Adaptive agent modeling method, system and product for digital government service

Abstract

The application belongs to the field of digital government service system modeling, and provides a self-adaptive intelligent agent modeling method, system and product for digital government service, and the technical scheme is: obtaining business processing data of intelligent agents of various government departments; based on the business processing data of the intelligent agents of various government departments, a digital government service model is constructed; according to different response variables, the modeling characteristics corresponding to the digital government service model are matched, and according to the correlation between the response variables and the corresponding modeling characteristics, events leading to significant changes in the response variables are determined. This framework captures and models the micro individual characteristics and emergence mechanism in the digital government service system, which not only helps to standardize the modeling process, but also eliminates the gap between the complex characteristics modeling of the intelligent agents and the system of the government departments and the application of AI technology.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of digital government service system modeling, and particularly relates to a self-adaptive agent modeling method, system and product for digital government service. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] Digital government service is a new form of government service supported by artificial intelligence technology. Entities with different intelligence, such as government departments, business personnel, and software robots, participate in co-evolution and gradually form a complex social system, i.e., a digital government service system. In these systems, self-adaptive individuals and groups can spontaneously form organized structures and behavior patterns through interaction and feedback mechanisms, thereby providing efficient services. This self-organizing ability enables the system to self-regulate and adapt to changes in the demand environment. Research on digital government service systems helps to design and improve social policies and organizational management, and helps to understand the role of various factors and relationships in the system. Therefore, it is increasingly attracting the attention of researchers.

[0004] The complexity of digital government service system modeling stems from the diversity of its business and demand, the multi-link nature of its business processes, and the uncertainty and dynamics of the system. Its operation architecture can be modeled into three layers: (1) social layer: self-adaptive service agents (i.e., government department individual agents) provide services and achieve sustainable development through autonomous decision-making and regulatory measures; (2) physical layer: functions based on physical devices are encapsulated into services and uploaded to the cloud, and diversified intelligent devices achieve crowdsourcing and crowdfunding through collaboration; (3) information layer: emphasizes the complex relationships and interactions between self-adaptive service agents. Since complex adaptive systems involve human and social factors (such as changes in demand, competition and cooperation between departments), accurately modeling self-adaptive government department agents (Agents) and the complex characteristics of artificial intelligence society is crucial for effective analysis and governance of digital government service systems.

[0005] Existing models using complex networks and activity-based models (ABM) to model and analyze digital government service systems often focus on specific technical implementations or isolated characteristics. However, as social theory and computer technology further integrate into ABM research, and digital government service systems become increasingly complex, the modeling process often involves various characteristics at the individual, social, and system levels. However, existing modeling methods are difficult to accurately represent these complex characteristics, and there is a gap between existing technology and complex feature modeling. SUMMARY

[0006] To solve at least one technical problem in the background art, the present application provides a self-adaptive intelligent agent modeling method, system and product for digital government service, which models the autonomy, evolution, interaction and emergence characteristics of artificial intelligence society from self-adaptive department individual agent, according to the actual intelligent degree of digital government service, matches the corresponding characteristics, and improves the efficiency of government service governance.

[0007] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0008] The first aspect of the present application provides a self-adaptive intelligent agent modeling method for digital government service, comprising the following steps:

[0009] Obtaining the business processing data of each government department intelligent agent;

[0010] Based on the business processing data of each government department intelligent agent, a digital government service model is constructed; wherein the digital government service model comprises multiple layers, each layer models the characteristics of different complexity of the subject and system; wherein L1 layer models the basic characteristics of the government department intelligent agent, L2 layer models the evolution characteristics of the government department intelligent agent based on the modeling characteristics of L1 layer, L3 layer models the interaction characteristics between each government department intelligent agent based on the modeling characteristics of L2 layer, and L4 layer models the emergence characteristics of AI society based on the modeling characteristics of L3 layer;

[0011] Determining the factors affecting government service, matching the corresponding modeling characteristics of the digital government service model according to different influencing factors, and determining the events leading to significant changes in the influencing factors according to the association between different influencing factors and corresponding modeling characteristics.

[0012] Further, L1 layer models the basic characteristics of the government department intelligent agent, including:

[0013] Obtaining the state information and environment information list of the government department intelligent agent itself;

[0014] Determining the information type and memory requirement that the government department intelligent agent needs to store and retrieve, and saving the state information and environment information list of the government department intelligent agent itself as a "state-action" pair;

[0015] According to the current "state-action" pair, the intelligent agent takes corresponding actions to complete the task or adjust its own state, and makes decisions;

[0016] According to the decision, the state information and environment information list data of the government department intelligent agent are updated to obtain the autonomous characteristic structured representation.

[0017] Furthermore, the process of modeling the evolutionary characteristics of government agency intelligent agents at the L2 layer includes:

[0018] Based on the fundamental characteristics of a single government department intelligent agent, the genetic evolution of the government department intelligent agent is modeled.

[0019] Information about neighboring government department agents is obtained. The capability attributes of each government department agent are treated as a knowledge point. Each knowledge point is represented as multiple sectors. The optimal sector is determined based on the agent's attributes. Based on the combination of the optimal sectors, a learning reference target is determined for imitation learning.

[0020] Define the learning area;

[0021] Calculate the survivability of all agents in each department, combine the survivability of all agents in each department to calculate the fitness value of all agents in each department, select the agents with the highest fitness value in the first N periods of the learning region, and broadcast the knowledge points of these agents for other agents to learn.

[0022] Furthermore, the process of modeling the interaction characteristics between government agency agents at the L3 layer includes:

[0023] Determine the initial communication network topology between the intelligent agents of each government department;

[0024] If it is collaboration between individual government departments, the weights between nodes in the communication network topology are updated by recording the relationship list of the department agents that established the connection.

[0025] If it is a collaboration between an individual government department and a collaborative alliance, determine the collaboration intention of the government department's agent and choose whether to join the collaborative alliance;

[0026] Calculate the profit distribution during the collaboration process and decide whether to adopt a negotiation mechanism based on its own situation.

[0027] Furthermore, the profit distribution calculation formula in the collaboration process is as follows:

[0028]

[0029] In the formula, Profit(i) is the Agent. i The revenue obtained after participating in the collaboration; Cost(i) is the Agent's... i Total cost of participating in collaboration; Cost total It represents the total cost invested in processing the order; Value(i) is the Agent. i The normalized effective value created through participation and collaboration; Value total It is the total effective value created during the order processing process; w a It is the profit distribution coefficient; profittotal is the total profit obtained when the order is completed.

[0030] Further, the process of L4 layer modeling the emergent features of AI society includes:

[0031] Defining the influence of the environment on the government department agent;

[0032] The government department agent selects an action according to the current state, interacts with the environment, and obtains an immediate reward and feedback for the next state;

[0033] Combined with the reward and feedback, the constraints that the government department agent considers before making a decision are constantly adjusted according to the purpose and state of the system operation, that is, the intervention strategy.

[0034] Further, the autonomous feature structure is structured as:

[0035] L1(Agent) = <R, S t ,E t ,V t ,M t ,D t >,

[0036] The evolutionary feature structure of the government department agent is structured as:

[0037] L2(Agent) = <L1(Agent), F, Y t >,

[0038] The interaction feature structure of the government department agent is structured as:

[0039] L3(Society) = <L2(Agent), C t ,G t ,P t ,TS>,

[0040] The emergent feature structure of the AI society is structured as:

[0041] L4(System) = <L3(Society), Env., Q t ,N t >,

[0042] Wherein, R is the attribute of Agent to maintain stability, S t is the dynamic ability attribute of Agent changing over time, E t is the perception ability of agent, V t is the reaction ability of Agent, M t is the memory ability of Agent, D t is the decision-making ability of Agent, F is the prior condition of Agent learning and evolution, Yt is the learning ability of Agent, C t is the communication rule between Agents, G t is the cooperation rule between Agents, P t is the competition rule between Agents, TS represents the system topology structure, and Env. is the interaction between the environment and the department Agent; Q t represents the feedback mechanism from the system to the Agent, N t is the system intervention strategy adopted for stable development of the system.

[0043] Further, the factors affecting the government service include system performance indicators and customer satisfaction.

[0044] The second aspect of the present application provides an adaptive intelligent agent modeling system for digital government service, comprising:

[0045] A data acquisition module is configured to acquire business processing data of each government department intelligent agent.

[0046] A digital government service modeling module is configured to construct a digital government service model based on the business processing data of each government department intelligent agent; wherein the digital government service model comprises multiple layers, and each layer models the characteristics of the main body and the system with different complexities; wherein L1 layer models the basic characteristics of the government department intelligent agent, L2 layer models the evolution characteristics of the government department intelligent agent based on the modeling characteristics of L1 layer, L3 layer models the interaction characteristics between each government department intelligent agent based on the modeling characteristics of L2 layer, and L4 layer models the emergent characteristics of the AI society based on the modeling characteristics of L3 layer.

[0047] A governance module is configured to determine the factors affecting the government service, match the corresponding modeling characteristics of the digital government service model according to different influencing factors, and determine the events leading to significant changes in the influencing factors according to the association relationship between different influencing factors and corresponding modeling characteristics.

[0048] Compared with the prior art, the present application has the following beneficial effects:

[0049] According to the complexity of the main body and the system, the model describes the behavior mode and decision-making process of the adaptive government department Agent from four levels (namely L1, L2, L3 and L4) of autonomous characteristics, evolution characteristics, interaction characteristics and emergence characteristics. Among them, the L1 layer models the perception, memory, behavior and decision-making ability of the government department Agent; the L2 layer models the learning mechanism of the government department Agent based on heterogeneity and bounded rationality; the L3 layer models the communication, cooperation, competition and network topology structure among the government department Agents; and the L4 layer models the environmental influence, feedback cycle and intervention mechanism of the system from a macro perspective. This framework captures and models the micro individual characteristics and emergence mechanism in the digital government service system, which not only helps to standardize the modeling process, but also eliminates the gap between the modeling of the complex characteristics of the government department Agent and the system and the application of AI technology.

[0050] Advantages of additional aspects of the application will be partially given in the following description, partially will become obvious from the following description, or will be understood by the practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0051] The drawings accompanying the specification of this application form a part thereof, serve to provide further understanding of the application, and together with the description of the exemplary embodiments of the application given below, explain the application, and do not constitute an improper limitation of the application.

[0052] Figure 1 is a general structure diagram of the adaptive intelligent agent modeling method for digital government service provided by the embodiment of the application;

[0053] Figure 2 is a flowchart of the adaptive intelligent agent modeling method for digital government service provided by the embodiment of the application;

[0054] Figure 3 is a modeling logic diagram of individual autonomous characteristics provided by the embodiment of the application;

[0055] Figure 4 is a modeling logic diagram of individual evolution characteristics provided by the embodiment of the application;

[0056] Figure 5 is a modeling logic diagram of inter-agent interaction characteristics provided by the embodiment of the application;

[0057] Figure 6 is a modeling logic diagram of system emergence characteristics provided by the embodiment of the application. DETAILED DESCRIPTION

[0058] The application will be further described below in conjunction with the drawings and embodiments.

[0059] It should be noted that the following detailed description is illustrative only, and is intended to provide further description in connection with the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs.

[0060] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0061] As mentioned in the background, the existing digital government service system modeling process often involves various features at the individual, social, and system levels. However, the existing modeling methods are difficult to accurately represent these complex features, and there is a gap between existing technology and complex feature modeling.

[0062] The present application provides a method for aligning the features of a digital government service system based on adaptive Agent modeling with AI technology. Based on the collection of log data of the digital government service system, the method can be modeled in four layers (i.e., L1, L2, L3, and L4):

[0063] (1) L1: Statistics of the characteristics, state changes, and parameter distribution of the government department Agent, modeling the autonomous characteristics of the department Agent from the individual perspective, including perception, memory, behavior, and decision-making.

[0064] (2) L2: Analysis of the heterogeneous characteristics and learning evolution mechanism of the government department Agent, modeling its evolution characteristics from the individual perspective, including modeling of Agent learning mechanism based on heterogeneity and bounded rationality.

[0065] (3) L3: Statistics of the interaction data between government department Agents, modeling the interaction characteristics between Agents from the social perspective, including communication, collaboration, competition, and topology.

[0066] (4) L4: Modeling the emergent characteristics of artificial intelligence society from the system perspective, including environmental impact within the system, feedback loop between department Agents and the system, and intervention mechanism. When modeling the digital government service system, part of the characteristics can be modeled according to the research problem.

[0067] The AI society can be regarded as a virtual system for analyzing a digital government service system based on ABM technology, including intelligent agents, environment, and system operation rules. It is a new social form generated by artificial intelligence technology, characterized by interaction between agents and association between agents and environment. Service individual agents, as the main body of system modeling, can represent real individuals or entities, specific robots, or completely software-simulated entities. In the digital government service system, service individual agents are adaptive department agents, which include same-type departments with similar functions and different-type departments with different functions, as well as cooperation and competition. They are endowed with different complex characteristics and behaviors, and can autonomously adjust their behaviors, strategies, and decisions according to environmental changes and obtained information to achieve better performance and adaptability. In Figure 1 , the interaction between the hierarchical AI features and the intermediate part constitutes a relatively complete and abstract structure of the evolution cycle. The basic theory provides basic modeling techniques for the features of each layer. The model library encapsulates models corresponding to various features in the object-oriented component structure, facilitating management and expansion as needed. The input of this method is the information list of government department agents and dynamic features in the service system at time t; the output is the information list of government department agents and dynamic features in the service system at time t+1. That is, the system takes cycle t as the evolution unit, runs a complete round in each cycle, and updates the information list of government department agents and dynamic features in the service system; set the system to end running after T cycles, then analyze the evolution process of the system by statistical deduction of the running data of the system in T cycles (including influencing factors and response variables of the reaction system features), and assist in system governance.

[0068] Embodiment One

[0069] As shown in Figure 2 , the embodiment provides a self-adaptive agent modeling method for digital government service, including the following steps:

[0070] Step 1: Obtain the business processing data of each government department agent;

[0071] In this embodiment, the business processing data of each government department agent includes the change of the business processing capacity of each government department agent, the change of the business backlog, the collaboration frequency between departments, and other data, reflecting the running state of government departments in government service; for example, the time when the user submits an application, the type and progress of business processing, and the API calling situation between different agencies, including request parameters, response time, and results.

[0072] Step 2: Based on the business processing data of each government department agent, a digital government service model is constructed;

[0073] In this embodiment, the digital government service model includes multiple layers, each modeling the characteristics of different complexity of the subject and system; L1 layer models the basic characteristics of government department Agent, L2 to L4 layers respectively model supplementary characteristics on the basis of the modeling characteristics of the previous layer; the modeling of department Agent from L1 to L4 is increasingly intelligent, and the system is increasingly complex.

[0074] Specifically, it includes the following steps:

[0075] Step 201, L1 layer models the basic characteristics of government department Agent;

[0076] Adaptive Agent simulates the behavior of real human or human group, which is reflected in their ability to reason, plan and act on themselves, other intelligent Agents and environment. As shown in Figure 2 The perception (sensors) and memory of government department Agent provide key information about the environment and experience. The cooperation of perception, memory and effector enables government department Agent to make wise decisions and take appropriate actions.

[0077] Specifically, it includes the following steps:

[0078] Step 2011, information perception and storage

[0079] Get the current state information of government department Agent itself (including assets, location) and environment information list information (such as other Agent and social environment information that government department Agent can perceive);

[0080] Department Agent perceives the state of itself and the environment by reading log data and system state information. For example, the business processing module can perceive the current request load, resource usage and error state. The perceived content includes the performance indicators, resource usage, business state of itself, and the state of other Agents in the environment and the overall operation of the system. For example, a server node can perceive the load of other server nodes in order to make decisions when load balancing.

[0081] Step 2012, determine the type of information that government department Agent needs to store and retrieve and the memory requirement, and save the information of step 2011 as a "state-action" pair;

[0082] The department Agent can store the perceived information and historical experiences in databases, file systems, or memory. For example, store the results and timestamps of business processing in the database for subsequent analysis and query. The memory content includes historical data, experience rules, decision results, etc. For example, the user authentication module can remember common error types and solutions to quickly handle similar problems next time.

[0083] The generated historical data is stored in memory and continuously updated, evolving over time, the government department Agent can selectively retain, update and output memory, modeling long-term and short-term memory by controlling the forget gate, input gate and output gate. This memory mechanism allows the Agent to capture key information and long-term dependencies when processing sequential data, thereby facilitating better decision-making and prediction.

[0084] The government department Agent memory mechanism can be represented as formula (1):

[0085] M t = M t-1 ⊙ f t + A t ⊙ i t (1),

[0086] where M t is the memory of the Agent; M t-1 is the early memory of the Agent; f is the forgetting vector, which determines the features retained in the memory and their weights; A t is the new memory obtained according to the input; i t is the update vector, which determines the weights of the features in the new memory; ⊙ represents element multiplication.

[0087] Step 2013, according to the current "state-action", the Agent takes corresponding actions to complete the task or adjust its own state, makes decisions;

[0088] According to the perceived information and the experience in memory, the department Agent takes corresponding actions to complete the task or adjust its own state. For example, when the business processing module perceives that the request load is too high, it can start more threads to improve processing capacity; when the resource usage approaches the limit, it can take resource optimization measures such as releasing unnecessary memory. The behavior rules of the Agent can guide the behavior of the Agent through pre-defined rules or strategies. For example, develop a load balancing strategy, when a certain server node load is too high, distribute the request to other nodes with lower load.

[0089] The department Agent makes decisions based on the perceived information, the experience in memory, and the predefined goals. For example, when handling a user request, the Agent decides whether to approve the request based on the user's identity, the type of request, and the resource situation of the system. Meanwhile, the Agent can use different decision-making algorithms to implement the decision-making process of the Agent. For example, using machine learning algorithms to predict future request loads based on historical data to adjust resource allocation in advance. Finally, the department Agent can evaluate the results of the decisions to continuously optimize the decision-making process. For example, by comparing the effects of different decision-making schemes, the optimal decision-making scheme is selected.

[0090] The decision-making mechanism of the government department Agent refers to the logic and algorithms inside the Agent that determine the intelligence level and behavior of the Agent. According to the intelligence level of the government department Agent, the decision-making is divided into three levels:

[0091] 1) "Zero intelligence" mode: low-intelligent Agent randomly makes decisions from a set of actions.

[0092] 2) "Observation-Action" mode: medium-intelligent Agent makes decisions based on the current state, and the decision-making mechanism consists of a set of rules in the form of "if…then…" statements.

[0093] Medium and high-intelligent government department Agents evaluate the current state and behavior based on the information data perceived in step 2011 and the memory data stored in step 2012, and make decisions that are most beneficial to themselves. The decision-making process of medium and high-intelligent department Agents is shown in formula (2) - formula (3):

[0094] E t *S t *M t →D t .Rules*V t →S (t+1) (2),

[0095] π(v|s)→Q(s,v),Q(s,v)→π(v|s) (3),

[0096] In the formula, E t is the information perceived in step 2011, S t is a list of information reflecting the current state of the department Agent; M t is the memory of the department Agent in step 2012; D t .Rules is the rule set in the decision-making mechanism; π(v|s) represents the strategy function, which represents the probability of selecting action v given state s; Q(s,v) represents the expected cumulative reward given a specific state and action. V tis the reaction ability of Agent, including its spontaneous behavior and all behaviors triggered by external events;D t is the decision-making ability of Agent, including the behavior rules of Agent in various situations;

[0097] 3) "observation-prediction" mode: high-intelligent department Agent makes decisions based on reasoning, and the effectiveness of previous decisions is represented by a loss function. The better the strategy with smaller loss function value, the greater the weight given. The decision is represented by the following formula:

[0098] D t = W h *[D t-1 ,S t ]+b h (4),

[0099] In the formula, D t and D t-1 are the weights of the decision information located at t and t-1; W h is the decision weight vector; b h is the bias vector;

[0100] Step 2014, update the Agent information list of the government department, and proceed to the next iteration until the running data of the system in the set period is completed, and the autonomous feature structure representation is obtained;

[0101] The final autonomous feature L1(Agent) structure representation of the department Agent is as follows:

[0102] L1(Agent) = <R, S t , E t , V t , M t , D t > (5),

[0103] In the formula, R is the attribute that Agent maintains stability, such as the type of government department; S t is the dynamic ability attribute of Agent changing over time, such as the assets of the government department; E t is the perception ability of agent, for example, the ability of the department to obtain information; V t is the reaction ability of Agent, including its spontaneous behavior and all behaviors triggered by external events; M t is the memory ability of Agent, including the long and short term memory duration, etc.; D t is the decision-making ability of Agent, including the behavior rules of Agent in various situations.

[0104] Step 202, model the evolution characteristics of the government department Agent in L2 layer;

[0105] Different departmental agents may perform different government service functions. For example, some departments handle citizens' household registration matters, while others may focus on tax filing services. This functional difference leads to different characteristics in their processing flows, required data, and interaction objects. Even when performing the same function, different agents may differ in performance. This includes response time, processing capacity, and reliability. For example, some agents may process requests faster due to higher hardware configurations or better algorithm optimization, while others may experience performance degradation under high load. Layer 2 (L2) builds upon Layer 1 to model more intelligent departmental agents. The L2 layer introduces diversity by giving departmental agents distinct personalities, goals, and value orientations, gradually revealing the differences and uniqueness of agents.

[0106] The evolutionary characteristics of government agency agents are mainly reflected in their learning capabilities. For example... Figure 3 As shown, the bounded rationality and heterogeneity in digital government service systems are prerequisites for government agency agents to learn, enabling them to make near-optimal decisions within limited resources and time without requiring comprehensive searches and analyses. Based on this, government agency agents continuously improve their decision-making through self-reflection, imitation of other government departments, and learning from experiences in various scenarios.

[0107] Bounded rationality comprises both incomplete information and incomplete rationality, but it does not equate to irrationality. Rather, it refers to decision-making within the constraints of information, knowledge, and ability, following a "fast-and-frugal heuristic." The bounded rationality of a government agent is represented by P(a i |s i ), indicating the government department's Agent i In state s i Take action a i The probability, rather than the determinism, provides an interface for incorporating research findings from psychology and behavioral economics, such as prospect theory and learned bias models.

[0108] Incorporating heterogeneity into artificial social modeling helps solve complex dynamic game problems. By simulating a multi-agent, multi-cycle dynamic game process, each government agency agent continuously adjusts its strategies throughout the game, collectively pushing the overall situation towards a specific "equilibrium." This includes two aspects:

[0109] i) Attribute heterogeneity: usually represented as differences in specific parameter values, such as risk preferences, energy constraints, beliefs, etc. Among them, the heterogeneity of the objectives of different department Agents determines the motivation and direction of the Agents.

[0110] ii) Behavioral heterogeneity: can be divided into two categories: a. N-type model: the strategy set is constant; b. Autonomous model: the strategy of the Agent using this model is heterogeneous. The difference from the N-type model is that the Agent can create new strategies through learning. Agents with different levels of intelligence have different ways of cognition and behavior, and their strategy sets are updated differently.

[0111] In the computational model, from the optimization based on complete rationality to the simulation based on limited rationality, the flexibility of modeling is improved, and the difficulty of system solution is reduced. The government department Agent continuously improves its behavior and performance through genetic learning and other technologies. If you want to consider the evolutionary characteristics in the digital government service system, after completing the modeling of step 201 in the L1 layer, the modeling process of the evolutionary characteristics is as follows:

[0112] It should be noted that according to different modeling scenarios and research problems, the following parts or all of the learning characteristics can be modeled;

[0113] Step 2021, genetic evolution modeling of government department Agents

[0114] Genetic evolution simulates the inheritance between departments and sub-departments. Identify the characteristics of a single department Agent and represent it as a matrix. During the evolution process, a random value between 0 and 1 will be generated. If the generated random value is less than the mutation threshold, the Agent will mutate. In the weight matrix, a weight variable will be randomly selected and replaced with a random value (fluctuation range within 10% of the initial value). This will cause the Agent to change slightly, which may produce a healthier Agent. A smaller fluctuation range can avoid catastrophic failure of the neural network.

[0115] When the energy of an Agent falls to 0, it will die. When the energy of an Agent reaches the reproduction threshold, it will produce offspring. The offspring department Agent will inherit some attributes from its parents, such as formula (6):

[0116] Offspring = parent1 * a + parent2 * (1-a) (6),

[0117] where Offspring is the attribute vector of the offspring object; a is a randomly generated value between 0 and 1, and parent1 and parent2 represent the weight matrix of the attribute.

[0118] Step 2022, Modeling of Government Department Agents by Imitation Learning

[0119] Imitation learning is used to model the mechanism of mutual learning between neighboring departments. Department Agents utilize information obtained from neighboring Department Agents, with each Department Agent's capability attributes as a knowledge point. ij is the knowledge point of the jth Agent, S r1 ,S r2 ,...,S rk represent the knowledge points of other Agents in the organization. Each knowledge point is then represented as a number of sectors (e.g., its location, number of orders stored). After comparing the standardized attribute values of all Agents, the best sector in the organization is selected. By combining these best sectors, a final learning reference target can be formed. An Agent compares one of its sectors with the sectors of the virtual learning object to retain the better point.

[0120] Step 2023, Modeling of Government Department Agents by Social Learning

[0121] The social learning mechanism of a department is used to simulate the influence of corporate culture or social culture on individual departments. Department Agents determine their game strategies based on the social culture they receive and their historical strategies in the game process. The social learning modeling of Department Agents is shown as follows:

[0122] All Department Agents in a whole region or part of a region are identified as a learning society. This learning region is represented by the following formula:

[0123] If |X d - X1| = Rand() * (X2 - X1), X d ∈ this society (7),

[0124] where X d represents a point in the region; X1 represents a reference point; X2 represents a center point; Rand() represents a random value in the range [0, 1].

[0125] Then, the viability of all Department Agents is calculated as follows:

[0126]

[0127] where Viability i represents the viability value of the ith Agent; w j represents the weight value of the jth attribute; Attribute j represents the value of the jth attribute.

[0128] After that, the normalized fitness value Fitness of all Agents is calculated based on the viability of all Agents i :

[0129]

[0130] where Fitness i represents the normalized fitness value of the i-th Agent; viability i represents the viability of the i-th Agent.

[0131] Finally, the top N Agents in the region with the highest fitness value in the period are selected, and the knowledge points of these elite Agents are broadcasted for other Agents to observe and learn. After receiving the broadcast information, other Agents compare the recommended knowledge points with their own knowledge points, and decide whether to imitate learning and change their own knowledge according to their own strategy switching rules.

[0132] Step 2024, update the information list of the department Agent and the service system, and perform the next period iteration until the running data of the system in the set period is completed, and the evolution characteristic structure of the department Agent is obtained;

[0133] L2(Agent) = <L1(Agent), F, Y t > (10),

[0134] where F is the prior condition of Agent learning and evolution, including heterogeneity and bounded rationality; Y t is the learning ability of Agent, which is improved by learning the environment and other Agents.

[0135] Step 203, modeling the interaction characteristics between the department Agents in L3 layer

[0136] The interaction modeling of department Agents based on social network is directly related to the structural characteristics and internal evolution mechanism of social system. Due to the diversity and interaction of Agents, quantitative change will lead to qualitative change, thus forming a complex and dynamic AI society. L3 is modeled based on L2, and the department Agents in L3 not only have basic adaptive characteristics and learning mechanism, but also have communication and competition and cooperation mechanism between each other, which constitutes the organizational structure of the system.

[0137] The interaction of department Agents mainly reflects in the cooperation between Agents, and cooperation is the combination of individual task solving ability and communication ability. In complex coordination tasks, there may be differences in goals or intentions between department Agents, which brings difficulties to cooperation. Figure 4As shown in the social network-based Agent interaction mechanism, communication is the basis of cooperation and competition. In the fierce market competition, department Agents may form cooperative groups with different topological structures, and department Agents make decisions and take appropriate actions according to their own cognition and information received from the scene.

[0138] In the computational model, the communication between department Agents can be realized through complex networks, shared data structures, etc. Department Agents share important data, insights and knowledge through communication to increase mutual understanding and decision-making ability. This includes direct point-to-point communication, as well as indirect broadcast or intermediary transmission. When modeling, various techniques and methods such as game theory and social network analysis can be used to capture the characteristics of communication, competition and cooperation. Agents can share the same goal and reasonably allocate tasks according to the availability of individual skills and resources. Through effective task division and cooperation, Agents can improve work efficiency. If you want to consider the interaction characteristics in the digital government service system, after completing the modeling in L1 layer and the modeling in L2 layer, the modeling process of the interaction characteristics specifically includes the following steps:

[0139] Step 2031, design the communication network topology structure of the initial government department Agent in the system;

[0140] The initial contact of the government department Agent can be represented by a network structure, such as a small-world network. In addition to the fixed topology structure in the previous system or the topology structure switched according to certain rules, the modeling process of passively changing the network topology structure through node failure, edge failure, adding edges, etc. can also be used to study some sudden events in society. This kind of work combined with control design will eventually be classified as an optimal topology design problem.

[0141] Step 2032: modeling of the cooperation process between government department individuals:

[0142] Obtain the relationship list of government department Agents by recording the subjects establishing contact and update the weights between the nodes of the communication network topology structure;

[0143] Government department Agents build their own relationship networks through cooperation times, which can be represented by the following formula:

[0144]

[0145] In the formula, P ij is the strength of the relationship between Agent i and Agent j ; N i and N j are the relationship strengths between Agent i and Agent jThe number of connected objects in the list of strong connections; N is the total number of Agents in the network; n ij Agent i and Agent j The actual number of times they meet; T is the total number of times the process is repeated.

[0146] Step 2033: Modeling the collaborative process between individual government departments and collaborative alliances;

[0147] The adjustment and updating of government agency agent cooperation intentions depend on the following formula:

[0148] (v i_co ≥v i_in )? Evole(no):Evole(yes) (12),

[0149] In the formula, V i_co and V i_in Let represent the gains of the i-th agent when choosing the cooperation strategy and the independent operation strategy, respectively. Only when the above formula is not satisfied will the agent tend to join a new cooperative alliance.

[0150] In the collaborative process of steps 2032 and 2033, the profit distribution calculation formula is as follows:

[0151]

[0152] In the formula, Profit(i) is the Agent. i The revenue obtained after participating in the collaboration; Cost(i) is the Agent's... i Total cost of participating in collaboration; Cost total It represents the total cost invested in processing the order; Value(i) is the Agent. i The normalized effective value created through participation and collaboration; Value total It is the total effective value created during the order processing process; w a It is the profit distribution coefficient; Profit total It is the total profit obtained when an order is completed.

[0153] Negotiations between government agents can often achieve win-win results, leading to a Nash equilibrium in the system. The negotiation mechanism is represented by the following formula:

[0154] C i =f(H i (14),

[0155] In the formula, C iFor the coordination strategy, let f(i) represent the ratio of the expected profit Profit(i) allocated to the Agent; f() represents the coordination function; H i Indicates Agent i Experience.

[0156] Step 2034: Update the information list of the department agent and service system, proceed to the next cycle iteration, and obtain the structured representation of the interaction features of the government department agent;

[0157] L3 (Society) = <L2(Agent),C t G t ,P t ,TS> (15),

[0158] In the formula, C t For communication rules between agents, including communication range, etc.; G t The rules governing cooperation between agents, including cooperation conditions, etc.; P t The competition rules between agents include resource constraints; TS represents the system topology, such as tree structure, star structure and grid structure.

[0159] Step 204: Modeling the Emergent Characteristics of the AI ​​Society at Layer L4

[0160] Numerous environmental factors influence digital government systems. From a societal perspective, the public's increasing demands for convenience and transparency in government services drive continuous optimization of digital government systems. When citizens submit applications, the system provides real-time feedback on the application's acceptance status and processing progress. The intervention mechanism of digital government systems is reflected in the proactive control of government processes and service quality. In the event of emergencies, such as natural disasters restricting offline government services, the system can quickly intervene, guiding relevant services online and issuing emergency handling guidelines. For risky government decisions, the system can intervene through data analysis and model prediction.

[0161] Digital government service systems demonstrate strong intelligence in three aspects: environmental impact, system feedback, and intervention mechanisms. In social systems, agents not only engage in competition and cooperation, but also exhibit time-varying correlations and nonlinear interactions between internal agents or groups, as well as between agents and the environment. These are the fundamental driving factors for the emergence of such systems.

[0162] In the digital government service system, the scenario is a system composed of government department agents and the environment. The environment can be a natural environment or a social environment. The environment in the digital government service system is the external conditions and influences in which the service individual agents exist, which determines the perception ability and action space of the agents, mainly embodied as the demand order in the society;

[0163] System external intervention mainly includes two aspects:

[0164] i) Change the variables or parameters in the system through human intervention in the experiment, and observe how the system responds;

[0165] ii) Consider different types of human intervention, such as policy changes, technology upgrades, etc., and test the evolution results of the system under different intervention strategies.

[0166] As shown in Figure 5 , the government department agent can send a request for environmental state information. At the same time, the behavior of the government department agent can cause changes in the state of the environment and the behavior of other agents. In addition, the feedback of the government department agent is simulated through the reward signals or evaluation indicators of the system. These signals can be used to indicate whether the behavior of the agent is correct or close to the target, and update the learning algorithm or decision-making mechanism of the government department agent. The system or environment continuously optimizes the strategy set by receiving feedback signals from the agent, and induces or forces intervention in the decision-making of the agent.

[0167] In the computational model, the environment model acts as a container that connects agents with geographical scenarios based on domain knowledge, including spatial distribution and spatiotemporal association. When the research goal is to understand the mechanism, an abstract environment model is sufficient as the background of the agent without extensive environmental details. However, when the goal is to optimize the strategy, ensuring a high degree of consistency between the artificial society and the real system, a more specific environment model is needed. Agents use appropriate interfaces, APIs, or virtual environments to simulate their interaction with the environment. If you want to consider the interaction characteristics in the digital government service system, after completing the modeling of the L1 layer, the modeling of the L2 layer, and the modeling of the L3 layer, the modeling process of the emergent characteristics specifically includes:

[0168] Step 2041, define the influence of the environment on the government department agent;

[0169] The environmental information obtained by the government department agent from the social demand order can be described according to the following dimensions: 1) completeness: whether the environment provides all the information required by the agent; 2) observability: whether the agent can observe the state of the environment in whole or in part; 3) determinism: whether the environment has a unique result for the agent's behavior; 4) staticity: whether the environment remains unchanged before and after the agent's behavior; 5) discreteness: whether the environment is composed of finite states and actions, or has continuity and infinity; 6) unity: whether there is only one agent in the environment, or there is competition or cooperation between multiple agents.

[0170] The environmental impact perceived by the government department agent is represented as follows:

[0171] E = <S e ,X,δ e >, S e *X→δ e (16),

[0172] where S e is the set of environment states; X is the set of external events, and δ e represents the new environment state based on the current environment state and events.

[0173] Step 2042, modeling system feedback mechanism

[0174] The government department agent selects an action based on the current state, interacts with the environment, and obtains immediate rewards and feedback for the next state.

[0175] Based on this feedback, the government department agent can update its value function and strategy to make better decisions in future interactions.

[0176] Specific algorithms and methods can vary depending on the characteristics of the problem and research needs, such as Q-learning, deep Q-network (DQN), policy gradient, etc. The feedback information received by the government department agent from the environment is represented using a reward function as follows:

[0177] R(t) = r( S t ,a(t)) (17),

[0178] where R(t) represents the immediate reward obtained by the agent at time t based on the selected action and the current environment state; R() represents the reward function; S t is the environment state at time t; a(t) represents the action selected by the agent at time t.

[0179] According to the feedback from the environment, the government department Agent can update the value function to more accurately evaluate the value of the state and action. The specific formula of the value function update can be different according to the specific algorithm and method. Based on the obtained reward and the estimate of the value function, the updated value function formula can be expressed as follows:

[0180] V(s(t)) = V(s(t)) + a * (R(t) + g * V(s(t + 1)) - V(s(t))) (18), Q(s(t), a(t)) = Q(s(t), a(t)) + a * (R(t) + g * Q(s(t + 1), a(t + 1)) - Q(s(t), a(t))) (19),

[0181] where V(s(t)) represents the state value estimate at time t, i.e. the expected value of the cumulative reward that the Agent can obtain in state s; Q(s(t), a(t)) represents the action value estimate at time t, i.e. the expected value of the cumulative reward that the Agent can obtain after selecting action a in state s; s(t + 1) represents the next state that the Agent enters after selecting action a(t) at time t; a is the learning rate, and g is the discount factor used to measure the importance of future returns.

[0182] Based on the feedback from the environment and the policy evaluation, the Agent can update its policy to obtain better returns in future interactions. The updated policy formula based on the value function estimate can be expressed as follows:

[0183] p(t + 1) = argmax Q(s(t), a(t)) (20),

[0184] where p(t + 1) represents the policy at time t + 1, and the argmax operation represents selecting the action that maximizes Q(s(t), a(t)).

[0185] Step 2043, modeling the intervention mechanism of the system operator

[0186] Under the induced policy, the Agent makes decisions based on its own state, but has no obligation to execute; the intervention policy imposes constraints on the constraints that the Agent must consider before making decisions. The system operator continuously regulates the policy according to the system's goals and state, which can be represented as follows:

[0187] Intervention(R t ,J t )=<S Syatem ,S t ,Goal System > (21),

[0188] where Intervention(R tJ t ) represents the intervention mechanism; R t and J t represent the incentive strategy and intervention strategy of the system respectively; S Syatem represents the state of the system under the action of the Agent; S t represents the state of the environment at time t; Goal System represents the goal of the system.

[0189] Step 2044, update the information list of the department Agent and the service system, and perform the next cycle iteration to obtain the emergent characteristics of the AI society;

[0190] The structured representation L4(System) of the department Agent emergent characteristics is as follows:

[0191] L4(System) = <L3(Society), Env., Q t , N t > (22),

[0192] In the formula, Env. is the interaction between the environment and the department Agent; Q t represents the feedback mechanism from the system to the Agent. The Agent can extract knowledge and improve decision-making; N t is the system intervention strategy adopted for stable development of the system.

[0193] Step 3: Determine the factors affecting government services, match the corresponding modeling characteristics of the digital government service model according to different influencing factors, and determine the events that cause significant changes in the influencing factors according to the association relationship between different influencing factors and corresponding modeling characteristics.

[0194] Specifically, the following steps are included:

[0195] Step 301, determine the response variables affecting government services;

[0196] In this embodiment, system efficiency and customer satisfaction can be used.

[0197] System efficiency is used to evaluate the overall operation efficiency of the system, which can be obtained by collecting system performance index data monitored by the system (such as server load average, response time average, business processing success rate, etc.) and processing calculation. Customer satisfaction summarizes the satisfaction evaluation results of the service population Agent, calculates the average satisfaction score in different time periods, and understands the customer satisfaction degree of government services.

[0198] Step 302, obtain related data of the response variables affecting government services;

[0199] Start the multi-agent model, and each agent interacts according to the behavior rules in each time step, while collecting the data related to the influencing factors (such as the number of different types of business applications related to social demand, the distribution of application time, and the length of the business queue of government department agents, the fluctuation of business processing time, the frequency of cooperation between departments, etc.) and the response variable related data (such as the success rate of business processing, the average processing time, the system resource utilization rate, and the customer satisfaction score data related to the system efficiency and the customer satisfaction of the business crowd agents).

[0200] Step 303, time series analysis of the acquired response variable related data affecting the government service, to obtain the evolution process thereof;

[0201] Step 3031, when analyzing the system evolution process, the system efficiency indicators (such as the success rate of business processing and the average processing time) and the customer satisfaction data can be analyzed by time series analysis (such as smoothing the data by using the moving average method), and the change trend thereof in the evolution period can be observed to determine whether it is improved or deteriorated;

[0202] Step 3032, the correlation between the change of social demand (such as a significant increase in the number of business applications), the change of government department behavior (such as the adjustment of business processing rules) and the system efficiency and customer satisfaction can be studied by correlation analysis;

[0203] In this embodiment, correlation analysis can be performed by calculating the correlation coefficient or constructing a regression model.

[0204] Step 3033, the key events that cause significant changes in system efficiency or customer satisfaction (such as a new policy causing a sharp increase in the number of business applications) can be found by identifying key events and turning points;

[0205] Step 3034, the reasons for the change of system efficiency or satisfaction are analyzed, and the turning points in the evolution of the system (such as the improvement of efficiency after system upgrade) and the related factors and behavior changes are identified.

[0206] Step 304, based on the evolution analysis results, the auxiliary governance of government service is performed;

[0207] Through data analysis of the system evolution results, suggestions for performance optimization, function improvement, risk warning and prevention, etc. can be obtained.

[0208] In terms of performance optimization suggestions, according to the system efficiency analysis results, if the business processing time is too long, the government department business processing flow can be optimized, the personnel can be increased or the system resource performance can be improved, and the dynamic resource allocation scheme can be proposed for the unreasonable system resource utilization rate (such as server time overload or idling) ;

[0209] In the direction of function improvement, according to customer satisfaction analysis, if the masses are not satisfied with the service quality of a certain type of business, suggestions for improving the business handling process or service mode (such as increasing the clarity of the service guide, optimizing the online handling interface) are put forward;

[0210] In the risk early warning and prevention aspect, based on the system evolution process analysis, the risk early warning index system is established, when the business processing success rate continuously decreases or the customer satisfaction is lower than the critical value, timely early warning is made to prompt the relevant departments to take measures (such as personnel training, system function optimization) to prevent serious problems of the system.

[0211] Embodiment two

[0212] The embodiment provides a self-adaptive intelligent agent modeling system for digital government service, comprising:

[0213] A data acquisition module is configured to acquire business processing data of intelligent agents of various government departments;

[0214] A digital government service modeling module is configured to construct a digital government service model based on the business processing data of intelligent agents of various government departments; wherein the digital government service model comprises multiple layers, each layer models different complexity of subject and system characteristics; wherein L1 layer models the basic characteristics of the intelligent agents of the government departments, L2 layer models the evolution characteristics of the intelligent agents of the government departments based on the modeling characteristics of L1 layer, L3 layer models the interaction characteristics between the intelligent agents of various government departments based on the modeling characteristics of L2 layer, and L4 layer models the emergent characteristics of the AI society based on the modeling characteristics of L3 layer.

[0215] A governance module is configured to determine factors affecting government services, match corresponding modeling characteristics of the digital government service model according to different influencing factors, and determine events that cause significant changes in the influencing factors according to the association between different influencing factors and corresponding modeling characteristics.

[0216] Embodiment three

[0217] The embodiment provides a program product, which is a computer program product comprising a computer program, the computer program being executed by a processor to implement the steps in the self-adaptive intelligent agent modeling method for digital government service of embodiment one.

[0218] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An adaptive intelligent agent modeling method for digital government services, characterized in that, Includes the following steps: Acquire business processing data from intelligent agents in various government departments; Based on the business processing data of various government department intelligent agents, a digital government service model is constructed. The digital government service model includes multiple layers, with each layer modeling the subject and system characteristics of different levels of complexity. Specifically, the L1 layer models the basic characteristics of government department intelligent agents, the L2 layer models the evolutionary characteristics of government department intelligent agents based on the characteristics modeled in the L1 layer, the L3 layer models the interaction characteristics between government department intelligent agents based on the characteristics modeled in the L2 layer, and the L4 layer models the emergent characteristics of the AI ​​society based on the characteristics modeled in the L3 layer. Identify the factors affecting government services, match the corresponding modeling features of the digital government service model according to different influencing factors, and identify the events that lead to significant changes in the influencing factors based on the correlation between different influencing factors and their corresponding modeling features; The process of modeling the evolutionary characteristics of government agency agents at the L2 layer includes: Based on the fundamental characteristics of a single government department intelligent agent, the genetic evolution of the government department intelligent agent is modeled. Information about neighboring government department agents is obtained. The capability attributes of each government department agent are treated as a knowledge point. Each knowledge point is represented as multiple sectors. The optimal sector is determined based on the agent's attributes. Based on the combination of the optimal sectors, a learning reference target is determined for imitation learning. Define the learning area; Calculate the survivability of all agents in each department, combine the survivability of all agents in each department to calculate the fitness value of all agents in each department, select the agents with the highest fitness value in the first N periods of the learning area, and broadcast the knowledge points of these agents for other agents to learn. The process of modeling the emergent characteristics of the AI ​​society at layer L4 includes: Define the impact of the environment on government agency intelligent agents; The government department's intelligent agent selects actions based on the current state, interacts with the environment, and receives immediate rewards and feedback for the next state; By combining rewards and feedback, the constraints that the agent considers before making decisions are continuously adjusted according to the purpose and state of system operation; that is, the intervention strategy.

2. The adaptive intelligent agent modeling method for digital government services as described in claim 1, characterized in that, The L1 layer models the basic characteristics of government agency agents, including: Obtain a list of status and environmental information for the government department's intelligent agent; Determine the types of information that government intelligence agents need to store and retrieve, as well as their memory requirements, and save the government intelligence agents' own state information and environmental information list as "state-behavior" pairs; Based on the current "state-behavior" pair, the agent takes corresponding actions to complete the task or adjust its own state and make decisions. Based on the decision, the government department's intelligent agent updates its own state information and environmental information list data to obtain a structured representation of autonomous features.

3. The adaptive intelligent agent modeling method for digital government services as described in claim 1, characterized in that, The process of modeling the interaction features between government agency agents at the L3 layer includes: Determine the initial communication network topology between the intelligent agents of each government department; If it is collaboration between individual government departments, the weights between nodes in the communication network topology are updated by recording the relationship list of the department agents that established the connection. If it is a collaboration between an individual government department and a collaborative alliance, determine the collaboration intention of the government department's agent and choose whether to join the collaborative alliance; Calculate the profit distribution during the collaboration process and decide whether to adopt a negotiation mechanism based on its own situation.

4. The adaptive intelligent agent modeling method for digital government services as described in claim 3, characterized in that, The formula for calculating profit distribution during the collaboration process is as follows: ， In the formula, for Income obtained after participating in the cooperation; for The total cost of participating in the collaboration; It is the total cost incurred in processing orders; yes The normalized effective value created through participation in collaboration; It is the total effective value created during the order processing process; It is the profit distribution coefficient; It is the total profit obtained when an order is completed.

5. The adaptive intelligent agent modeling method for digital government services as described in claim 1, characterized in that, The structured representation of autonomous features is as follows: ， The evolutionary characteristics of government intelligence agents are represented in a structured manner as follows: ， The structured representation of the interaction features of government department intelligent agents is as follows: ， The emerging characteristics of the AI ​​society can be structurally represented as follows: ， in, To maintain stable properties for the Agent These are the dynamic capability attributes of the Agent that change over time. It is the agent's perception ability. It is the agent's responsiveness. For the agent's memory ability, It is the agent's decision-making ability. F These are the prior conditions for agent learning and evolution. It is the agent's learning ability. For communication rules between agents, Rules for cooperation between agents The competition rules between agents are defined by TS, which represents the system topology. . It is the interaction between the environment and the departmental agent; This represents the feedback mechanism from the system to the agent. System intervention strategies adopted to ensure the stable development of the system.

6. The adaptive intelligent agent modeling method for digital government services as described in claim 1, characterized in that, Factors affecting government services include system performance indicators and customer satisfaction.

7. An adaptive intelligent agent modeling system for digital government services, characterized in that: include: The data acquisition module is used to acquire business processing data from various government departments' intelligent agents; The digital government service modeling module is used to construct a digital government service model based on the business processing data of various government department intelligent agents. The digital government service model includes multiple layers, each layer modeling the subject and system characteristics of different levels of complexity. Specifically, the L1 layer models the basic characteristics of government department intelligent agents, the L2 layer models the evolutionary characteristics of government department intelligent agents based on the characteristics modeled in the L1 layer, the L3 layer models the interaction characteristics between government department intelligent agents based on the characteristics modeled in the L2 layer, and the L4 layer models the emergent characteristics of the AI ​​society based on the characteristics modeled in the L3 layer. The process of modeling the evolutionary characteristics of government agency agents at the L2 layer includes: Based on the fundamental characteristics of a single government department intelligent agent, the genetic evolution of the government department intelligent agent is modeled. Information about neighboring government department agents is obtained. The capability attributes of each government department agent are treated as a knowledge point. Each knowledge point is represented as multiple sectors. The optimal sector is determined based on the agent's attributes. Based on the combination of the optimal sectors, a learning reference target is determined for imitation learning. Define the learning area; Calculate the survivability of all agents in each department, combine the survivability of all agents in each department to calculate the fitness value of all agents in each department, select the agents with the highest fitness value in the first N periods of the learning area, and broadcast the knowledge points of these agents for other agents to learn. The process of modeling the emergent characteristics of the AI ​​society at layer L4 includes: Define the impact of the environment on government agency intelligent agents; The government department's intelligent agent selects actions based on the current state, interacts with the environment, and receives immediate rewards and feedback for the next state; By combining rewards and feedback, the constraints that the agent considers before making decisions are continuously adjusted according to the purpose and state of system operation; that is, the intervention strategy. The governance module is used to identify factors affecting government services, match the corresponding modeling features of the digital government service model according to different influencing factors, and identify events that lead to significant changes in the influencing factors based on the correlation between different influencing factors and their corresponding modeling features.

8. A program product, said program product being a computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps in the adaptive intelligent agent modeling method for digital government services as described in any one of claims 1-6.

Images