JP2014173806A - Simulation executing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To shorten a total calculation time at Monte Carlo simulation under a large scale scenario with a modeling in real world being applied as a main object.SOLUTION: A dependence relation between units appeared in a scenario is specified at Monte Carlo simulation under utilization of a result of first time execution and an execution unit of the simulation is divided in response to the dependence relation. In reference to this operation, units influencing to each other is specified by every probabilistic elements, thereby computation amount is reduced by a thinned-out calculation so as to shorten a total execution time.

Description

  The present invention relates to a method for executing a Monte Carlo simulation using a computer according to a large-scale scenario modeling a real world.

  In a computer simulation including a stochastic element, a Monte Carlo simulation is used in which a statistical execution result is obtained by repeatedly executing a simulation. When performing a Monte Carlo simulation using a computer system for a large-scale scenario that assumes the behavior of various configuration systems in the real world, the scale and complexity of the scenario (for example, the number of units, parameters, and statistics of the system) Depending on the calculation conditions (such as the number of target elements), a considerable number of calculation execution times are required.

  2. Description of the Related Art Conventionally, a simulation execution method that shortens the total execution time of a simulation by using the information found in the first simulation execution to improve the efficiency of the second and subsequent executions is known (for example, Patent Documents). 1). If the occurrence time of an event that became known by the first execution is acquired and there is no stochastic element up to that time, use the occurrence time of the event that became known when executing the second and subsequent times Thus, the total simulation execution time is shortened. For example, between events where the time of occurrence is known in the first simulation and execution with a fine time step width is not required, the time step width is increased to reduce the amount of computation to shorten the total execution time.

JP 2007-78288 A

  In the above conventional simulation execution method, some computation processing during the period when detailed simulation is unnecessary can be omitted, but simulation models of various units appearing in the scenario are generated for all units, and the execution result of the scenario is obtained. It is necessary to process a simulation for prediction. For this reason, there is a problem that the contribution of the shortening effect to the total execution time of the entire simulation is small.

  The present invention has been made to solve such problems, and aims to reduce the total execution time of the simulation by localizing units appearing in the scenario generated at the time of simulation execution and reducing the amount of calculation. To do.

  The simulation execution method according to the present invention is a simulation execution method for repeatedly executing a simulation based on a scenario modeling the behavior of a preset unit at a plurality of time steps by Monte Carlo simulation using a computer. Using the first simulation execution process for executing the simulation in the type step and the first simulation execution result, the dependency relationship between the units appearing in the scenario is specified, and the stochastic element is determined from the specified dependency relationship. Extraction processing that extracts the behavior of the unit that interacts as an event pattern, division processing that divides the part that does not include the event pattern from the execution unit of the simulation as a template, and interaction by the stochastic element The unit to be localized, re-execute the simulation execution unit corresponding to the event pattern of the unit, combine the template with the re-execution result, and use the combination result for the next and subsequent type step arithmetic processing Loop processing for repeatedly executing.

  According to the present invention, the amount of calculation can be reduced by limiting the number of units involved in the simulation that are generated when the large-scale scenario is simulated.

6 is a diagram showing a flow of a simulation execution method according to Embodiment 1. FIG. 6 is a diagram illustrating an example of creating an event pattern according to Embodiment 1. FIG. 6 is a diagram illustrating an example of creating an event schedule according to Embodiment 1. FIG. 6 is a diagram illustrating an example of creating an execution result template according to Embodiment 1. FIG. It is a figure which shows the data model in the relational database by Embodiment 1. FIG. FIG. 10 is a diagram illustrating a flow of a simulation execution method according to the second embodiment. FIG. 10 is a diagram illustrating a determination example of a similar event model according to the second embodiment. FIG. 10 is a diagram illustrating a configuration of a computer system according to a third embodiment.

Embodiment 1 FIG.
A simulation execution method according to the first embodiment of the present invention will be described with reference to FIGS. FIG. 1 is a diagram showing a flow of a simulation execution method according to the first embodiment. FIG. 2 is a diagram illustrating an example of creating an event pattern according to the first embodiment. FIG. 3 is a diagram illustrating an example of creating an event schedule according to the first embodiment. FIG. 4 is a diagram illustrating an example of creating an execution result template according to the first embodiment. FIG. 5 is a diagram showing a data model in the relational database according to the first embodiment. The simulation execution method according to the first embodiment relates to a real system composed of a plurality of units in the real world, and the computer system performs the execution result of the scenario for a large-scale scenario in which the behavior of the real system is modeled on the computer. When performing a Monte Carlo simulation for predicting the above, the simulation is executed efficiently. The simulation execution method according to the first embodiment is executed by a computer system including a computer, an input / output device connected to the computer, and an external storage device connected to the computer. A simulation program of Monte Carlo simulation based on the time step method is implemented in an execution format in the internal storage device and the external storage device of the computer. The external storage device constitutes a relational database. The scenario is stored in the relational database in advance, and the execution result data is updated on the relational database as the simulation is executed. Examples of the actual system include a fire control system and an information system (hereinafter referred to as a target response system) that deals with threat targets using a plurality of fire control devices.

  In the simulation execution method of the first embodiment, the computer system executes a simulation by Monte Carlo simulation based on the time step method. When the computer system executes a simulation after the second type step, the total calculation time is shortened by thinning out unnecessary simulations using the result executed in the first type step. At this time, the dependence of the interaction between the units appearing in the scenario is specified using the first execution result, and the simulation execution unit is divided according to the dependence. This makes it possible to localize units that interact with each other by a stochastic element during the second and subsequent executions, reduce the amount of computation by executing simulation for each execution unit, and shorten the total execution time of the simulation. Yes. This will be described in detail below.

  In FIG. 1, the simulation execution method according to the first embodiment is composed of the processes of Step 1 to Step 11 and Steps 70 and 80 processed by the computer system. In step 1, the first simulation execution of the Monte Carlo simulation is performed based on a preset scenario. Steps 2, 3, and 4 are extraction processes, steps 5 and 6 are division processes, and steps 7, 8, 80, and 70 are loop processes.

  In step 2, an event pattern composed of specific dependencies between units is created from the result of the first type step simulation execution of the Monte Carlo simulation using FIG. 2 described later.

  In step 3, an event pattern execution schedule (hereinafter referred to as an event schedule) created using FIG. 3 described later is created.

  In step 4, an execution result for each event pattern by the first simulation execution is created using FIG. 4 described later.

  In step 5, an execution result template is created based on the execution result for each event pattern by the first simulation execution.

  In step 6, the execution result and execution result template for each event pattern created in steps 4 and 5 are registered in the relational database, and the data registered in the database can be reused in the second and subsequent executions.

Here, in step 2 of FIG. 1, the computer system uses the information input from the operator via the input / output device or the automatic processing input of the automatic program to generate a probabilistic element in the scenario from the first simulation execution result. Dependencies between units that influence each other or interact with each other are identified, and event patterns of the identified dependencies are created. The event pattern is composed of the following contents.
(1) Dependent unit type (for example, aircraft, radar site, command office, threat target, etc.)
(2) Unit performance specifications and behavior or behavior conditions (for example, maximum speed of each unit, rules for dealing with threat targets, etc.)
(3) Types of dependencies (for example, detection and response of threat targets, reporting of detection results, threat judgment, command control, etc.)
(4) Time when dependency relationship occurs

  Here, for example, in the case of a target response system including an aircraft, a radar site, a command control station, and a threat target in a unit, the radar site or aircraft reports a detection result (hereinafter referred to as a detection report) based on the detected threat target information. ) And send it to the command center. The command control station receives the radar site detection report, performs threat determination of the threat target based on the detection report, generates information for command control (hereinafter referred to as command control), and transmits it to the aircraft or the coping device. The aircraft receives a command and control from the command and control station, approaches the threat target, and uses a firearm to deal with the threat target. In response to the command and control from the command and control station, the coping device directs the firearm to the threat target and deals with the threat target. Such behavior of each unit is included in the scenario in advance.

FIG. 2 shows an example of creating a dependency relationship between units in an event pattern. The computer system extracts the dependency relationship between the units shown below from the first execution result, and creates an event that occurs in association with a specific unit as an event pattern.
(1) Detection: Radar site A detects threat target A, Aircraft A detects threat target A (2) Detection report: Radar site A detects detection to command control station A (3) Threat determination: Command control station A (4) Command and Control: Command and Control Center A controls Aircraft A and controls (5) Action: Aircraft A deals with Threat Target A

  In step 3 of FIG. 1, the computer system creates a schedule that each unit related to the event pattern created in step 2 executes based on the scenario as an event schedule.

FIG. 3 shows an example of creating an event schedule for each event pattern from the execution result of the first simulation. The event schedule is composed of the following contents.
(1) Types of all units appearing in the scenario (aircraft A or aircraft B, radar site A or radar site B, command and control device A or command and control device B, aircraft A or countermeasure device B, etc.)
(2) Event pattern (event pattern A, event pattern B, etc.)
(3) Event pattern occurrence schedule

  In step 4 of FIG. 1, the computer system creates a simulation execution result for each unit for each event pattern from the first execution result. In step 5, the computer system creates an execution result template by subtracting the event pattern execution result from the first execution result.

  FIG. 4 shows an example of creating an execution result template. The execution result template includes all the units appearing in the scenario and the first execution result of the behavior of each unit according to the simulation time. However, the execution result template is provided with an area to which the execution result for each event pattern is applied, and the execution result of each unit in the area is excluded.

  In step 6 of FIG. 1, the computer system registers the event pattern, event schedule, execution result template, and execution result for each first event pattern created in steps 1 to 5 in the relational database. The data registered in this database can be reused in the second and subsequent executions.

  FIG. 5 shows a data model of data registered in the relational database handled by the computer system of the first embodiment. In the explanatory diagram, 1 or n indicates a one-to-n dependency of the data model. This one-to-n dependency indicates that, for example, a plurality of event schedules exist for one scenario according to the performance specifications and action conditions of the units that appear in the scenario.

  In step 7 of FIG. 1, the computer system performs steps 8 to 11 as executions in the second and subsequent type steps. In the second and subsequent executions, only the created event pattern is repeatedly executed (trial number minus 1) times. That is, the entire simulation is not executed in the second and subsequent executions.

  In step 8 of FIG. 1, the computer system repeatedly executes steps 9 to 11 by the number of event patterns created in step 2.

  In step 9 of FIG. 1, in the event pattern execution, simulation execution is performed only at the time when the dependency relationship occurs for the unit having the dependency relationship. The execution result of the event pattern varies with the number of trials by the Monte Carlo simulation.

  In step 10 of FIG. 1, the computer system combines the execution results of the second and subsequent event patterns with the area to which the execution result of the execution result template is applied, and obtains the second and subsequent simulation execution results.

  In step 11 of FIG. 1, the computer system registers the execution result obtained from the second and subsequent executions in the relational database based on the data model shown in FIG. This is to enable reuse by the simulation execution method shown in the second embodiment.

  Steps 8 and 80 constitute a loop process. Steps 7 and 70 constitute a loop process. After the process of step 11 in FIG. 1, the process of returning to step 8 is performed in step 80 until the repetition process for the number of event patterns is completed. If the number of event patterns has been repeated, the process returns to step 7 and repeats the next (third and subsequent) simulation executions until the number of trials (number of trials minus 1) is completed in step 70. .

  As described above, the simulation execution method according to the first embodiment repeats simulation based on a scenario that models the behavior of a preset unit at a plurality of time steps by Monte Carlo simulation using a computer. A method for identifying dependencies between units appearing in the scenario by using the first simulation execution process (step 1) for executing the simulation in the first type step and the first simulation execution result. Then, the extraction process (steps 2, 3, 4) for extracting the behavior of the unit that interacts with the stochastic element from the specified dependency relationship as an event pattern, and the part not including the event pattern from the execution unit of the simulation The balance The processing to divide the data as a group (steps 5 and 6) and the units that interact with the stochastic element are localized, and the simulation execution unit corresponding to the event pattern of the unit is re-executed (step 9). A loop process (steps 7, 8, 80, and 70) that combines the template with the re-execution result (steps 10 and 11) and repeatedly executes the calculation process of the next type step using the combination result; It is equipped with.

  With this simulation execution method, in a Monte Carlo simulation of a large-scale scenario, the unit generated at the time of execution is localized as an event pattern from the dependency relationship between the units, and the second and subsequent executions are performed by executing only a specific event pattern The amount can be reduced and the total execution time can be shortened.

Embodiment 2. FIG.
A simulation execution method according to the second embodiment of the present invention will be described with reference to FIGS. FIG. 6 is a diagram illustrating a flow of a simulation execution method according to the second embodiment. FIG. 7 is a diagram illustrating a determination example of a similar event model according to the second embodiment.

  The execution method of the Monte Carlo simulation according to the second embodiment is a scenario in which the computer system accumulates from the dependency relationship between the units appearing in the scenario when the performance specifications and behavior conditions of the scenario accumulated in the relational database are changed. The similarity relationship is determined. The computer system reduces the amount of computation by reusing the accumulated execution results for the parts where the scenarios match, and for the parts where the scenarios differ, by locally executing the units that influence the stochastic factors at the time of execution. The total execution time is shortened.

  In FIG. 6, the simulation execution method according to the second embodiment is composed of processes from step 12 to step 26 and steps 200 and 210 which are processed by the computer system. Steps 12, 13, 14, 15, 16, 17, 20, 21, 25, 26, 210, and 200 in FIG. 6 are steps 1, 2, 3, 4, 5, 6, 7, 8, It corresponds to 10, 11, 80, 70. 6 differs from FIG. 1 in that steps 18 and 19 are newly provided, and step 9 in FIG. 1 is replaced with steps 22, 23 and 24.

  In step 12, simulation execution is performed in the first type step of the Monte Carlo simulation based on a preset scenario. From this first simulation execution (step 12), creation of an event pattern (step 13), creation of an event schedule (step 14), creation of an execution result for each first event pattern (step 15), and execution result template In each process from creation (step 16) to registration of those products in the relational database (step 17), the computer system executes processing operations as in FIG. 1 of the first embodiment.

  In step 18 in FIG. 6, the computer system determines a similar event pattern that is a branching point indicating which of the first and second embodiments is to be executed. In the determination of the similar event pattern, it is determined whether or not the event pattern is a similar event pattern based on the dependency relationship of the event pattern newly created in step 13 and the dependency relationship of the past event pattern acquired from the relational database. . FIG. 7 shows an example of determining event patterns and similar event patterns. For example, even if the performance specifications and action conditions of the aircraft A and the aircraft AA are different, it is determined that they are similar if the types of dependency between the units match. Thus, by comparing the dependency relationship between units, it is determined that the event pattern is similar.

  If a similar event pattern exists, the process proceeds to simulation execution (step 20) in the second and subsequent type steps. If no similar event pattern exists, the second and subsequent simulations are executed (step 19) according to the flow of the first embodiment. In step 19, the processing from step 7 to step 11 described in FIG. 1 is performed.

  In step 20 of FIG. 6, steps 21 to 26 are performed as the second and subsequent executions. As in the first embodiment, the second and subsequent executions are repeatedly executed (number of trials−1) times. In step 21, the execution of steps 22 to 26 is repeated for the number of event patterns created in step 13.

  In step 22 of FIG. 6, in the second and subsequent simulation executions, the event pattern created from the first execution is compared with the past event pattern acquired from the relational database. Here, if the types of dependency relationships between the units match and any of the performance specifications or action conditions of the units having the dependency relationship do not match, it is determined as a similar event pattern, and the process proceeds to step 23. Further, not only the types of dependency relationships but also those in which the performance specifications and action conditions of the units having the dependency relationships all match are determined as the same event pattern, and the process proceeds to step 24.

  In step 23 of FIG. 6, in the execution of the similar event pattern, the simulation is executed only at the time when the dependency relationship occurs for each unit having the dependency relationship between the units due to the stochastic element. The execution result of the similar event pattern varies from trial to trial by the Monte Carlo simulation.

  In step 24 of FIG. 2, in the case of the same event pattern, the execution result of the same event pattern is acquired from the relational database. As a result, the execution result is obtained without executing the same event pattern.

  In step 25 of FIG. 2, the execution results of the second and subsequent event patterns are combined with the area to which the execution results of the execution result template are applied to obtain the simulation execution results of the second and subsequent times.

  In step 26 of FIG. 2, the execution results obtained from the second and subsequent executions are registered in the relational database based on the data model shown in FIG. This is because data can be reused in the second and subsequent simulation execution methods according to the second embodiment.

  Note that dependency relationships between units and execution results are accumulated in a relational database and can be reused when executing simulations based on the same or similar scenarios. This is to reduce the execution time due to repeated execution of the same or similar scenario.

  As described above, the simulation execution method according to the second embodiment compares the similarity between the event pattern previously stored in the database and the event pattern extracted by the extraction process in the loop processing according to the first embodiment ( Step 22) If there is a match, the event pattern stored in the database is combined with the template (Steps 24, 25, 26). If they do not match, the unit that interacts with the stochastic element is localized, The execution unit of the simulation corresponding to the event pattern extracted by the extraction process of the unit is re-executed (step 23), the template is combined with the re-execution result (steps 25 and 26), and the combination result is used. The calculation process of the type step after the next time is repeatedly executed (steps 20 and 21). 210, 200) is intended.

  In the Monte Carlo simulation of a large-scale scenario using the simulation execution method described above, even if some performance specifications and behavioral conditions of the scenario are changed, the relational database accumulated in the past due to the dependency between units appearing in the scenario Similarity is determined from the above, and the simulation result is executed by reusing the execution result for the part where the scenario matches, and for the part where the scenario is different, localizing the units that interact with each other due to the stochastic factor at the time of execution. This reduces the amount of computation and shortens the total execution time.

Embodiment 3 FIG.
A computer system according to Embodiment 3 of the present invention will be described. FIG. 8 is a diagram showing a configuration of a computer system according to the third embodiment. In FIG. 8, the computer system has a system configuration that operates in a distributed simulation environment. A computer system that operates in this distributed simulation environment includes a distributed simulation platform 310, one or more computers 301 connected to the distributed simulation platform, and a simulation management apparatus 302 connected to the distributed simulation platform. The simulation management apparatus 302 controls the execution operation of the simulation based on the scenario, and controls the combination of the execution results. Each computer 301 executes the event pattern set by the simulation management apparatus 302 based on the scenario. The distributed simulation platform is a network platform that connects the simulation management apparatus 302 and each computer 301 and manages data transmission thereof.

  In the distributed simulation environment, the execution time can be shortened by executing the event pattern in parallel for each computer 301 and outputting the execution result.

  1 First simulation execution, 2 Event pattern creation, 3 Event schedule creation, 4 First event pattern execution result creation, 5 Execution result template creation, 6 Database registration, 7 Second simulation onwards Repeated for execution (number of trials-1), repeated for 8 event patterns, repeated for the number of event patterns, 9 executed event patterns, 10 combined event pattern execution results with execution templates, 11 registered in database, 12 1st time Simulation execution, 13 event pattern creation, 14 event schedule creation, 15 first event pattern execution result creation, 16 execution result template creation, 17 database registration, 18 event pattern determination, 19 implementation Second and subsequent simulations of Form 1 Execution, 20 Repeat the second and subsequent simulation executions (number of trials-1), 21 event pattern execution, 22 event pattern similarity / identity determination, 23 similar event pattern execution, 24 database same Acquire event pattern execution result, 25 Combine execution result of similar event pattern or event pattern with execution result template, 26 Register in database, 70, 80, 200, 210 Loop processing, 301 Computer, 302 Simulation management device, 310 Distributed simulation infrastructure.

Claims (2)

A simulation execution method for repeatedly executing a simulation based on a scenario modeling a behavior of a preset unit at a plurality of time steps by a Monte Carlo simulation using a computer,
A first simulation execution process for executing a simulation in the first type step;
An extraction process for identifying a dependency relationship between units appearing in the scenario using the first simulation execution result, and extracting a behavior of a unit that interacts with a stochastic element from the identified dependency relationship as an event pattern; ,
A division process for dividing a part not including the event pattern from the execution unit of the simulation as a template,
The unit that interacts with the stochastic element is localized, the execution unit of the simulation corresponding to the event pattern of the unit is re-executed, the template is combined with the re-execution result, and the next time using the combination result. Loop processing that repeatedly executes the calculation processing of the subsequent type steps,
A simulation execution method comprising:   The loop process compares the similarity between the event pattern stored in the database in advance and the event pattern extracted in the extraction process, and if they match, the event pattern stored in the database is combined with the template, If they do not match, localize the unit that interacts with the stochastic element, re-execute the simulation execution unit corresponding to the event pattern extracted by the extraction process of the unit, and combine the template with the re-execution result The simulation execution method according to claim 1, wherein the calculation process of the type step after the next time is repeatedly executed using the combination result.

Images