correct the model parameters.

[0110] Finally, as shown in the example in Figure 5(c), when the initial value of the system noise variance is large and the system noise adjustment process is performed, although the fluctuations in the estimated values ​​of the model parameters are somewhat large at first, the estimated values ​​of the model parameters reach their true values ​​at an early stage and thereafter stably maintain values ​​close to the true values. This is thought to be because, while the estimated values ​​of the model parameters are largely corrected toward the true values ​​when the variance of the system noise is large, the variance of the system noise decreases as the number of data assimilation runs increases due to the system noise adjustment process, resulting in stable estimation of the model parameters.

[0111] The effects achieved by the simulation execution device 1 according to the embodiment of the present invention have been described above.

[0112] (2. Hardware configuration example) Next, an example of the hardware configuration of the simulation execution device 1 according to an embodiment of the present invention will be described.

[0113] Below, an example of the hardware configuration of an information processing device 900 will be described as an example of the hardware configuration of a simulation execution device 1 according to an embodiment of the present invention. Note that the example of the hardware configuration of the information processing device 900 described below is merely one example of the hardware configuration of the simulation execution device 1. Therefore, the hardware configuration of the simulation execution device 1 may be such that unnecessary components are deleted from the hardware configuration of the information processing device 900 described below, or new components are added.

[0114] 6 is a diagram showing the hardware configuration of an information processing device 900 as an example of the simulation execution device 1 according to an embodiment of the present invention. The information processing device 900 includes a CPU (Central Processing Unit) 901, a ROM (Read Only Memory) 902, a RAM (Random Access Memory) 903, a host bus 904, a bridge 905, an external bus 906, an interface 907, an input device 908, an output device 909, a storage device 910, and a communication device 911.

[0115] The CPU 901 functions as an arithmetic processing unit and control unit, and controls the overall operation of the information processing device 900 in accordance with various programs. The CPU 901 may also be a microprocessor. The ROM 902 stores programs used by the CPU 901, calculation parameters, etc. The RAM 903 temporarily stores programs used in the execution of the CPU 901, parameters that change as appropriate during the execution, etc. These are interconnected by a host bus 904 that is composed of a CPU bus, etc.

[0116] The host bus 904 is connected to an external bus 906, such as a PCI (Peripheral Component Interconnect / Interface) bus, via a bridge 905. It is not necessary to configure the host bus 904, bridge 905, and external bus 906 separately, and these functions may be implemented on a single bus.

[0117] The input device 908 is composed of input means such as a mouse, keyboard, touch panel, buttons, microphone, switches, and levers that allow the user to input information, and an input control circuit that generates an input signal based on the user's input and outputs it to the CPU 901. By operating this input device 908, the user operating the information processing device 900 can input various data to the information processing device 900 and instruct the information processing device 900 to perform processing operations.

[0118] The output device 909 includes, for example, a display device such as a CRT (Cathode Ray Tube) display device, a liquid crystal display (LCD) device, an OLED (Organic Light Emitting Diode) device, or a lamp, and an audio output device such as a speaker.

[0119] The storage device 910 is a device for storing data. The storage device 910 may include a storage medium, a recording device for recording data on the storage medium, a reading device for reading data from the storage medium, and a deletion device for deleting data recorded on the storage medium. The storage device 910 is configured, for example, with an HDD (Hard Disk Drive). This storage device 910 drives a hard disk and stores programs executed by the CPU 901 and various data.

[0120] The communication device 911 is, for example, a communication interface configured with a communication device for connecting to a network, etc. The communication device 911 may be compatible with either wireless communication or wired communication.

[0121] An example of the hardware configuration of the simulation execution device 1 according to an embodiment of the present invention has been described above.

[0122] (3. Supplementary Information) Although the preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, the present invention is not limited to these examples. It is clear that a person skilled in the art to which the present invention pertains can conceive of various modifications and alterations within the scope of the technical ideas set forth in the claims, and it is understood that these also naturally fall within the technical scope of the present invention. [Explanation of symbols]

[0123] 1 Simulation execution device 110 Observation data acquisition unit 111 Observation data storage unit 112 Observation Section 120 Simulation Department 121 Model parameter storage unit 122 Initial condition acquisition section 123 Simulation Execution Unit 124 Output result storage unit 130 Data Assimilation Department 131 Data Assimilation Setting Acquisition Unit 132 System noise adjustment section 133 Data Assimilation Executive Unit 134 Data Assimilation Results Storage Unit 140 Processing control section

Claims

1. a simulation unit that executes a simulation using the model in which the model parameters are set and obtains a simulation result; a data assimilation unit that performs sequential data assimilation based on observation data obtained by observing the state of an observation target, the simulation results, and the variance of system noise in the simulation unit to obtain a data assimilation result; Equipped with the data assimilation unit makes the variance of the system noise at the end of the sequential data assimilation smaller than the variance of the system noise at the start of the sequential data assimilation; Simulation execution device.

2. the simulation unit executes a first simulation using a model in which initial values ​​of the model parameters are set, the data assimilation unit performs initial sequential data assimilation based on a result of the initial simulation, observation data corresponding to the result of the initial simulation, and an initial value of the variance of the system noise; The simulation execution device according to claim 1 .

3. the simulation unit executes a second simulation using a model in which the model parameters estimated immediately before are set and a simulation result estimated immediately before; the data assimilation unit performs a second sequential data assimilation based on a result of the second simulation, observation data corresponding to the result of the second simulation, and a system noise variance that is smaller than the system noise variance used in the immediately preceding data assimilation; the simulation execution device, a processing control unit that executes the re-processing of sequentially executing the re-simulation and the re-sequential data assimilation one or more times; The simulation execution device according to claim 2 .

4. The model parameters include a plurality of parameters, a variance of the system noise exists for each of a plurality of parameters; the data assimilation unit independently controls the variance of system noise corresponding to each of the plurality of parameters; The simulation execution device according to claim 1 .

5. the data assimilation unit performs the sequential data assimilation based on an ensemble Kalman filter or a particle filter; The simulation execution device according to claim 1 .

6. performing a simulation using the model for which the model parameters have been set to obtain a simulation result; performing sequential data assimilation based on observation data obtained by observing the state of the observation target, the simulation results, and the variance of system noise in a simulation unit that executes the simulation to obtain a data assimilation result; The variance of the system noise at the end of the sequential data assimilation is smaller than the variance of the system noise at the start of the sequential data assimilation; A computer-implemented method for performing a simulation, comprising:

7. Computer, a simulation unit that executes a simulation using the model in which the model parameters are set and obtains a simulation result; a data assimilation unit that performs sequential data assimilation based on observation data obtained by observing the state of an observation target, the simulation results, and the variance of system noise in the simulation unit to obtain a data assimilation result; It functions as the data assimilation unit makes the variance of the system noise at the end of the sequential data assimilation smaller than the variance of the system noise at the start of the sequential data assimilation; program.

Images