Government affair service process optimization method and system based on artificial intelligence

Abstract

The invention discloses a government affair service process optimization method and system based on artificial intelligence, and relates to the field of government affair service digitization, and the method comprises the steps: constructing a government affair knowledge graph for storing the static knowledge of the government affair field; based on the government affair knowledge graph, constructing a government affair affair graph of an event causal chain, and integrating the government affair affair graph into a process digital twinborn body to form a government affair affair graph enhanced digital twinborn body; based on a historical successful government affair process case, performing strategy pre-training on a reinforcement learning agent by using a government affair affair atlas to obtain an initial process optimization strategy; in the government affair atlas enhanced digital twinborn body, the reinforcement learning agent guides an exploration process based on an initial process optimization strategy according to a causal association relationship disclosed by the government affair atlas. According to the method, the verified strategy is deployed to a real environment to execute optimization, and the digital twin is dynamically updated and enhanced based on operation feedback, so that closed-loop optimization and continuous self-evolution of the government affair service process are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of government service digitization, in particular to a government service process optimization method and system based on artificial intelligence. BACKGROUND

[0002] In the field of government service digitization, process optimization technology based on artificial intelligence has become a research hotspot. The existing technical solutions usually adopt a method combining reinforcement learning and digital twinning. By establishing a process model in a virtual environment, an intelligent agent explores an optimization strategy in a trial-and-error manner. To improve efficiency, some advanced solutions introduce imitation learning, which pre-trains an intelligent agent using historical successful case data to quickly obtain a high-performance process optimization strategy. This data-driven method shows certain potential in improving process efficiency.

[0003] The existing technology mainly relies on learning the statistical correlation in historical data, lacks understanding and modeling of the inherent causal logic between process links, and the strategy generated by it may perform well in efficiency indicators, but the decision logic may deviate from business specifications, posing compliance risks. The intelligent agent may learn to shorten the time by skipping some non-critical verifications, although the problem is not recorded in the historical data, it violates the rigid constraints of the policy, and the optimization objective function is difficult to integrate into the causal common sense of the field, which cannot ensure the compliance of the strategy. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a government service process optimization method based on artificial intelligence, which solves the core problem of decision compliance risk caused by lack of causal understanding in the prior art.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a government service process optimization method based on artificial intelligence, which comprises: constructing a government knowledge graph storing static knowledge in the government field; Based on the government knowledge graph, a government reason graph of event causal chain is constructed, and the government reason graph is integrated into a process digital twinning to form a government reason graph enhanced digital twinning; Based on historical successful government process cases, the government reason graph is used to pre-train the strategy of a reinforcement learning intelligent agent to obtain an initial process optimization strategy; In the government reason graph enhanced digital twinning, the reinforcement learning intelligent agent generates a target process optimization strategy through simulation iteration based on the initial process optimization strategy and the causal correlation relationship revealed by the government reason graph; The decision logic of the target process optimization strategy is explained causally by using the government affair management graph, and the root cause of the process abnormality monitored in the real government affair operation environment is traced back; The target process optimization strategy verified by the causal explanation is deployed to the real government affair operation environment to perform process optimization, and the government affair management graph enhanced digital twin is updated and fed back based on the performance monitoring data and the root cause tracing result.

[0007] As a preferred scheme of the government affair service process optimization method based on artificial intelligence, wherein: the government affair knowledge graph storing static knowledge in the government affair field is constructed, including the following steps: Multi-source heterogeneous data is obtained from the government affair data source and is cleaned, and based on the cleaned multi-source heterogeneous data, the government affair field entities are recognized and extracted by using natural language processing technology; The static relationship between the recognized government affair field entities is recognized and extracted by using natural language processing technology; The recognized government affair field entities and the static relationship between the government affair field entities are stored in the form of nodes and edges in the graph database to form the government affair knowledge graph storing static knowledge in the government affair field.

[0008] As a preferred scheme of the government affair service process optimization method based on artificial intelligence, wherein: based on the government affair knowledge graph, the government affair management graph of event causal chain is constructed, including the following steps: The government affair knowledge graph is analyzed to identify entities and relationships in the government affair knowledge graph as dynamic event elements, and is instantiated as event nodes; Based on historical process logs and business rules, the causal logic relationship between the event nodes is mined, the causal logic relationship and the event nodes are combined, and the government affair management graph of event causal chain is constructed in the form of event nodes and causal logic relationship edges.

[0009] As a preferred scheme of the government affair service process optimization method based on artificial intelligence, wherein: the government affair management graph is integrated into the process digital twin to form a government affair management graph enhanced digital twin, including the following steps: Based on historical government affair process data, the process digital twin is constructed, and the data mapping rules between the government affair management graph and the process digital twin are established; The causal reasoning logic of the government affair management graph is embedded into the process digital twin, and the process digital twin is enhanced to a government affair management graph enhanced digital twin.

[0010] As a preferred scheme of the government affair service process optimization method based on artificial intelligence provided in the present application, wherein: based on historical successful government affair process cases, the strategy of the reinforcement learning agent is pre-trained using the government affair affair graph, and an initial process optimization strategy is obtained, including the following steps: From the historical records of government affair process, the historical successful government affair process cases are screened out, and the corresponding event sequence is parsed from the historical successful government affair process cases based on the government affair affair graph, forming an event causal chain; The event causal chain is converted into the connection weight of the strategy network of the reinforcement learning agent through the imitation learning algorithm, and an initial weight matrix is generated; The initial weight matrix is used to configure the parameters of the strategy network of the reinforcement learning agent to complete the initialization training of the strategy network, and an initial optimization strategy defined by the initial weight matrix is obtained.

[0011] As a preferred scheme of the government affair service process optimization method based on artificial intelligence provided in the present application, wherein: in the government affair affair graph enhanced digital twin, the reinforcement learning agent guides the exploration process according to the causal relationship revealed by the government affair affair graph based on the initial process optimization strategy, and generates a target process optimization strategy through simulation iteration, including the following steps: The initial weight matrix corresponding to the initial optimization strategy is loaded into the strategy network of the reinforcement learning agent, and the reinforcement learning agent dynamically adjusts the action exploration space according to the causal relationship revealed by the government affair affair graph in the simulation environment of the affair graph enhanced digital twin; Based on the dynamically adjusted action exploration space, the reinforcement learning agent performs multi-round simulation iteration optimization strategy in the affair graph enhanced digital twin.

[0012] As a preferred scheme of the government affair service process optimization method based on artificial intelligence provided in the present application, wherein: the decision logic of the target process optimization strategy is causally explained using the government affair affair graph, and the root cause of the process anomaly monitored in the real government running environment is traced, including the following steps: Based on the updated strategy network weight matrix, the key decision sequence of the target process optimization strategy is simulated and reproduced in the affair graph enhanced digital twin; Using the key decision sequence, a causal explanation report describing the basis for decision-making is generated; The process anomaly monitored in the real government running environment is mapped to the corresponding abnormal event node in the government affair affair graph; Based on the event causal chain of the government affair affair graph, the reverse causal reasoning is performed on the abnormal event node identified by the process abnormal event mapping, and the root cause event causing the abnormal event is located.

[0013] As a preferred scheme of the government service process optimization method based on artificial intelligence, the target process optimization strategy verified by the causal explanation is deployed to the real government running environment to execute process optimization, including the following steps: The updated strategy network weight matrix corresponding to the target process optimization strategy verified by the causal explanation is converted into a workflow rule executable by the real government running environment; The workflow rule is deployed to the process execution engine of the real government running environment, and the process execution engine in the real government running environment automatically processes a newly arrived government process instance according to the deployed workflow rule to execute process optimization; In the process optimization execution process, the running state data of the government process instance is collected in real time to monitor the optimization effect.

[0014] As a preferred scheme of the government service process optimization method based on artificial intelligence, the government affairs reasoning graph enhanced digital twin is updated and fed back based on the performance monitoring data and root cause tracing results, including the following steps: The performance monitoring data and root cause tracing results obtained from the real government running environment are analyzed to identify deviations from the government affairs reasoning graph enhanced digital twin prediction, and the event causal chain in the government affairs reasoning graph is calibrated; According to the calibrated government affairs reasoning graph and the new rules reflected in the performance monitoring data, the simulation parameters of the process digital twin are updated; The calibrated government affairs reasoning graph and the updated simulation parameters of the process digital twin are re-integrated to form an updated government affairs reasoning graph enhanced digital twin.

[0015] In a second aspect, the present application provides a government service process optimization system based on artificial intelligence, which comprises a government knowledge graph construction module for constructing a government knowledge graph storing static knowledge in the government field; A construction module constructs a government affairs reasoning graph of event causal chain based on the government knowledge graph, integrates the government affairs reasoning graph into a process digital twin to form a government affairs reasoning graph enhanced digital twin; A training module pre-trains a reinforcement learning agent based on historical successful government process cases using the government affairs reasoning graph to obtain an initial process optimization strategy; An optimization module generates a target process optimization strategy through simulation iteration in the government affairs reasoning graph enhanced digital twin based on the initial process optimization strategy and the causal correlation relationship revealed by the government affairs reasoning graph; An abnormality diagnosis module performs causal explanation on the decision logic of the target process optimization strategy using the government affairs reasoning graph, and performs root cause tracing on the process abnormalities monitored in the real government running environment; An updating module deploys the target process optimization strategy verified by the causal explanation to a real government affair operation environment to execute process optimization, and updates and feeds back the enhanced digital twin of the government affair management graph based on performance monitoring data and root cause tracing results.

[0016] The application has the beneficial effects that: by constructing a government affair knowledge graph and a government affair management graph, integrating event causal logic into a process digital twin to form an enhanced digital twin, using the government affair management graph to pre-train an initial optimization strategy of a reinforcement learning intelligent agent, and guiding an exploration process in the enhanced digital twin according to causal correlation, a target optimization strategy is generated through simulation iteration, the strategy decision logic is explained causally by using the government affair management graph, and root cause tracing is performed on process abnormalities, the verified strategy is deployed to a real environment to execute optimization, and the enhanced digital twin is dynamically updated based on running feedback, so that closed-loop optimization and continuous self-evolution of government service processes are realized. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Fig. 1 The flowchart of the government affair service process optimization method based on artificial intelligence.

[0019] Fig. 2 The schematic diagram of the government affair service process optimization system based on artificial intelligence.

[0020] Fig. 3 The flowchart of the government affair knowledge graph construction.

[0021] Fig. 4 The flowchart of the initial process optimization strategy generation. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings of the specification.

[0023] In the following description, many specific details are set forth in order to provide a thorough understanding of the application, but the application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the application, therefore the application is not limited by the specific embodiments disclosed below.

[0024] Second, the "one embodiment" or "an embodiment" referred to herein can include a particular feature, structure, or characteristic. The various embodiments appearing at different places in this specification are not necessarily all cumulative with one another. That is, an embodiment appearing at one place in this specification can or can not be combined with other embodiments appearing elsewhere in this specification.

[0025] Reference Figs. 1-4 For one embodiment of the present application, the embodiment provides an artificial intelligence-based government service process optimization method, comprising the following steps: S1, a government knowledge graph storing static knowledge in the government field is constructed.

[0026] S1.1, multi-source heterogeneous data is obtained from a government data source and cleaned. Based on the cleaned multi-source heterogeneous data, natural language processing technology is used to identify and extract government field entities.

[0027] Further, after the multi-source heterogeneous data obtained from the government data source is cleaned, the named entity recognition method in the natural language processing technology is used for processing. This process is not simply keyword matching, but relies on a language model pre-trained on government field text corpus, which can deeply understand the specific semantic context in government documents, regulations and guidelines. By analyzing the context, grammar structure and specific expression habits of the government field, entities with clear government semantics are accurately identified and extracted.

[0028] S1.2, the static relationship between the identified government field entities is identified and extracted using natural language processing technology.

[0029] Further, based on the identified government field entities, the relationship extraction method in the natural language processing technology is further used to analyze the sentence fragments containing two or more target government field entities. Through semantic role labeling and dependency syntax analysis techniques, the specific static relationship between entities is determined, for example, when processing the text that a company needs to submit a company charter to apply for a business license, the relationship extraction model needs to accurately identify the relationship between the company and the business license. The relationship between the business license and the company charter needs to be submitted. This deep semantic analysis-based relationship extraction goes beyond simple co-occurrence statistics and accurately captures the inherent static association between government field entities defined by regulations or rules, so as to ensure that the constructed government knowledge graph can truly reflect the business rules and knowledge structure of the government field.

[0030] S1.3, the identified government field entities and the static relationship between the government field entities are stored in the form of nodes and edges in the graph database, forming a government knowledge graph storing static knowledge in the government field.

[0031] Further, each identified government affairs field entity is instantiated as an independent node in the graph database, and each node is labeled with its key attribute information extracted from the original data. The static relationship between the identified government affairs field entities is instantiated as a directed edge connecting the corresponding node entities, and each edge is labeled with a specific relationship type. This transformation process integrates the static knowledge in the government affairs field scattered in multiple heterogeneous data sources into an interconnected semantic network, i.e., a government knowledge graph, which is stored in a graph database and can efficiently support complex reasoning operations such as multi-hop relationship query and path discovery, thereby converting unstructured text information into structured knowledge that can be understood and calculated by machines, providing strong knowledge support for upper-layer applications, and forming a government knowledge graph that can store static knowledge in the government affairs field in the form of nodes and edges in a graph database.

[0032] S2, based on the government knowledge graph, constructing the government affairs reasoning graph of event causal chain.

[0033] S2.1, analyzing the government knowledge graph to identify entities and relationships in the government knowledge graph as dynamic event elements, and instantiating them as event nodes.

[0034] Further, semantic analysis is performed on the government knowledge graph to identify entities and relationships that have dynamic and executable properties, and to promote them to event nodes. The government knowledge graph stores a large amount of static facts, but not all entities are suitable for conversion to events. Semantic filtering is performed, for example, the administrative permission items, handling departments, and application material entities in the government knowledge graph are static concepts, but when they are associated with dynamic relationships such as submission, review, and approval indicating actions or state changes, they constitute potential event elements.

[0035] Specifically, the key dynamic relationships are identified, and the participating entities of the relationships are instantiated as specific event nodes. The relationship between the enterprise and the submission and establishment registration application is instantiated as an enterprise submission and establishment registration application event. This method of separating dynamic event elements from static knowledge achieves the conversion from a static knowledge network that describes what to a dynamic event sequence that describes what happens.

[0036] S2.2, based on historical process logs and business rules, mining the causal logic relationship between event nodes, combining the causal logic relationship with event nodes, and constructing the government affairs reasoning graph of event causal chain in the form of event nodes and causal logic relationship edges.

[0037] Further, the discrete event nodes are connected into a chain through causal logical relationships, the timing information in the historical process log and the logical constraints contained in the business rules are comprehensively utilized, the order and conditional probability of the occurrence between the event nodes are counted through analyzing the historical process log, the potential causal or logical relationship is mined, for example, event B has a high probability of following event A, the business rules provide strict logical constraints, for example, the rules may explicitly stipulate that only after the material pre-examination event occurs, the substantive examination event can be triggered, the two kinds of information are fused, the edges with causal semantics between the event nodes are established, for example, it is defined that the event of material incompleteness is the reason leading to the event of pre-examination failure, the event nodes and the causal logical relationship edges are combined, the event causal chain capable of expressing the dynamic logic that because A occurs, B occurs is constructed, and the government affair matter graph generated has not only the event sequence, but also the internal causal driving mechanism between the events, so that the graph has logical reasoning capability, and the government affair matter graph in the form of the event nodes and the causal logical relationship edges is formed.

[0038] S3, the government affair matter graph is integrated into the process digital twin to form a government affair matter graph enhanced digital twin.

[0039] S3.1, a process digital twin is constructed based on historical government process data, and a data mapping rule between the government affair matter graph and the process digital twin is established.

[0040] Further, the construction of the process digital twin is based on historical government process data, the state transition, resource consumption and time consumption of the process instance at each link are recorded, the completed process digital twin can simulate the dynamic execution process of the government process, for example, simulates random behaviors such as user arrival and link processing time fluctuation, a data mapping rule between the government affair matter graph and the process digital twin is established, and how the event nodes and the causal logical relationship in the government affair matter graph correspond to the state variables and the transition conditions in the process digital twin is defined.

[0041] Further, the event of material pre-examination failure in the government affair matter graph needs to be mapped to the audit result state variable of the material pre-examination link in the process digital twin, and the corresponding state transition logic is triggered, the establishment of this mapping rule realizes semantic alignment between the static and declarative domain knowledge (the government affair matter graph) and the dynamic and executable process model (the process digital twin), and ensures that the causal reasoning can act on the specific state of the process simulation.

[0042] S3.2, the causal reasoning logic of the government affair matter graph is embedded into the process digital twin, and the process digital twin is enhanced to a government affair matter graph enhanced digital twin.

[0043] Further, the cause and effect reasoning logic contained in the government affair matter handling graph is converted into decision rules or state transition constraints inside the process digital twin. When the process digital twin is in a certain state during simulation, it will query the government affair matter handling graph in real time, infer the set of events that may or should occur at the next moment according to the current event nodes that have occurred, and convert this inference result into guidance or constraints for the subsequent simulation path of the process digital twin. When the simulation process triggers a material incomplete event, the cause and effect logic of the government affair matter handling graph points out that this is likely to lead to a pre-trial disapproval event. After receiving this inference result, the process digital twin can adjust the probability distribution of its internal state transition, or directly trigger the corresponding state change, so that the process digital twin is no longer just a random simulator based on historical statistical rules, but an intelligent simulation environment that can make cause and effect inferences and understand business consequences. The process digital twin is enhanced to a government affair matter handling graph enhanced digital twin. S4. Based on historical successful government process cases, the government affair matter handling graph is used to pre-train the strategy of the reinforcement learning agent, and an initial process optimization strategy is obtained.

[0044] S4.1. Select historical successful government process cases from government process historical records, and based on the government affair matter handling graph, parse the corresponding event sequence from the historical successful government process cases to form an event causal chain.

[0045] Further, the government process historical records are screened, and those historical successful government process cases that are completed efficiently and in compliance are selected as positive samples. Based on the government affair matter handling graph, the key to parsing the cases is to map the records in the cases to the corresponding event nodes in the government affair matter handling graph. According to the actual execution order of the operations in the cases, the trigger chain between the event nodes is restored, thereby forming an event causal chain verified by successful experience.

[0046] Specifically, a successful enterprise establishment registration case is parsed into a series of event nodes such as submitting application materials event, material pre-trial pass event, qualification verification completion event, and issuing business license event in the government affair matter handling graph, and the original successful time sequence and causal correlation are maintained. The specific, instantiated successful experience is abstracted into an event sequence pattern that has universal guiding significance and conforms to the domain logic.

[0047] S4.2. Convert the event causal chain into the connection weights of the strategy network of the reinforcement learning agent through the imitation learning algorithm, and generate an initial weight matrix.

[0048] Furthermore, an imitation learning algorithm incorporating causal constraints is employed, with its objective function being a composite loss function consisting of two parts: the first part is the standard imitation learning loss, which minimizes the negative log-likelihood, thereby improving the policy network of the reinforcement learning agent. The state experienced by historical success cases Next, select the successful action. probability Maximize, thereby driving the behavioral distribution of the reinforcement learning agent's policy network. Approximating the reference behavior distribution The second part is the innovative causal consistency loss term, which calculates the loss of the reinforcement learning agent's policy network. One's own behavior relative to the political and administrative logic map The expected causal violation degree, derived from the causal violation degree function. Quantify it.

[0049] Specifically, through regularization coefficients By weighting and summing the two losses, the optimization process, while pursuing behavioral imitation, incorporates constraints on the causal rationality of the behavior. This guides the reinforcement learning agent's policy network not only to learn the appearance of historical successful cases, but also to deeply understand the causal logic behind their behavior. This generates an initial weight matrix that has both successful experience and inherently conforms to domain common sense, providing a high-starting-point, low-risk initial optimization strategy for subsequent reinforcement learning.

[0050] The expression for the composite loss function is: ; in, The initial weight matrix, In a given state Time-reinforcement learning agent policy network Select the action to perform. The conditional probability, As a reference to the mathematical expectation of the behavior distribution, For reference behavior distribution, For action, Given a state, To enhance the learning agent policy network, For the political affairs and rationale diagram, For the degree of causality violation, The regularization coefficient is used. S4.3. Use the initial weight matrix to configure the parameters of the reinforcement learning agent policy network to complete the initialization training of the policy network and obtain the initial optimization policy defined by the initial weight matrix.

[0051] Further, the initial weight matrix is assigned to all connection weights of the reinforcement learning agent policy network to complete the policy network initialization training, and the knowledge obtained through the joint optimization of imitation learning and causal constraints is solidified into the parameters of the reinforcement learning agent policy network, so that the initial optimization strategy defined by the initial weight matrix is obtained. This means that the reinforcement learning agent already has a baseline strategy that tends to select historical successful actions and its decision logic conforms to the causal relationship of the government affair rationality graph before it starts interacting with the environment for trial-and-error learning. This greatly improves the starting performance of the reinforcement learning agent and effectively avoids the problem of low initial performance and large exploration blindness when the reinforcement learning starts from scratch, as the initial strategy already has embedded causal logic.

[0052] S5. In the government affair rationality graph enhanced digital twin, the reinforcement learning agent generates a target process optimization strategy through simulation iteration based on the initial process optimization strategy and the causal relationship revealed by the government affair rationality graph.

[0053] S5.1, the initial weight matrix corresponding to the initial optimization strategy is loaded into the reinforcement learning agent policy network, and the reinforcement learning agent dynamically adjusts the action exploration space according to the causal relationship revealed by the government affair rationality graph in the simulation environment of the affair rationality graph enhanced digital twin.

[0054] Further, the initial weight matrix corresponding to the initial optimization strategy is loaded into the reinforcement learning agent policy network, so that the reinforcement learning agent has a preliminary decision-making capability based on historical successful experience. In the simulation environment of the government affair rationality graph enhanced digital twin, the reinforcement learning agent dynamically adjusts the action exploration space according to the causal relationship revealed by the government affair rationality graph. When the simulation process reaches a certain state, the reinforcement learning agent will query the government affair rationality graph in real time, predict a reasonable set of events that may occur in the next stage according to the current event nodes and causal logic relationship, and the reinforcement learning agent policy network will amplify the exploration probability of actions that may lead to favorable results and suppress the exploration probability of actions that are known to violate causal logic or lead to adverse consequences.

[0055] Specifically, when the government affair rationality graph indicates that material pre-examination is a necessary condition for starting on-site investigation, the reinforcement learning agent will reduce the probability of selecting the action of starting on-site investigation in the state where the material pre-examination is not passed. This changes the exploration process of the reinforcement learning agent from pure random trial-and-error to a kind of intelligent search guided by domain knowledge and with a direction, greatly improving the exploration efficiency and effectively avoiding the waste of resources on invalid or high-risk actions.

[0056] S5.2, based on the dynamically adjusted action exploration space, the reinforcement learning agent performs multi-round simulation iteration optimization strategy in the government affair matter graph enhanced digital twin.

[0057] Further, based on the dynamically adjusted action exploration space, the reinforcement learning agent performs multi-round simulation iteration optimization strategy in the government affair matter graph enhanced digital twin. Each iteration contains four stages: strategy execution, reward acquisition, strategy evaluation and strategy update. The reinforcement learning agent selects actions within the exploration space constrained by causal logic, interacts with the government affair matter graph enhanced digital twin, and receives reward signals based on process performance indicators. Strategy gradient reinforcement learning algorithms are used to update the weight parameters of the reinforcement learning agent's strategy network based on cumulative rewards. The government affair matter graph enhanced digital twin provides a safe, controllable and semantically rich simulation environment, allowing the reinforcement learning agent to experience a large number of trial-and-error learning that is costly or risky in the real world.

[0058] Specifically, through multiple iterations, the reinforcement learning agent's strategy network not only learns how to obtain high rewards, but also internalizes the causal laws contained in the government affair matter graph, ultimately converging to an efficient and business logic-compliant target process optimization strategy. By combining causal knowledge guidance with data-driven optimization, a high-quality target process optimization strategy is generated that surpasses the initial strategy and pure historical experience.

[0059] S6, using the government affair matter graph to causally explain the decision logic of the target process optimization strategy, and tracing the root cause of the process anomalies monitored in the real government operation environment.

[0060] S6.1, based on the updated strategy network weight matrix, simulate the key decision sequence of the target process optimization strategy in the matter graph enhanced digital twin.

[0061] Further, the updated strategy network weight matrix corresponding to the trained target process optimization strategy is loaded into the reinforcement learning agent's strategy network and run in the government affair matter graph enhanced digital twin. By inputting the initial state, the reinforcement learning agent's strategy network is driven to make decisions, thereby completely reproducing the complete execution path of the target process optimization strategy in the virtual environment, and recording the key decision sequence from the initial state to the final state.

[0062] Specifically, the government affair matter graph enhanced digital twin provides a high-fidelity, repeatable simulation environment that ensures consistent conditions for each simulation run, thereby stably obtaining the decision trajectory of the strategy. This converts the neural network weight matrix, which is difficult to understand intuitively, into a specific, traceable state-action sequence, providing a clear analysis object for subsequent causal explanation and achieving the transformation from a parameterized model to an interpretable behavior sequence.

[0063] S6.2, generate a causal explanation report describing the basis for the decisions using the critical decision sequence.

[0064] Furthermore, for the critical decision sequence, use the government affair management graph to perform causal correlation analysis. For each decision point in the sequence, i.e., the conversion from state to action, query the event nodes and causal logic relationships related to the state and action in the government affair management graph, thereby explaining the causal rationality of selecting the action.

[0065] Specifically, when the critical decision sequence shows that the parallel approval action is selected in the material pre-examination link state, the analysis process will associate the logic that the material complete event is the reason for starting the parallel approval event in the government affair management graph, and further generate a natural language description that the material pre-examination has passed, so the parallel approval is started to improve efficiency, and summarize into a causal explanation report. The decision logic of the strategy is associated with the recognized causal common sense in the government affair management graph, thereby enhancing the credibility and acceptability of the optimized strategy.

[0066] S6.3, map the process abnormalities monitored in the real government operation environment to the corresponding abnormal event nodes in the government affair management graph.

[0067] Furthermore, the process abnormalities monitored in the real government operation environment are extracted and semantically analyzed, and matched with the pre-defined event node types in the government affair management graph. Process abnormalities usually manifest as a surge in link time consumption abnormalities, abnormal failure of handling results, and resource occupation abnormalities. The abnormal features need to be mapped to abnormal event nodes with clear semantics in the government affair management graph, such as material pre-examination link timeout event, final approval rejection event, and server resource depletion event. A corresponding relationship library of abnormal features and event node types is established, and the observed numerical index abnormalities in the real world are promoted to abnormal events with business semantics that can be understood by the government affair management graph.

[0068] S6.4, based on the event causal chain of the government affair management graph, perform reverse causal reasoning on the abnormal event nodes identified by the process abnormal event mapping to locate the root cause event that caused the abnormal event.

[0069] Furthermore, starting from the abnormal event nodes identified by the mapping, reverse traversal and reasoning are performed on the event causal chain of the government affair management graph. Along the reverse direction of the causal logic relationship edge, each upstream cause event node that may have caused the abnormal event is checked one by one, and the possibility or conditional probability of each cause event node is evaluated, until the most fundamental, usually the most upstream, inducing factor, i.e., the root cause event, is located.

[0070] Specifically, for example, for the final approval veto of abnormal events, reverse reasoning may find that the direct cause is that the on-site investigation is not up to standard, and the root cause of the on-site investigation not up to standard can be traced back to the applicant's inadequate preparation or the on-site investigation standard understanding deviation event. It uses the structured causal knowledge of the government affair reasoning graph to realize the automatic and systematic tracing from the surface abnormal phenomenon to the deep root cause, avoids stopping at the surface symptoms, and thus provides a clear target for precise optimization and rectification.

[0071] S7, deploy the target process optimization strategy verified by causal explanation to the real government running environment to execute process optimization.

[0072] S7.1, convert the updated strategy network weight matrix corresponding to the target process optimization strategy verified by causal explanation into a workflow rule executable in the real government running environment.

[0073] Further, the decision logic contained in the updated strategy network weight matrix corresponding to the target process optimization strategy verified by causal explanation is converted into a workflow rule executable in the real government running environment, the input-output mapping relationship of the strategy network weight matrix is analyzed, and the decision mode of the strategy network selecting high-probability actions in a specific state is mapped to business rules with clear condition judgment and execution path.

[0074] Specifically, for example, the strategy network weight matrix learns to tend to select the green channel action in the state of complete materials and encourage industry, and this decision mode is converted into a workflow rule, the condition part is defined as the application material state being complete and the industry type belonging to the encourage category, and the execution path is defined as routing the process instance to the green channel processing queue, which realizes the smooth transition from data-driven, parameterized intelligent decision model to rule-based, executable workflow definition based on existing government information system, ensuring that the optimization strategy can be applied in a low-risk and highly compatible manner.

[0075] S7.2, deploy the workflow rule to the process execution engine in the real government running environment, and the process execution engine in the real government running environment automatically processes the newly arrived government process instance according to the deployed workflow rule to execute process optimization.

[0076] Further, the converted workflow rules are deployed to the process execution engine of the real government operation environment. After loading the new workflow rules, the process execution engine automatically routes and handles the newly arrived government process instances according to the conditions defined in the rules. When a new enterprise establishment registration application enters the system, the process execution engine automatically checks its material state and industry type. If the rule conditions are met, it is automatically assigned to the green channel, thereby realizing automatic process optimization without human intervention. The optimization strategy verified by simulation and causal explanation is seamlessly embedded into the actual business operation system, so that the optimization results can directly and real-time affect the massive government process instances, improve the overall processing efficiency and consistency, and reduce the error risk introduced by improper manual operation.

[0077] S7.3, in the process optimization execution process, the running state data of the government process instance is collected in real time to monitor the optimization effect.

[0078] Further, after the workflow rules take effect, the government process instances handled in the real government operation environment are monitored throughout the life cycle, and their running state data are collected in real time. These data include key performance indicators such as time consumption of each link, flow path, resource consumption, and final result. The collected data are continuously aggregated and used to calculate various performance indicators to objectively evaluate the actual optimization effect brought by the change of workflow rules. It establishes a feedback channel based on real running data, so that the actual value of the optimization strategy can be quantitatively evaluated, and the entire optimization method can form a complete closed loop from virtual simulation to real deployment and effect verification.

[0079] S8, based on the performance monitoring data and root cause tracing results, the enhanced digital twin of the government affair graph is updated and fed back.

[0080] S8.1, analyze the performance monitoring data and root cause tracing results obtained from the real government operation environment, identify the deviations from the prediction of the enhanced digital twin of the government affair graph, and calibrate the event causal chain in the government affair graph.

[0081] Further, the performance monitoring data and root cause tracing results obtained from the real government operation environment are compared and analyzed, and the links or event sequences with significant differences between the prediction of the enhanced digital twin of the government affair graph and the real running results are identified. The performance monitoring data shows that the actual average time consumption of the material pre-examination link is much higher than the simulation prediction value of the enhanced digital twin of the government affair graph, and the root cause tracing result points out that the delay of the material pre-examination link is mainly due to the inconsistency of the new policy interpretation, which is not depicted in the original government affair graph.

[0082] Specifically, based on this deviation analysis, the event causal chain in the government affair matter handling graph is calibrated, a new causal logic relationship edge is added between the material pre-examination event node and the pre-examination time length extension event node, and the new policy interpretation inconsistency event is taken as the trigger condition of the causal relationship. It continuously verifies and corrects the completeness and accuracy of the event causal chain in the government affair matter handling graph by using real running feedback, so that the domain knowledge model can dynamically adapt to policy changes and business evolution, thereby maintaining its reliability as a basis for simulation and decision-making.

[0083] S8.2, update the simulation parameters of the process digital twin according to the new rules reflected in the calibrated government affair matter handling graph and the efficiency monitoring data.

[0084] Further, according to the new rules reflected in the calibrated government affair matter handling graph and the efficiency monitoring data, the simulation parameters of the process digital twin are updated. The calibrated government affair matter handling graph clearly shows that the new policy interpretation inconsistency event will lead to the extension of the processing time of the material pre-examination link. Combined with the statistical distribution of the actual time consumption of this link in the efficiency monitoring data, the parameters of the service time probability distribution model corresponding to the material pre-examination activity in the process digital twin need to be refitted, and the parameters of the resource scheduling strategy simulation in the process digital twin may also need to be adjusted according to the new resource consumption mode, ensuring that the dynamic behavior model of the process digital twin can accurately reflect the latest characteristics of the real environment, making the simulation environment and the real environment keep synchronized in statistical behavior, and providing a high-fidelity basis for subsequent simulation-based analysis, prediction and optimization.

[0085] S8.3, the calibrated government affair matter handling graph and the updated simulation parameters of the process digital twin are re-integrated to form an updated government affair matter handling graph enhanced digital twin.

[0086] Further, the calibrated government affair matter handling graph and the updated simulation parameters of the process digital twin are re-integrated to form an updated government affair matter handling graph enhanced digital twin, and the mapping and driving relationship between the updated event causal logic in the calibrated government affair matter handling graph and the adjusted simulation components in the updated process digital twin is re-established. The newly added new policy interpretation inconsistency event and its causal association need to be mapped to the rule that triggers the specific delay logic in the process digital twin. It completes the closed loop from real world observation feedback to virtual model iterative optimization, making the government affair matter handling graph enhanced digital twin a living being that can continuously evolve with the changes of business environment, and continuously enhancing the simulation prediction and decision support capability, providing an intelligent foundation for the dynamic adaptation of continuous optimization of government affair processes.

[0087] The embodiment also provides an artificial intelligence-based government service process optimization system, which comprises: a government knowledge graph construction module, which constructs a government knowledge graph storing static knowledge in the government field; a construction module configured to construct a government affair matter graph based on the government affair knowledge graph, integrate the government affair matter graph into the process digital twin to form a government affair matter graph enhanced digital twin; a training module configured to pre-train a reinforcement learning agent based on historical successful government affair process cases and the government affair matter graph to obtain an initial process optimization strategy; an optimization module configured to guide an exploration process of the reinforcement learning agent based on the initial process optimization strategy and the causal correlation relationship revealed by the government affair matter graph in the government affair matter graph enhanced digital twin, and generate a target process optimization strategy through simulation iteration; an abnormality diagnosis module configured to perform causal explanation on a decision logic of the target process optimization strategy and trace a root cause of a process abnormality monitored in a real government affair running environment based on the government affair matter graph; an updating module configured to deploy the target process optimization strategy verified through the causal explanation to a real government affair running environment to perform process optimization, and update the government affair matter graph enhanced digital twin based on performance monitoring data and the root cause tracing result.

[0088] The embodiment also provides a computer device suitable for the government affair service process optimization method based on artificial intelligence, which comprises a memory and a processor.

[0089] The computer device can be a terminal, and the computer device comprises a processor, a memory, a communication interface, a display screen and an input device which are connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with external terminals. The wireless communication can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, a trackball or a touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, a touchpad or a mouse, etc.

[0090] The embodiment also provides a storage medium on which a computer program is stored, the program being executed by a processor to implement the method for optimizing a government affair service process based on artificial intelligence proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0091] To sum up, the application forms an enhanced digital twin by integrating event causal logic into a process digital twin through constructing a government affair knowledge graph and a government affair affair graph, pre-trains an initial optimization strategy for a reinforcement learning intelligent agent by using the government affair affair graph, guides an exploration process in the enhanced digital twin according to causal correlation, generates a target optimization strategy through simulation iteration, performs causal explanation on strategy decision logic and root cause tracing on process abnormalities by using the government affair affair graph, deploys the verified strategy to a real environment to perform optimization, and dynamically updates the enhanced digital twin based on running feedback, so as to realize closed-loop optimization and continuous self-evolution of the government affair service process.

[0092] It should be noted that the above embodiments are only used to illustrate the technical solutions of the application but not limit the application, although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the application, and all of them should be covered in the scope of the claims of the application.

Claims

1. A method for optimizing government service processes based on artificial intelligence, characterized by: This includes constructing a government knowledge graph that stores static knowledge in the government affairs domain; Based on the aforementioned government knowledge graph, a government affairs logic graph of event causal chains is constructed, and the government affairs logic graph is integrated into the process digital twin to form a government affairs logic graph enhanced digital twin; Based on successful historical government process cases, the government process graph is used to pre-train the reinforcement learning agent to obtain the initial process optimization strategy. In the digital twin of the government affairs graph, the reinforcement learning agent is guided to explore the process based on the causal relationship revealed by the government affairs graph, and generates the target process optimization strategy through simulation iteration, based on the initial process optimization strategy. Using the political affairs graph, we can provide a causal explanation for the decision-making logic of the target process optimization strategy and trace the root causes of process anomalies monitored in the real political affairs operation environment. The target process optimization strategy, verified by causal explanation, is deployed to the real government operation environment to optimize the process. Based on the performance monitoring data and root cause tracing results, the digital twin of the government affairs logic graph is updated and fed back.

2. The method for optimizing government service processes based on artificial intelligence as described in claim 1, characterized in that: Constructing a government knowledge graph that stores static knowledge in the government domain includes the following steps: The system obtains heterogeneous data from multiple sources from government data sources and cleans it. Based on the cleaned heterogeneous data, it uses natural language processing technology to identify and extract entities in the government domain. Among the identified entities in the government domain, natural language processing technology is used to identify and extract the static relationships between them. The identified entities in the government domain and the static relationships between them are stored in a graph database in the form of nodes and edges, forming a government knowledge graph that stores static knowledge in the government domain.

3. The method for optimizing government service processes based on artificial intelligence as described in claim 2, characterized in that: Based on the aforementioned government knowledge graph, a government event logic graph of event causal chains is constructed, including the following steps: The government knowledge graph is analyzed to identify entities and relationships that serve as dynamic event elements, and these are instantiated as event nodes. Based on historical process logs and business rules, we mine the causal logical relationships between event nodes, combine the causal logical relationships with event nodes, and construct a government affairs logic graph of event causal chains in the form of event nodes and causal logical relationship edges.

4. The method for optimizing government service processes based on artificial intelligence as described in claim 3, characterized in that: Integrating the administrative affairs graph into the process digital twin to form an enhanced digital twin of the administrative affairs graph includes the following steps: A digital twin of processes is constructed based on historical government affairs process data, and data mapping rules are established between the government affairs logic graph and the digital twin of processes. The causal reasoning logic of the government affairs logic graph is embedded into the process digital twin, and the process digital twin is enhanced into an enhanced digital twin of the government affairs logic graph.

5. The method for optimizing government service processes based on artificial intelligence as described in claim 4, characterized in that: Based on successful historical government process cases, a government affairs logic graph is used to pre-train a reinforcement learning agent to obtain an initial process optimization strategy, including the following steps: Successful historical government process cases are selected from the historical records of government processes. Based on the government affairs graph, the corresponding event sequences are analyzed from the successful historical government process cases to form event causal chains. The event causal chain is transformed into connection weights of the reinforcement learning agent policy network through imitation learning algorithm, and an initial weight matrix is ​​generated. The initial weight matrix is ​​used to configure the parameters of the reinforcement learning agent's policy network to complete the initial training of the policy network and obtain the initial optimized policy defined by the initial weight matrix.

6. The method for optimizing government service processes based on artificial intelligence as described in claim 5, characterized in that: In the digital twin augmented with a government affairs logic graph, the reinforcement learning agent, based on an initial process optimization strategy, guides the exploration process according to the causal relationships revealed by the government affairs logic graph, and generates a target process optimization strategy through simulation iteration, including the following steps: The initial weight matrix corresponding to the initial optimization strategy is loaded into the policy network of the reinforcement learning agent. In the simulation environment of the digital twin with enhanced event graph, the reinforcement learning agent dynamically adjusts its action exploration space based on the causal relationships revealed by the event graph. Based on the dynamically adjusted action exploration space, the reinforcement learning agent performs multi-round simulation iteration optimization strategies in the event graph-enhanced digital twin.

7. The method for optimizing government service processes based on artificial intelligence as described in claim 6, characterized in that: Using a government affairs process graph, the decision-making logic of the target process optimization strategy is explained causally, and the root causes of process anomalies monitored in the real government affairs operation environment are traced, including the following steps: Based on the updated policy network weight matrix, the key decision sequence of the target process optimization strategy is simulated and reproduced in the event graph-enhanced digital twin. By utilizing key decision sequences, generate causal explanation reports describing the basis for decision-making; Map process anomalies detected in real government operations to corresponding abnormal event nodes in the government affairs logic graph; Based on the event causal chain of the government affairs graph, reverse causal reasoning is performed on the abnormal event nodes identified by the mapping of abnormal events in the process to locate the root cause event that caused the abnormal event.

8. The method for optimizing government service processes based on artificial intelligence as described in claim 7, characterized in that: Deploying the target process optimization strategy, validated by causal explanation, to the real-world government operational environment to perform process optimization includes the following steps: The updated strategy network weight matrix corresponding to the target process optimization strategy verified by causal explanation is converted into a workflow rule executable in the real government operation environment; Workflow rules are deployed to the process execution engine in the real government operation environment. The process execution engine in the real government operation environment automatically processes newly arriving government process instances and performs process optimization based on the deployed workflow rules. During the process optimization, real-time data on the operational status of government process instances is collected to monitor the optimization effect.

9. The method for optimizing government service processes based on artificial intelligence as described in claim 8, characterized in that: The digital twin of the government affairs management graph is updated and fed back based on performance monitoring data and root cause analysis results, including the following steps: By analyzing the performance monitoring data and root cause analysis results obtained from the real government operation environment, deviations were identified from the predictions of the enhanced digital twin of the government affairs graph, and the event causal chain in the government affairs graph was calibrated. Based on the new patterns reflected in the calibrated administrative affairs graph and performance monitoring data, update the simulation parameters of the process digital twin; The calibrated administrative affairs graph is re-integrated with the simulation parameters of the updated process digital twin to form an updated administrative affairs graph enhanced digital twin.

10. An AI-based government service process optimization system, based on the AI-based government service process optimization method according to any one of claims 1 to 9, characterized in that: This includes a government affairs knowledge graph construction module, which constructs a government affairs knowledge graph that stores static knowledge in the government affairs field; The construction module, based on the aforementioned government knowledge graph, constructs a government affairs logic graph of event causal chains, and integrates the government affairs logic graph into the process digital twin to form a government affairs logic graph enhanced digital twin; The training module, based on historical successful government process cases, uses a government affairs logic graph to pre-train the reinforcement learning agent on the strategy to obtain the initial process optimization strategy. In the optimization module, within the enhanced digital twin of the government affairs logic graph, the reinforcement learning agent, based on the initial process optimization strategy, guides the exploration process according to the causal relationships revealed by the government affairs logic graph, and generates the target process optimization strategy through simulation iteration. The anomaly diagnosis module uses the government affairs logic graph to explain the causal logic of the decision-making logic of the target process optimization strategy and to trace the root causes of process anomalies monitored in the real government affairs operation environment. The update module deploys the target process optimization strategy, verified by causal explanation, to the real government operation environment to optimize the process. Based on the performance monitoring data and root cause tracing results, it updates and provides feedback to the enhanced digital twin of the government affairs logic graph.

Images