Complex intelligent decision-making system scene-level evaluation method and device based on agent migration

Abstract

The invention relates to the field of electric digital data processing, in particular to a complex intelligent decision-making system scene level evaluation method and device based on agent migration, and the method comprises the following steps: generating an evaluation scene graph based on a knowledge graph, the scene graph comprising an evaluation scene covering an evaluation target of a complex intelligent decision-making system; the evaluation agent actively walks in the evaluation scene graph according to a path planned by a path planning algorithm, evaluates the complex intelligent decision-making system through the evaluation data set, analyzes an evaluation result through an evaluation tool and outputs an evaluation index; and calculating the coverage rate of the assessment scene in the assessment scene graph after the current round of assessment agent migration is completed, and if the coverage rate is smaller than a coverage rate threshold value, rejecting the traversed assessment scene in the assessment scene graph, and performing the next round of active migration of the assessment agent. The method has the effects of improving the comprehensiveness of the assessment scene and the execution efficiency of the assessment scene.

Description

Technical Field

[0001] The present invention relates to the field of electronic digital data processing, and in particular to a scene-level evaluation method and device for a complex intelligent decision-making system based on agent wandering. Background Art

[0002] With the rapid development of artificial intelligence technology, technologies such as intelligent decision-making, intelligent diagnosis, expert systems and embodied intelligence have penetrated into various fields. As a complex intelligent system, the intelligent decision-making system includes multiple parts such as intelligent perception module, intelligent auxiliary decision module and intelligent execution module.

[0003] Ordinary intelligent decision-making systems usually only perform intelligent decisions for a single function. When evaluating ordinary intelligent decision-making systems, a relatively comprehensive evaluation can be achieved by using software batch testing. However, when facing complex intelligent decision-making systems, complex intelligent decision-making systems contain multiple intelligent modules, such as situation analysis, information fusion, abnormal event detection, multi-criteria decision analysis, etc. Each intelligent module is equivalent to an ordinary intelligent decision-making system. At this time, if the software batch testing method is used to evaluate the complex intelligent decision-making system, it is difficult to achieve the evaluation of all intelligent modules. Therefore, the evaluation scenario design method will be adopted to enable the complex intelligent decision-making system to be evaluated in the evaluation scenario.

[0004] Traditional intelligent decision-making system evaluation methods usually rely on manually designed evaluation scenarios and test cases. However, this method has many shortcomings. For example, manually designed scenarios may be subjective and limited, and cannot fully cover all key evaluation links of complex intelligent decision-making systems. In addition, manually designed evaluation scenarios are inefficient during execution and are difficult to meet actual application needs. Summary of the invention

[0005] In order to solve the problems of subjectivity in evaluation scenario design and low efficiency in the execution of evaluation scenarios, the present invention provides a scenario-level evaluation method and device for complex intelligent decision-making systems based on agent wandering.

[0006] The present invention provides a scenario-level evaluation method for complex intelligent decision-making systems based on agent wandering, which adopts the following technical solutions: A scenario-level evaluation method for complex intelligent decision-making systems based on agent wandering includes the following steps: Generate an evaluation scenario graph based on the knowledge graph, the scenario graph includes evaluation scenarios covering the evaluation objectives of the complex intelligent decision-making system; an evaluation agent carrying an evaluation data set and an evaluation tool actively wanders in the evaluation scenario graph, the evaluation agent evaluates the intelligent module of the complex intelligent decision-making system through the evaluation data set to obtain an evaluation result, and the evaluation tool analyzes the evaluation result and outputs an evaluation indicator; After the evaluation agent completes the current round of roaming, the coverage rate of the evaluation scenes in the evaluation scene graph is calculated. If the coverage rate is less than the coverage rate threshold, the evaluation scenes traversed in the evaluation scene graph are removed, and the next round of active roaming of the evaluation agent is performed.

[0007] In a specific implementation scheme, before the evaluation agent actively wanders in the evaluation scene graph, the wandering path of the evaluation agent is planned by a path planning algorithm so that the wandering path of the evaluation agent includes as many nodes in the evaluation scene graph as possible.

[0008] In a specific implementation scheme, the path planning algorithm plans the walking path of the evaluation agent based on the evaluation task with the goal of fully triggering the intelligent module of the complex intelligent decision-making system.

[0009] In a specific implementation scheme, before the path planning algorithm plans the wandering path of the evaluation agent, it calculates the weight of each node in the evaluation scene graph.

[0010] In a specific possible implementation scheme, after the path planning algorithm plans the wandering path of the evaluation agent, the evaluation scene graph is cropped to generate an evaluation scene subgraph for the evaluation agent to wander.

[0011] In a specific possible implementation scheme, when the evaluation scenario graph is generated, the completeness of the evaluation scenario graph with respect to the knowledge graph is calculated, and the completeness is compared with a preset completeness threshold. If the completeness is less than the completeness threshold, it indicates that the evaluation scenario needs to continue to be generated.

[0012] In a specific implementation scheme, there are multiple evaluation agents in the evaluation scene graph, and before the evaluation agents actively wander in the evaluation scene graph, computing resources are scheduled using a resource scheduling algorithm.

[0013] In a specific implementation scheme, the resource scheduling algorithm allocates the resources and working hours consumed by the evaluation agent's wandering as follows: Consider each of the evaluation agents as a particle and generate an initial population; According to the evaluation task, calculating the fitness of each particle; screening the particles for crossover and mutation; Update the position and velocity of the particle until the optimal evaluation agent walking plan is obtained.

[0014] In a specific implementation scheme, when the evaluation agent roams in the evaluation scene graph, it first evaluates each intelligent module of the complex intelligent decision-making system, and then conducts a comprehensive evaluation of the complex intelligent decision-making system.

[0015] The present invention also provides a scene-level evaluation device for a complex intelligent decision-making system based on agent wandering, which adopts the following technical solution: A scene-level evaluation device for complex intelligent decision-making systems based on agent wandering, comprising: one or more processors; memory; And one or more computer programs, wherein one or more of the computer programs are stored in the memory, and one or more of the computer programs include instructions, which, when executed by the processor, enable the scene-level evaluation device for the complex intelligent decision-making system based on agent wandering to perform the above-mentioned scene-level evaluation method for the complex intelligent decision-making system based on agent wandering.

[0016] In summary, the present invention has the following beneficial effects: 1. Automatically generate evaluation scenario diagrams through knowledge graphs, comprehensively covering the key evaluation links of complex intelligent decision-making systems to make up for possible omissions in human subjective design.

[0017] 2. Based on the active wandering ability of the evaluation agent, data collection of complex intelligent decision-making systems is realized to solve the problem of low efficiency in the execution of evaluation tasks.

[0018] 3. Based on the genetic particle swarm resource scheduling method, the resource scheduling and allocation of multiple evaluation agents in the evaluation process of complex intelligent decision-making systems is realized, solving the problem of insufficient resources for multi-agent evaluation tasks.

[0019] 4. After completing the current round of evaluation, calculate the coverage of the evaluation scenarios traversed by the evaluation agent, and traverse the scenarios that have not been traversed by comparing with the preset coverage threshold, so as to conduct a comprehensive evaluation of the complex intelligent decision-making system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is an overall flow chart of the scenario-level evaluation method for complex intelligent decision-making systems based on agent wandering.

[0021] Figure 2This is an architectural diagram of the scenario-level evaluation method for complex intelligent decision-making systems based on agent wandering. DETAILED DESCRIPTION

[0022] The following combination Figure 1-Figure 2 , the present invention is described in further detail.

[0023] Scenario refers to a comprehensive expression of a series of interrelated events, activities and situations within a specific time and space. It includes elements such as physical environment, participants, interaction methods and background information, and is used to describe a complete situation with internal logic. Scenario Elements refer to the basic components that make up a scene. Elements work together to form a complete situation with internal logic, usually including the physical environment, participants, interaction methods and background information.

[0024] Scenario relationships refer to the connection and interaction between different scenarios, describing how one scenario transitions to another, as well as the logical and causal relationships between them. These relationships can be temporal sequence, causal relationship, conditional triggering, etc.

[0025] Scenario Graph refers to a graph structure formed by organizing multiple interrelated scenarios according to certain association relationships during the operation of the target object. Each node represents a specific scenario, and the edges between scenario nodes represent the association relationship between scenarios. Scenario graphs are used to describe and analyze the state and behavior changes of target objects under different conditions.

[0026] The present invention is further described in detail below, taking the unmanned driving system as the complex intelligent decision-making system to be evaluated. The unmanned driving system has intelligent modules such as temperature sensing module, distance sensing module, weather sensing module, action decision module, and execution module. The temperature sensing module senses the temperature inside the vehicle; the distance module measures the distance between the unmanned driving car and the edge of the lane, the distance between the front vehicle, and the distance between the front and rear obstacles; the weather sensing module senses the current season and weather; the decision module obtains data information from the temperature sensing module, distance sensing module, weather sensing module, etc., as well as the road condition information of the current road, the current distance to the destination, etc., plans the driving path and driving status of the unmanned driving car, and generates driving plans and driving decisions for execution by the execution module.

[0027] The scenario-level evaluation method of complex intelligent decision-making systems based on agent wandering includes the following steps: S100, generating an evaluation scenario graph based on the knowledge graph.

[0028] A knowledge graph is a structured semantic knowledge base that is used to quickly describe concepts and their relationships in the physical world.

[0029] The knowledge graph effectively processes, handles and integrates the data of complex documents into simple and clear "entity, relationship, entity" triples, and finally aggregates field knowledge such as events and time series information, thereby achieving rapid response and reasoning of knowledge. The evaluation object is the unmanned driving system, so the knowledge graph of the unmanned driving system and related fields is selected to generate a graph-structured evaluation scenario graph, and each evaluation scenario is a node in the evaluation scenario graph.

[0030] As shown in Table 1 below, some knowledge in the knowledge graph is presented. Based on the knowledge in Table 1, different driving scenarios that the unmanned driving system has to deal with can be constructed. The driving scenario is the evaluation scenario.

[0031] Table 1

[0032] In the evaluation scenario diagram constructed by the knowledge in Table 1, different evaluation scenario designs and different selected routes result in a variety of decisions that can be made. When the unmanned driving system generates driving plans and driving decisions, the intelligent modules called are different; and after the unmanned driving system generates driving plans and driving decisions, the driving plans and driving decisions executed by the execution module and the components such as motors acted by the execution module are also different.

[0033] By changing the knowledge when constructing the evaluation scenarios according to the knowledge graph, new scenarios are generated, so as to construct as many evaluation scenarios as possible, that is, to generate as many nodes as possible in the evaluation scenario graph. The change of knowledge forms the scenario association relationship between the evaluation scenarios, that is, the connection between the nodes in the evaluation scenario graph.

[0034] Taking the data in Table 1 as an example, we first construct a winter normal weather road driving scenario. In this evaluation scenario, the unmanned driving system automatically makes decisions to plan the vehicle's route based on the simulated environment so that the vehicle can drive to the destination. Then, we construct a winter severe weather road driving scenario. The difference between the winter severe weather road driving scenario and the winter normal weather road driving scenario is that the weather changes from sunny to sleet to heavy snow, which causes the strategy executed by the unmanned driving system to change from K1 to K3, and the intelligent module called by the unmanned driving system also changes accordingly. By analogy, by gradually changing the knowledge in the evaluation scenario by controlling the variables, a new evaluation scenario is constructed, thereby constructing as many evaluation scenarios as possible, so that the constructed scenarios can cover all possible combinations of knowledge in the knowledge graph as much as possible. In different evaluation scenarios, as the knowledge changes, the intelligent modules called by the unmanned driving system are different, and the performance of the intelligent modules is also different. As many evaluation scenarios as possible can enable the unmanned driving system to call as many intelligent modules as possible and make the intelligent modules perform their full performance as much as possible, so that when the unmanned driving system is evaluated in the future, the performance of the unmanned driving system can be comprehensively evaluated.

[0035] The evaluation scenario in the evaluation scenario diagram represents the completeness of the effectiveness of the intelligent module of the complex intelligent decision-making system. The coverage of the evaluation scenario can be evaluated by presetting the completeness threshold and comparing the completeness with the completeness threshold. For example, if the completeness is greater than or equal to the completeness threshold, it indicates that the generated evaluation scenario can fully exert the effectiveness of the intelligent module of the unmanned driving system, and the generation of the evaluation scenario can be stopped for subsequent processing; otherwise, it indicates that there is an intelligent module that has not exerted its effectiveness or has not fully exerted its effectiveness, and the generation of the evaluation scenario continues to enable the intelligent module of the unmanned driving system to further exert its effectiveness, thereby greatly reducing the subjectivity and limitations of the evaluation scenario design.

[0036] S200, the evaluation agent actively walks to evaluate the complex intelligent decision-making system. The walking path of the evaluation agent is planned by a path planning algorithm.

[0037] According to the complexity of the evaluation scene, the weight of each node in the evaluation scene graph is calculated. The higher the complexity of the evaluation scene, the greater the weight of the node. The complexity of the evaluation scene depends on the number of scene elements and the position of the node in the evaluation scene graph. The more scene elements there are, the greater the out-degree and in-degree of the current node in the context environment, and the closer the distance to the end node, the higher the complexity of the evaluation scene.

[0038] The evaluation agent is an agent with the ability to actively wander. In the evaluation scene graph, the evaluation agent generates the optimal active wandering path in the evaluation scene graph through a path planning algorithm according to the preset evaluation task, thereby completing the evaluation task.

[0039] Multiple evaluation agents are set in the evaluation scenario graph. Each evaluation agent carries the corresponding evaluation data set, evaluation tool and evaluation standard set. The evaluation agent evaluates each intelligent module of the complex intelligent decision-making system through the evaluation data set to obtain the evaluation results. The evaluation tool analyzes the evaluation results in combination with the evaluation standard set and outputs the evaluation indicators. After the evaluation indicators are dimensionless, they are aggregated to obtain a comprehensive evaluation of the complex intelligent decision-making system.

[0040] The path planning algorithm is based on the evaluation scene graph and the weight of the nodes. It aims to fully utilize the efficiency of the intelligent module of the complex intelligent decision-making system in the evaluation scene. It plans the walking path of each evaluation agent and calculates a path with a starting point and an end point. This path passes through multiple evaluation scene nodes through the connection of the graph structure and is the optimal or most feasible path for the evaluation agent to complete the evaluation task. The evaluation agent walks between the evaluation scene nodes along the path, evaluates the complex intelligent decision-making system, and reaches the end of the path to complete the evaluation task.

[0041] Specifically, in order to obtain the evaluation scenario traversal path, that is, the evaluation agent's wandering path, from the evaluation scenario roadmap, the evaluation scenario roadmap can be regarded as a Hidden Markov Model (HMM), and the Viterbi Algorithm can be used to obtain the optimal evaluation scenario path.

[0042] However, the goal of the typical Viterbi algorithm is to obtain the traversal path with the shortest path and the largest joint probability. Therefore, in order to encourage the path to contain as many nodes as possible, a penalty function is designed to penalize those paths containing fewer nodes.

[0043] The path planning method of the evaluation scenario roadmap of complex intelligent systems is as follows: (1) Build an evaluation scenario roadmap: Represent the evaluation scenario roadmap as a directed graph ,in It is a collection of evaluation scenario nodes, representing the evaluation scenario; is a set of edges, representing scene associations; is the node weight, which represents the importance of the scene node.

[0044] Each node It contains several evaluation factors (such as time, location, weather, etc.), and each evaluation factor has a probability of occurrence.

[0045] (2) Define the hidden Markov model: State Collection Corresponds to the set of evaluation scene nodes .

[0046] Initial state probability Represents the probability that the system starts from a certain scenario.

[0047] State transition probability matrix set A collection of probabilities representing the transition from one scenario to another.

[0048] Observation probability matrix set It represents the set of probabilities of observing certain evaluation factors in a certain scenario.

[0049] (3) Viterbi algorithm: The Viterbi algorithm is used to find the most likely state sequence (i.e., evaluation scenario path) given an observation sequence. Indicates at time Status is The probability of the optimal path is:

[0050] Among them, N is the total number of states, that is, the total number of nodes, , ; From the state Transfer to state The probability of The node moves to the The probability of a node, ; is in state Observed Assessment Elements The probability of The evaluation factors are observed in the evaluation scenario of nodes The probability of .

[0051] (4) Penalty function: In order to traverse more nodes, a penalty function is introduced , which is used to penalize paths containing fewer nodes. Let the path length be The total number of nodes is , then the penalty function can be defined as:

[0052] In the Viterbi algorithm, a penalty function is added when updating the path probability, resulting in the formula: .

[0053] It is easy to understand that each time the path planning algorithm completes the path planning of an evaluation agent, it clips the evaluation scene graph once to generate an evaluation scene subgraph of the evaluation agent's walking path for the evaluation agent to walk. The clipped evaluation scene subgraph is removed from the evaluation scene graph, and then the next path planning is performed, and a new evaluation scene subgraph is clipped from the remaining evaluation scene graph, and the cycle repeats.

[0054] In the evaluation scenario graph of the unmanned driving system, the evaluation tasks of the evaluation agent include the intelligence of the unmanned driving system when the weather changes, the intelligence of the unmanned driving system when the route changes, and the intelligence of the unmanned driving system when the road conditions change. Each evaluation agent corresponds to one or more evaluation tasks. For example, the evaluation task of the evaluation agent is the intelligence of the unmanned driving system when the weather changes, and the evaluation scenario graph is constructed by Table 1. The wandering path constructed by the path planning algorithm is: winter normal weather road driving scene-winter rainy weather road driving scene-winter snowy weather road driving scene. The evaluation agent wanders in these three scenes in turn and evaluates the intelligent modules of the unmanned driving system. After completing the evaluation of each evaluation scene, a comprehensive evaluation is output. When the evaluation agent evaluates the unmanned driving system in the scene, each intelligent module called by the unmanned driving system is evaluated and the corresponding evaluation results are output. The evaluation results are analyzed using the evaluation tool and the evaluation indicators are output. Multiple evaluation indicators are dimensionless and aggregated, and the comprehensive evaluation of the unmanned driving system in the evaluation scene is output. After completing the wandering of the three scenes, the three comprehensive evaluations are aggregated to obtain a comprehensive evaluation result.

[0055] Since there are multiple evaluation agents in the evaluation scene graph, a large number of evaluation paths will be generated, which will occupy a lot of computing resources. Scheduling the resources of the evaluation agent through the resource scheduling algorithm can reduce the waste of computing resources and make the computing resource allocation during the evaluation agent's wandering and evaluation process more reasonable.

[0056] For ease of understanding, the resource scheduling algorithm of the present invention is specifically a genetic particle swarm algorithm. In some other embodiments, it may be an ant colony algorithm, a hybrid genetic algorithm, etc.

[0057] Specifically, each evaluation agent is regarded as a particle, and the initial population is generated randomly or based on experience and initialized.

[0058] According to the requirements of the evaluation task, the fitness of each particle is evaluated based on the fitness evaluation indicators, which include the consumed working hours, consumed resources, etc.

[0059] Particles with higher fitness are selected for crossover and mutation to generate a new population. The particle selection method can be roulette selection, tournament selection, etc., and the crossover method can be single-point crossover, multi-point crossover, etc.

[0060] Update the particle's position and velocity.

[0061] By formula:

[0062] Calculate the current position of the particle. In the above formula, Indicates Particles in The location at the moment, Indicates The particle in The speed of time, Indicates The particle in The position at the moment.

[0063] By formula:

[0064] Calculate the current velocity of the particle. In the above formula, represents the inertia weight, and is the acceleration constant, and is a random constant, Indicates The individual optimal position of each particle, represents the global optimal position.

[0065] By continuously updating the position and speed of particles, the movement path and resource consumption of particles are optimized, the movement rate of particles is increased and the resource consumption of particles is reduced. That is, the wandering efficiency of the evaluation agent is improved and the resource consumption of the evaluation agent during wandering is reduced, thus obtaining the optimal resource scheduling solution with maximum execution efficiency of the evaluation task and minimum resource consumption.

[0066] According to the optimal evaluation agent walking plan, resources are scheduled and allocated for the evaluation agent.

[0067] S300, calculating the coverage rate of the current round.

[0068] The coverage rate is the ratio of the evaluation scenes that the evaluation agent has actively traversed in the current round to all the evaluation scenes in the evaluation scene graph. The coverage rate is compared with the pre-input coverage rate threshold. If the coverage rate is greater than or equal to the coverage rate threshold, it indicates that the evaluation of the complex intelligent decision-making system is comprehensive enough, and a multi-dimensional comprehensive evaluation of the complex intelligent decision-making system is performed, and finally the multi-dimensional comprehensive evaluation results are output; otherwise, the evaluation scenes that have been traversed in the evaluation scene graph are eliminated, and step S200 is repeated to perform a new round of active roaming of the evaluation agent.

[0069] The present invention also discloses a scene-level evaluation device for a complex intelligent decision-making system based on agent wandering, comprising one or more processors; a memory; And one or more computer programs, wherein the one or more computer programs are stored in a memory, and the one or more computer programs include instructions, which, when executed by a processor, enable the scene-level evaluation device for a complex intelligent decision-making system based on agent wandering to perform the above-mentioned scene-level evaluation method for a complex intelligent decision-making system based on agent wandering.

[0070] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A scenario-level evaluation method for complex intelligent decision-making systems based on agent wandering, characterized by: The steps include: Generate an evaluation scenario graph based on the knowledge graph, where the scenario graph includes evaluation scenarios covering the evaluation objectives of the complex intelligent decision-making system; An evaluation agent carrying an evaluation data set, an evaluation tool, and an evaluation standard set actively wanders in the evaluation scene graph, the evaluation agent evaluates the intelligent module of the complex intelligent decision-making system through the evaluation data set to obtain an evaluation result, and the evaluation tool analyzes the evaluation result in combination with the evaluation standard set to output an evaluation indicator; After the evaluation agent completes the current round of roaming, the coverage rate of the evaluation scenes in the evaluation scene graph is calculated. If the coverage rate is less than the coverage rate threshold, the evaluation scenes traversed in the evaluation scene graph are removed, and the next round of active roaming of the evaluation agent is performed.

2. The method according to claim 1, characterized in that: Before the evaluation agent actively wanders in the evaluation scene graph, the wandering path of the evaluation agent is planned by a path planning algorithm so that the wandering path of the evaluation agent includes as many nodes in the evaluation scene graph as possible.

3. The method according to claim 2, characterized in that: The path planning algorithm plans the walking path of the evaluation agent according to the evaluation task with the goal of fully utilizing the effectiveness of the intelligent module of the complex intelligent decision-making system.

4. The method according to claim 2, characterized in that: Before the path planning algorithm plans the wandering path of the evaluation agent, the weight of each node in the evaluation scene graph is calculated.

5. The method according to claim 2, characterized in that: After the path planning algorithm plans the wandering path of the evaluation agent, the evaluation scene graph is cropped to generate an evaluation scene subgraph for the evaluation agent to wander.

6. The method according to claim 1, characterized in that: When the evaluation scene graph is generated, the completeness of the evaluation scene graph is calculated, and the completeness is compared with a preset completeness threshold. If the completeness is less than the completeness threshold, it indicates that the evaluation scene needs to be further generated.

7. The method according to claim 1, characterized in that: There are multiple evaluation agents in the evaluation scene graph. When the evaluation agents actively wander in the evaluation scene graph, computing resources are scheduled using a resource scheduling algorithm.

8. The method according to claim 7, characterized in that: The resource scheduling algorithm allocates the resources and working hours consumed by the evaluation agent's wandering as follows: Consider each of the evaluation agents as a particle and generate an initial population; According to the evaluation task, calculating the fitness of each particle; screening the particles for crossover and mutation; The position and velocity of the particle are updated until an optimal resource scheduling solution is obtained.

9. The method according to claim 1, characterized in that: When the evaluation agent moves in the evaluation scene graph, it first evaluates each intelligent module of the complex intelligent decision-making system, and then conducts a comprehensive evaluation of the complex intelligent decision-making system.

10. A scene-level evaluation device for complex intelligent decision-making systems based on agent wandering, characterized in that: include: one or more processors; Memory; And one or more computer programs, wherein one or more of the computer programs are stored in the memory, and one or more of the computer programs include instructions, which, when executed by the processor, enable the scene-level evaluation device for the complex intelligent decision-making system based on agent wandering to perform the scene-level evaluation method for the complex intelligent decision-making system based on agent wandering as described in any one of claims 1-9.

Images