Control data intelligence simulation system and method

Abstract

The invention discloses a control data intelligence simulation system and method, and relates to the technical field of data processing, the control data intelligence simulation system is composed of a dynamic complex network model construction module and a multi-agent decision module, the dynamic complex network model construction module is composed of a dynamic data acquisition module, an updating module and a construction module; according to the control digital intelligence simulation system, the dynamic complex network model of the task is constructed to display the dynamically changing task area situation, and the multi-agent decision-making module is constructed, so that the speed and the efficiency of deducing the control index item are high.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a control digital intelligent simulation system and method. Background Art

[0002] The scenario deduction method is an important decision-making and evaluation tool that can help people predict and evaluate various possible results and situations. Decision tree analysis, simulation and Monte Carlo methods are several common scenario deduction methods that are widely used in different fields. Decision tree analysis is a decision support tool based on a tree structure, which is used to solve complex decision problems. It calculates possible results based on different choices and probabilities by breaking down the problem into a series of nodes and branches, and determines the best decision solution. Decision tree analysis is generally applicable to situations with multiple different decision paths and probabilities. Simulation is a method for studying and evaluating system behavior by establishing mathematical models and simulation experiments. It can simulate and test various situations and scenarios in a virtual environment to gain insights into system behavior and possible results. Simulation is generally applicable to situations where multiple variables and complex relationships need to be considered. Monte Carlo method is a numerical calculation method based on random sampling and statistical analysis, which is used to evaluate and predict various possible results and situations. It generates a large number of random samples and uses statistical methods to analyze the samples to infer and predict the behavior and results of the entire system. Monte Carlo method is generally applicable to situations where random factors and multiple variables need to be considered. By choosing appropriate methods and techniques, people can better understand and solve complex decision-making problems and develop the best solutions and strategies. When performing scheme deduction, it is necessary to deduce the strategy. For example, the invention patent application with publication number CN115730630A discloses a control method, device, electronic device and storage medium of an intelligent agent. The specific implementation scheme is: generating a state sequence according to the target state data of multiple intelligent agents, and inputting the state sequence into the intelligent agent strategy model to determine the action sequence according to the output of the intelligent agent strategy model; wherein the action sequence includes the target control action of each intelligent agent belonging to the first object; according to each target control action in the action sequence, each intelligent agent belonging to the first object is controlled. Thus, it can be realized in a multi-intelligent scene that each intelligent agent (such as one's own intelligent agent) belonging to the first object is controlled in action, so that although artificial intelligence can be used to generate a control scheme in the collaborative completion of the task by each intelligent agent belonging to the first object, there is no strategy deduction by using artificial intelligence to quickly obtain the direction of the control index, so a control digital simulation system is proposed. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a control digital simulation system that solves the problems raised in the above background technology.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: a control digital intelligent simulation system, consisting of a dynamic complex network model construction module and a multi-agent decision-making module, wherein the dynamic complex network model construction module consists of a dynamic data acquisition module, an update module and a construction module; The dynamic data acquisition module acquires all entity information participating in the task in real time and sends the acquired entity set to the update module. and the entity collection All data related to the entity; The update module calculates the entity set sent by the dynamic data acquisition module Any entity in With any entity in the existing dynamic complex network model The degree of association between entities is generated based on the degree of association Related entity sequence ; The building blocks determine and The attribute relationship of the edge connecting each entity in and weight generation based on entity A dynamic complex network model with as the central point; The multi-agent decision-making module consists of a data selection module, a graph attention mechanism model layer and a neural network layer; The data selection module, Select from the dynamic complex network model according to the control objective at any time entities, The entities include a target and An intelligent agent will Agent The feature vectors of the target at each moment are input into the graph attention mechanism model layer respectively; The graph attention mechanism model layer includes The graph attention mechanism model, The graph attention mechanism model is Output the completion vector of the target at all times , the output vector of the graph attention mechanism model layer is ; The neural network layer, according to The input vector at time generate Moment control indicator items.

[0005] Optionally, the entity includes multiple agents participating in the task and a target of the task that the multiple agents need to select.

[0006] Optionally, the update module and the build module traverse the entity collection Each entity in generates a dynamic complex network model.

[0007] Optionally, the data selection module includes a hard attention mechanism model, which can select an intelligent agent entity from a dynamic complex network model based on a probability of selecting a target.

[0008] A control digital simulation method, characterized in that it includes S1: a dynamic complex network model construction process and S2 a multi-agent decision-making process, wherein the S1: dynamic complex network model construction process includes S11: dynamic data acquisition; S12: updating model construction; the S2: multi-agent decision-making process includes S21: data selection and calculation; and S22: sorting.

[0009] Optionally, the S1: dynamic complex network model construction process comprises the following steps: S11: Dynamic data acquisition: obtain all entity information participating in the task in real time, and send the acquired entity set to the update module and the entity collection All data related to entities in the task, including multiple agents participating in the task and targets that need to be selected by multiple agents; S12: Update and model building: traverse the entity set sent by the dynamic data acquisition module Any entity in With any entity in the existing dynamic complex network model The degree of association between entities is generated based on the degree of association Related entity sequence ; Building modules, determining and The attribute relationship of the edge connecting each entity in and weight generation based on entity A dynamic complex network model with the center point.

[0010] Optionally, the S2: multi-agent decision-making process comprises the following steps: S21: Data selection: Always select from the dynamic complex network model according to the control task entities, The entities include a target and An intelligent agent, The first group of agents is selected as the first group of agents, and the feature vector of the target selected by each agent in the first group of agents is input into the graph attention mechanism model layer of the first computing module. At the same time, at time t, one more agent entity is selected from the dynamic complex network model according to the control task to replace one agent in the first group of agents to form the second group of agents, and the feature vector of the target selected by each agent in the second group of agents is input into the graph attention mechanism model layer of the second computing module. And so on. The data selection is At any given moment, select an intelligent entity from the dynamic complex network model to replace the first intelligent entity according to the control task. The unreplaced agents in the agent group constitute the Intelligent group, The members of the agent group and The members of the agent groups are different. The feature vector of the target selected by each agent in the agent group is input into the A graph attention mechanism model layer with 4 computational modules; S22: Calculation and sorting: Using the graph attention mechanism model layer includes The graph attention mechanism model, The graph attention mechanism model is Output the completion vector of the target at all times , the output vector of the graph attention mechanism model layer is ,right The output of the calculation module The index items are comprehensively analyzed, sorted according to the comprehensive analysis results, and the intelligent agent group with the best comprehensive result is selected as the final optimal coordinated intelligent agent group.

[0011] Optional, (1) in, represents the activation function, represent The first of the agents The feature vector of the target selected by the agent For The feature vector of the target selected by the agent Contribution of represents the activation function; Represents the parameters from the input layer to the hidden layer of the graph attention mechanism model; represents the parameter matrix; Indicates that and Put it together.

[0012] 7. Optionally, the neural network includes an input layer, a hidden layer and an output layer; The input layer includes Neuron, Neurons The vector input at each moment is: (2) In formula (2), Representative Completion vector The weight matrix of The hidden layer includes a deformation layer and a comparison layer, wherein the deformation layer includes Neuron, Neurons The output at this moment is: (3) In formula (3), Represents the input layer The neurons in the deformation layer The weights between neurons; Representative Neurons Output at the moment, represents the adjustment factor, represents the activation function; The comparison layers include Neuron, The output of a neuron is: (4) In formula (4), is a constant; The output layer includes Neuron, The output of a neuron is: (5) In formula (5), For comparison neurons and the output layer The weights between neurons.

[0013] Optionally, the data selection module includes a hard attention mechanism, which selects intelligent entity from the dynamic complex network model based on the probability of target selection.

[0014] The present invention provides a control digital intelligent simulation system, which has the following beneficial effects: The control digital simulation system and method achieve the ability to display the dynamically changing task area situation by constructing a dynamic complex network model of the task, and achieves fast and efficient deduction of control indicators by constructing a multi-agent decision-making module. The agent decision module selects agents and selected targets from the dynamic complex network model for deduction, and compares the deduction by selecting agents and selected targets from the dynamic complex network model by manual rules, and performs a control efficiency comparison. It is concluded that the control indicators achieved by selecting agents and targets from the dynamic complex network model through the constructed multi-agent decision-making module are better than the control indicators achieved by manual rule control, thereby proving that the multi-agent decision-making model proposed in the present invention can enable multiple agents to cooperate with each other, thereby achieving autonomous decision-making to maximize the efficiency of individual agents, and collaborative selection to achieve the global optimal strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a block diagram of the composition of the control digital intelligent simulation system provided in the first embodiment of the present invention; Figure 2 It is a block diagram of the composition of the dynamic complex network model building module provided in the first embodiment of the present invention; Figure 3 It is a block diagram of the composition of the multi-agent decision-making module provided in the first embodiment of the present invention; Figure 4 It is a block diagram of the composition of the multi-agent decision-making module provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0017] Embodiment 1: See also Figure 1 ,The present invention provides a technical solution: a control digital intelligent simulation system, including a dynamic complex network model building module and a multi-agent decision-making module; Construct a multi-entity dynamic graph neural network through the dynamic complex network model building module, where the entities include multiple agents and targets; Through the multi-agent decision-making module, one target and ( is a positive integer) agents, according to The ability of each agent to achieve the goal output Indicators are provided to provide a basis for decision-making for mission control personnel; The ability of each agent to complete the selected target is determined by the agent's physical performance, the operator's skills, the distance to the target and the environment. The ability of multiple agents to complete the selected target is also related to factors such as the intercommunication and collaboration capabilities among multiple agents. The operators include, for example, drivers and operators, and the environment includes geographical environment, climate environment, electromagnetic environment, etc.; like Figure 2 As shown, the dynamic complex network model building module includes a dynamic data acquisition module, an update module and a building module; The dynamic data acquisition module acquires all entity information participating in the task in real time and sends the acquired entity set to the update module. and the entity collection All data related to entities in the task, including multiple agents participating in the task and multiple targets that need to be selected by the agents; Among them, the dynamic data acquisition module includes a measurement module, an evaluation module and a quantification module; The measurement module is used to measure the distance between one entity and another entity in real time, and to measure its own resources and the operating resources of the intelligent body; the evaluation module is used to evaluate the operating status of the intelligent body, the status of the operator, the intercommunication capability, etc. in real time; the quantification module quantifies the information provided by the measurement module to the evaluation module to facilitate data processing; The update module calculates the entity set sent by the dynamic data acquisition module Any entity in With any entity in the existing dynamic complex network model The degree of association between entities is generated based on the degree of association Related entity sequence ; Building blocks, identify and The attribute relationship of the edge connecting each entity in and weight generation based on entity Dynamic complex network model with central point; update module and construction module traverse entity collection Generate a dynamic complex network model for each entity in; The above update module and construction module traverse the entity collection Generate a dynamic complex network model for each entity in; like Figure 3 As shown, the multi-agent decision-making module includes a data selection module, a graph attention mechanism model layer, and a neural network layer; Among them, the data selection module, Select from the dynamic complex network model according to the control objective at any time entities, The entities include a target and An intelligent agent will Agent The feature vectors of the target at each moment are input into the graph attention mechanism model layer respectively; Among them, the graph attention mechanism model layer includes The graph attention mechanism model, The graph attention mechanism model is Output the completion vector of the target at all times , so the output vector of the graph attention mechanism model layer is ; Neural network layer, according to The input vector at time generate Moment Control indicator items; The data selection module includes a hard attention mechanism model, which selects intelligent entities from the dynamic complex network model according to the probability of selecting the selected target, and selects intelligent entities with higher selection probabilities from the dynamic complex network model in turn.

[0018] The feature vector of the agent includes the quantitative value of the agent's performance, the quantitative value of the personnel's skills, the distance value from the target, the quantitative value of the personnel's status, etc., that is, ; The completion vector of the selected target includes: the quantitative value of the target completion, the quantitative value of the target completion time, etc. The indicator items include the first control indicator, the second control indicator, the third control indicator, the fourth control indicator and the task completion time, etc., expressed as ; (1) In formula (1), represents the activation function, represent The first of the agents The feature vector of the target selected by the agent For The feature vector of the target selected by the agent Contribution of represents the activation function; Represents the parameters from the input layer to the hidden layer of the graph attention mechanism model; represents the parameter matrix; Indicates that and Splice it together; The neural network includes an input layer, a hidden layer, and an output layer. The input layer includes N neurons. Neurons The vector input at each moment is: (2) In formula (2), Representative Completion vector The weight matrix of The hidden layer includes a deformation layer and a comparison layer, wherein the deformation layer includes Neuron, Neurons The output at this moment is: (3) In formula (3), Represents the input layer The neurons in the deformation layer The weights between neurons; Representative Neurons Output at the moment, represents the adjustment factor, represents the activation function; The comparison layers include Neuron, The output of a neuron is: (4) In formula (4), is a constant; The output layer includes Neuron, The output of a neuron is: (5) In formula (5), For comparison neurons and the output layer The weights between neurons.

[0019] In one embodiment, the data selection module includes a hard attention mechanism model, which selects an intelligent agent entity from a dynamic complex network model based on a probability of selecting a selected target.

[0020] This embodiment 1 achieves the ability to display and dynamically change the situation of the task area by constructing a dynamic complex network model that participates in the task; and obtains the advantage of high efficiency in deducing control indicators by constructing a multi-agent decision-making module.

[0021] Embodiment 2: The difference between the second embodiment and the first embodiment is that Figure 4 As shown, the multi-agent decision-making module includes a data selection module, A calculation module and a sorting module; The computing module includes the graph attention mechanism model layer and the neural network layer; data selection is Always select from the dynamic complex network model according to the control task entities, The entities include a target and An intelligent agent, The agents are taken as the first agent group, and the feature vector of the target selected by each agent in the first agent group is input into the graph attention mechanism model layer of the first computing module. At any moment, according to the control task, another intelligent entity is selected from the dynamic complex network model to replace one intelligent entity in the first intelligent entity group to form the second intelligent entity group. The feature vector of the target selected by each intelligent entity in the second intelligent entity group is input into the graph attention mechanism model layer of the second computing module, and so on. At any given moment, select an intelligent entity from the dynamic complex network model to replace the first intelligent entity according to the control task. The unreplaced agents in the agent group constitute the Intelligent group, The members of the agent group and The members of the agent groups are different. The feature vector of the target selected by each agent in the agent group is input into the A graph attention mechanism model layer with 4 computational modules; Among them, the calculation and sorting module uses the graph attention mechanism model layer including The graph attention mechanism model, The graph attention mechanism model is Output the completion vector of the target at all times , so the output vector of the graph attention mechanism model layer is ,right The output of the calculation module Comprehensively analyze the indicators, sort them according to the comprehensive analysis results, and select the intelligent group with the best comprehensive results as the final optimal coordinated intelligent group; By utilizing the multi-agent decision-making module provided in the second embodiment to select agents and selected targets from the dynamic complex network model for deduction, and comparing with manual rules to select agents and selected targets from the dynamic complex network model for deduction, a control efficiency comparison was performed from the aspects of the first control indicator, the second control indicator, the fourth control indicator, and the task completion time, as shown in Table 1 for details.

[0022] Table 1 Indicator Artificially Selected Agents Multi-agent decision module selects agent groups The first control index 60% 83% The second control index 53% 70% The third control index 70% 45% The fourth control index 35% 50% Task completion time About 100 minutes About 45 minutes From the comparison of the contents in the table, it can be seen that the control indicators (including the first control indicator, the second control indicator, the third control indicator, the fourth control indicator and the task completion time) for selecting agents and completing the goals from the dynamic complex network model through the constructed multi-agent decision-making module are better than the control indicators completed by manual rule control. Therefore, the multi-agent decision-making model provided by the present invention enables multiple agents to cooperate with each other, thereby achieving autonomous decision-making to maximize the efficiency of individual agents, and collaborative selection to achieve the global optimal strategy.

[0023] Embodiment three: The present invention provides a technical solution: a control digital simulation method, which corresponds to the control digital simulation system of Example 1 and Example 2, that is, control deduction is performed using the control digital simulation system involved in Example 1 and Example 2.

[0024] Embodiment 4: This embodiment provides a computer storage medium storing a computer program code. The computer program code can be called by a processor to execute the method of the fourth embodiment. The storage medium includes a memory, and the processor includes a CPU.

[0025] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A control digital intelligent simulation system, characterized by: It is composed of a dynamic complex network model construction module and a multi-agent decision-making module, wherein the dynamic complex network model construction module is composed of a dynamic data acquisition module, an update module and a construction module; The dynamic data acquisition module acquires all entity information participating in the task in real time and sends the acquired entity set to the update module. and the entity collection All data related to the entity; The update module calculates the entity set sent by the dynamic data acquisition module Any entity in With any entity in the existing dynamic complex network model The degree of association between entities is generated based on the degree of association Related entity sequence ; The building blocks determine and The attribute relationship of the edge connecting each entity in and weight generation based on entity A dynamic complex network model with as the central point; The multi-agent decision-making module consists of a data selection module, a graph attention mechanism model layer and a neural network layer; The data selection module, Select from the dynamic complex network model according to the control objective at any time entities, The entities include a target and An intelligent agent will Agent The feature vectors of the target at each moment are input into the graph attention mechanism model layer respectively; The graph attention mechanism model layer includes The graph attention mechanism model, The graph attention mechanism model is Output the completion vector of the target at all times , the output vector of the graph attention mechanism model layer is ; The neural network layer, according to The input vector at time generate Moment control indicator items.

2. A control digital intelligent simulation system according to claim 1, characterized in that: The entities include multiple agents participating in the task and targets of the task that the multiple agents need to select.

3. The control digital intelligent simulation system according to claim 1, characterized in that: The update module and the build module traverse the entity collection Each entity in generates a dynamic complex network model.

4. The control digital intelligent simulation system according to claim 1, characterized in that: The data selection module includes a hard attention mechanism model, which can select an intelligent agent entity from a dynamic complex network model based on the probability of selecting a target.

5. A control digitalization simulation method, characterized in that: It includes S1: dynamic complex network model construction process and S2 multi-agent decision-making process, wherein S1: dynamic complex network model construction process includes S11: dynamic data acquisition; S12: update model construction; S2: multi-agent decision-making process includes S21: data selection and calculation; and S22: sorting.

6. A control digitalization simulation method according to claim 1, characterized in that: The S1: dynamic complex network model construction process includes the following steps: S11: Dynamic data acquisition: obtain all entity information participating in the task in real time, and send the acquired entity set to the update module and the entity collection All data related to entities in the task, including multiple agents participating in the task and targets that need to be selected by multiple agents; S12: Update and model building: traverse the entity set sent by the dynamic data acquisition module Any entity in With any entity in the existing dynamic complex network model The degree of association between entities is generated based on the degree of association Related entity sequence ; Building modules, determining and The attribute relationship of the edge connecting each entity in and weight generation based on entity A dynamic complex network model with the center point.

7. A control digitalization simulation method according to claim 1, characterized in that: The S2: multi-agent decision-making process includes the following steps: S21: Data selection: Always select from the dynamic complex network model according to the control task entities, The entities include a target and An intelligent agent, The first group of agents is selected as the first group of agents, and the feature vector of the target selected by each agent in the first group of agents is input into the graph attention mechanism model layer of the first computing module. At the same time, at time t, one more agent entity is selected from the dynamic complex network model according to the control task to replace one agent in the first group of agents to form the second group of agents, and the feature vector of the target selected by each agent in the second group of agents is input into the graph attention mechanism model layer of the second computing module. And so on. The data selection is According to the control task, one intelligent entity is selected from the dynamic complex network model to replace the first intelligent entity. The unreplaced agents in the agent group constitute the Intelligent group, The members of the agent group and The members of the agent groups are different. The feature vector of the target selected by each agent in the agent group is input into the A graph attention mechanism model layer with 4 computational modules; S22: Calculation and sorting: Using the graph attention mechanism model layer includes The graph attention mechanism model, The graph attention mechanism model is Output the completion vector of the target at all times , the output vector of the graph attention mechanism model layer is ,right The output of the calculation module The index items are comprehensively analyzed, sorted according to the comprehensive analysis results, and the intelligent agent group with the best comprehensive result is selected as the final optimal coordinated intelligent agent group.

8. The control digital intelligent simulation system according to claim 1, characterized in that: （1） in, represents the activation function, represent The first of the agents The feature vector of the target selected by the agent For The feature vector of the target selected by the agent Contribution of represents the activation function; Represents the parameters from the input layer to the hidden layer of the graph attention mechanism model; represents the parameter matrix; Indicates that and Put it together.

9. A control digitalization simulation method according to claim 1, characterized in that: The neural network comprises an input layer, a hidden layer and an output layer; The input layer includes Neuron, Neurons The vector input at each moment is: （2） In formula (2), Representative Completion vector The weight matrix of The hidden layer includes a deformation layer and a comparison layer, wherein the deformation layer includes Neuron, Neurons The output at this moment is: （3） In formula (3), Represents the input layer The neurons in the deformation layer The weights between neurons; Representative Neurons Output at the moment, represents the adjustment factor, represents the activation function; The comparison layers include Neuron, The output of a neuron is: （4） In formula (4), is a constant; The output layer includes Neuron, The output of a neuron is: （5） In formula (5), For comparison neurons and the output layer The weights between neurons.

10. The control digitalization simulation method according to claim 1, characterized in that: The data selection module includes a hard attention mechanism, which selects intelligent agent entities from a dynamic complex network model based on the probability of target selection.

11. A control digitalization simulation method according to claim 1, characterized in that: The control digital intelligent simulation system described in any one of claims 1 to 4 is used for control.

Images