A water network graph atlas and agent cooperative business analysis method

Abstract

The application provides a water network graph and intelligent agent collaborative business analysis method. It relates to the technical field of intelligent water conservancy. The method comprises: mapping the spatial influence range and business processing type of the water network business problem into physical topology constraints and business transmission constraints respectively; taking water network entities, business links and intelligent agents as nodes; taking the anchoring relationship between business links and water network entities, and the execution relationship between intelligent agents and business links as edges, constructing a water network global graph; taking the target water network entity as the starting point, traversing along the edges of the water network global graph, and in the traversal process, when the water network entity exceeds the physical topology constraint or the business link exceeds the business transmission constraint, the corresponding node and other type nodes related to the node are removed; constructing a local graph based on the nodes and edges retained after traversal; executing the business link based on the intelligent agent in the local graph to obtain the analysis result of the water network business problem.

Description

Technical Field

[0001] This invention relates to the field of smart water conservancy technology, and in particular to a business analysis method that combines water network maps with intelligent agents. Background Technology

[0002] Water network business analysis refers to comprehensive reasoning based on the operational status of physical engineering facilities (such as reservoirs, rivers, and sluice gates / pumps) within a water network system and related business logic (such as flood control scheduling and water resource allocation). This provides decision support for complex scenarios such as basin flood control and inter-regional water transfer, thereby ensuring the safety and stability of the water network operation. With the advancement of smart water conservancy construction, knowledge graph technology has been introduced into the field of water network business analysis. It integrates heterogeneous data such as the physical topology of the water network and historical business processes into a global knowledge base, which is then used by intelligent agents or large models to execute business processes.

[0003] However, there are complex cross-relationships between the physical engineering facilities and related business logic in the water network. Inputting the global knowledge base into the agent makes it difficult for it to lock in effective information that is strongly related to the current problem from the massive intertwined relationships. This leads to context information overload and divergent reasoning paths. This divergent state makes the analytical conclusions output by the agent unreliable, resulting in insufficient accuracy and reliability of water network business analysis, and making it difficult to provide effective support for actual water conservancy project decision-making. Summary of the Invention

[0004] Therefore, it is necessary to provide a business analysis method that integrates water network mapping with intelligent agents to address the aforementioned technical issues.

[0005] The following technical solution is adopted in this specification: This specification provides a business analysis method for the collaboration of water network maps and intelligent agents, including: Using water network entities, business processes, and intelligent agents as nodes, and the anchoring relationship between water network entities and business processes, the process relationship between business processes, and the execution relationship between intelligent agents and business processes as edges, a global business graph of the water network is constructed. Obtain water network business analysis requests, identify the target water network entity and target business type corresponding to the water network business analysis requests, and extract the business process expansion direction and hop range that match the target business type from the preset strategy library; In the global business graph of the water network, starting from the business links directly associated with the target water network entity, the path is extended along the business process extension direction. Business links that exceed the hop count range during the path extension process are filtered out, as well as water network entities and intelligent agents that have anchoring or execution relationships with the filtered business links. Candidate business paths formed by the remaining business links are obtained. The candidate business paths are closed and verified. The graph formed by the candidate business paths that pass the closure verification and their associated water network entities and intelligent agents is used as a local graph. Based on the intelligent agents in the local graph, the business processes in the local graph are executed to obtain the analysis results of the water network business analysis request.

[0006] Furthermore, the step of extracting the business process extension direction and hop range matching the target business type from the preset strategy library includes: Match the target business type with multiple business process templates in the preset strategy library, and obtain the target business process template corresponding to the target business type based on the matching results; the business process template at least defines the expansion direction and hop range of the business process corresponding to the business type. The extension direction defined in the target business process template is used as the business process extension direction. The extension direction includes tracing upstream along the process relationship, extending downstream, or bidirectional extension.

[0007] Furthermore, the path expansion along the business process extension direction, starting from the business links directly associated with the target water network entity, includes: Starting with the target water network entity, the first round of expansion is carried out along the anchoring relationship to obtain all business link nodes directly related to the target water network entity; Starting with all the business process nodes obtained, a second round of expansion is carried out synchronously along the process relationship and in the direction of business process expansion. During the expansion process, each business process node extends downstream or upstream along its respective process relationship.

[0008] Furthermore, business segments exceeding the hop count limit during path expansion will be filtered, as will water network entities and intelligent agents with anchoring or execution relationships to the filtered business segments, including: During the path expansion process, record the current hop count of each business node from the starting point; When the current hop count of any business process node exceeds the hop count range, the business process node is marked as an out-of-range node and removed. Using the removed business process node as the trigger point, search in reverse along the anchoring relationship for water network entity nodes that have an anchoring relationship with the removed business process node and remove them; Using the removed business process node as the trigger point, search backwards along the execution relationship for intelligent agent nodes that have an execution relationship with the removed business process node and remove them.

[0009] Furthermore, the closure verification of the candidate service path includes: Traverse each business link node in the candidate business path and check whether the business link node is connected to the corresponding water network entity node through anchoring relationship and connected to the corresponding intelligent agent node through execution relationship in the candidate business path. If all business process nodes in the candidate business path meet the above conditions, then the candidate business path is determined to have passed the closure check. If any business link node in the candidate business path does not meet the above conditions, the candidate business path is determined to have failed the closure check.

[0010] Furthermore, the identification of the target water network entity and target service type corresponding to the water network service analysis request includes: Entity identification is performed on water network business analysis requests to obtain the physical entity names in the water network business analysis requests. Based on the physical entity names, a search and matching is performed in the global water network business map, and the matched water network entity nodes are used as target water network entities. The intent analysis of water network business analysis requests is performed, and the results of the intent analysis are matched with a preset set of business types to determine the target business type.

[0011] Furthermore, the node also includes an algorithm model node, and the edge also includes the authorization relationship between the algorithm model and the agent; The construction of the global water network business map includes: A physical entity graph is constructed using water network entities as nodes and the upstream and downstream connections and control relationships between water network entities as edges. Based on the physical entity map, the business process is linked to the corresponding water network entity through the anchoring relationship between the business process and the water network entity, and a business relationship map is constructed by combining the process relationship between the business processes. Based on the business relationship graph, the intelligent agents are associated with the corresponding business processes through the execution relationship between the intelligent agents and the business processes, and the intelligent agent relationship graph is constructed by combining the collaborative relationship between the intelligent agents. Based on the agent relationship graph, the scope of the agent's algorithm model call is limited by the authorization relationship between the algorithm model and the agent, and a tool graph is constructed by combining the call relationship between the algorithm models. By integrating the physical entity graph, business relationship graph, agent relationship graph, and tool graph, a global business graph of the water network is obtained.

[0012] Furthermore, the process of the agent in the local graph executing the business processes within the local graph to obtain the analysis results of the water network business analysis request includes: Based on the execution relationship between agent nodes and business process nodes in the local graph, the expert agents participating in each business process in the business path are identified, and the algorithm models in the local graph that have authorization relationships with agent nodes are registered as the computing tools of the expert agents. Construct a host agent, which determines the order of business processing based on the process relationships between business links in the local graph, and schedules the corresponding expert agents according to the order of business processing. The expert agent responds to the scheduling instructions of the host agent, calls the registered computing tools to execute the corresponding business steps, and obtains the sub-results; The host agent collects the sub-results output by each expert agent and merges them to generate the analysis results for the water network business analysis request.

[0013] This specification provides a business analysis device that integrates water network mapping with intelligent agents, including: The global water network graph construction module is used to construct a global water network business graph with water network entities, business processes, and intelligent agents as nodes, and the anchoring relationship between water network entities and business processes, the process relationship between business processes, and the execution relationship between intelligent agents and business processes as edges. The business problem analysis module is used to obtain water network business analysis requests, identify the target water network entity and target business type corresponding to the water network business analysis request, and extract the business process extension direction and hop range that match the target business type from the preset strategy library; The local graph construction module is used to expand the path in the global water network business graph, starting from the business links directly associated with the target water network entity and extending along the business process. It filters out business links that exceed the hop count range during the path expansion process, and filters out water network entities and intelligent agents that have anchoring or execution relationships with the filtered business links. It obtains candidate business paths formed by the remaining business links, performs closure verification on the candidate business paths, and uses the graph formed by the candidate business paths that pass the closure verification and their associated water network entities and intelligent agents as the local graph. The analysis module is used to execute business processes within a local graph based on agents in that local graph, and to obtain the analysis results of water network business analysis requests. This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned business analysis method for the collaboration between water network maps and intelligent agents.

[0014] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: This invention determines a search range matching the water network business analysis request within the global water network business graph by identifying the target water network entity, the business process expansion direction, and the hop count range. Within this search range, cascading filtering and closure verification ensure that the final local graph does not exhibit logical loops such as "business without object" or "business without execution." It can precisely narrow down the global graph to a dedicated local graph that is highly relevant to the water network business analysis request and has a complete process. This reduces the complexity of large-scale model reasoning problems to deterministic execution problems on a constrained graph, overcoming the context overload and reasoning divergence problems caused by full data input, and improving the accuracy and reliability of water network business analysis. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0016] Figure 1 This is one of the flowcharts illustrating a business analysis method that combines water network mapping with intelligent agents, as provided in this specification. Figure 2 This document provides a schematic diagram of a process for constructing a global water network map. Figure 3 This document provides a schematic diagram of a multi-agent collaborative workflow. Figure 4 This is the second flowchart illustrating a business analysis method that combines water network mapping with intelligent agents, as provided in this specification. Figure 5 A schematic diagram of a business analysis device that integrates water network mapping and intelligent agents, provided in this specification. Figure 6 This is a schematic diagram of a computer device provided for this specification. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0018] The technical solution provided by this invention aims to address the challenges of insufficient problem organization capabilities in existing water network systems, lack of structured semantic support in large language models, and lack of effective control over multi-step, multi-role collaboration. First, it unifies the modeling of physical entities, business logic, agent collaboration, and tool capability relationships to form a structured semantic foundation. When a user raises a question, the system parses the target object and intent, generating a local graph oriented towards the current problem through scope templates, extension rules, and closure verification and pruning, automatically determining the analysis scope and business path. Then, based on the local graph, a unified task context is constructed, dynamically generating the skill sets of each agent, transforming structural information and constraints into understandable and callable task semantics. Finally, the host agent uniformly arranges tasks, assigns responsibilities, and constrains tool calls, ensuring execution stability through anomaly identification and expert intervention mechanisms, and structurally accumulating and reusing corrective experiences. Through the above design, it achieves automatic organization and standardized collaborative execution of problem-oriented processes in complex water network business scenarios, improving the accuracy, consistency, and controllability of analysis and decision-making.

[0019] The following description, in conjunction with the accompanying drawings, describes the business analysis method of water network mapping and intelligent agent collaboration of the present invention.

[0020] Figure 1 This is one of the flowcharts illustrating a business analysis method for the collaboration of water network maps and intelligent agents, as provided in this specification. Figure 1 As shown, the method includes: S100. Using water network entities, business processes, and intelligent agents as nodes, and the anchoring relationship between water network entities and business processes, the process relationship between business processes, and the execution relationship between intelligent agents and business processes as edges, a global business graph of the water network is constructed.

[0021] S110, Construct a physical entity map.

[0022] The physical entity map nodes cover hydraulic engineering objects such as reservoirs, rivers, canals, sluice gates, pumping stations, control sections, irrigation districts, water supply areas, and ecological sections. Each physical entity node defines attributes such as a unique identifier, object type, spatial location, upstream and downstream affiliation, watershed region, and object level. The object level distinguishes the spatial or managerial granularity of the object; for example, a watershed belongs to the watershed level, river B to the engineering level, and section D to the monitoring point level. Different level attributes support subsequent granularity-based retrieval and relationship filtering. Edges between nodes include upstream / downstream connections, branching relationships, inclusion relationships, service relationships, and control relationships.

[0023] Taking a segment of a watershed as an example, Reservoir A, River B, and Gate C are all engineering-level entities. Cross-section D is a monitoring point-level entity, and Irrigation District E is an engineering-level entity. The Reservoir A node records its normal water level and storage capacity curves; the River B node records its cross-sectional water level-discharge relationship curve; and the Gate C node records its opening degree-flow capacity correspondence table. The edges are constructed as follows: River B points to Reservoir A through its upstream inflow relationship; Reservoir A points to Gate C through its downstream outflow relationship; Gate C points to Cross-section D through its control relationship; and Cross-section D points to Irrigation District E through its water supply service relationship.

[0024] S120. Construct a business relationship graph.

[0025] Building upon the physical entity graph, the business relationship graph nodes, in addition to specific business processes such as rainfall runoff calculation, reservoir inflow forecasting, water supply capacity assessment, and scheduling plan generation, also include higher-level business objects such as business domains and business objectives. Business domains are used to categorize major business areas, such as flood control scheduling, water resource allocation, and ecological protection. Business objectives describe specific quantitative indicators, such as a water supply guarantee rate of no less than 95% and an ecological flow of no less than 5 m³ / s at irrigation district sections.

[0026] Edges are categorized into three types: The first type is anchoring edges, used to connect business nodes and physical entity nodes. This anchoring is two-tiered: business objectives are anchored to the target physical entity they influence or serve, such as the irrigation district's water supply guarantee rate being anchored to irrigation district E; business processes are anchored to the physical objects upon which their execution depends, such as a task being anchored to river channel B or cross-section D, thus forming a multi-level anchoring chain of business objectives → business tasks → physical entities. The second type is dependency edges, connecting business processes, including pre- and post-dependencies, input / output transmission, parallel collaboration relationships, and constraint relationships. Parallel collaboration edges indicate that two tasks are independent but need to be completed collaboratively within the same time window, such as flood control pre-discharge calculations and power generation planning being carried out in parallel, with the final results jointly input into the integrated scheduling task. The third type is constraint edges, connecting business processes and rule-constrained nodes, limiting the task execution conditions and boundaries.

[0027] A business domain node, Water Resources Allocation, is defined, with a business objective node T1: Irrigation District Water Supply Guarantee Rate ≥ 95%, anchored to Irrigation District E. Business process nodes are defined as follows: Task 1: Inflow Forecasting; Task 2: Reservoir Inflow Calculation; Task 3: Water Supply Assessment; Task 4: Scheduling Decision. Anchoring edges are as follows: Task 1 is anchored to River Channel B and Cross-section D; Task 2 is anchored to Reservoir A; Task 3 is anchored to Irrigation District E; Task 3 is also anchored to business objective T1 through target association, indicating that the water supply assessment is to verify whether T1 is met; Task 4 is anchored to Gate C. Dependency edges are as follows: Task 1 points to the pre-dependency edge of Task 2; Task 2 and Task 3 respectively point to the input dependency edges of Task 4; Additionally, a parallel collaborative edge is added between Task 3 and the Power Generation Planning task, both of which must be completed before 8:00 AM daily, with the results merged for use by Task 4. Specifically, the constraint edge is: add a constraint edge from the ecological base flow constraint of the rule node to Task 4, which limits the gate opening scheme to ensure that the discharge flow at section D is not less than 5m³ / s.

[0028] S130, Construct a collaborative graph of intelligent agents.

[0029] Based on the business relationship graph, the agent collaboration graph nodes include agent nodes and their role object nodes. Agent nodes represent specific functional entities, such as hydrological forecasting agents, irrigation analysis agents, integrated scheduling agents, and gate control agents. Role object nodes represent the functional positioning of an agent in a specific scenario, such as executor, evaluator, reviewer, and decision-maker. The same agent can be associated with different roles in different business scenarios. Each node also defines its scope of responsibility, the types of business it can participate in, and the scope of applicable objects.

[0030] Edges fall into two categories: The first category is execution edges and review edges, which connect agents and business processes by using role objects as intermediate nodes, forming an assignment chain of agent → role → task, whereby an agent executes or reviews the task with a specific role. The second category is collaboration edges, which connect agents or roles, representing triggering, master-slave, or result transmission relationships.

[0031] In one embodiment, the following intelligent agent nodes are defined: P is the forecasting intelligent agent, I is the irrigation intelligent agent, S is the scheduling intelligent agent, and G is the gate execution intelligent agent. Role nodes are defined: R1 is the executor, R2 is the evaluator, and R3 is the decision-maker. Specifically, the execution edge is as follows: the forecasting intelligent agent P executes tasks 1 and 2 as role R1; the irrigation intelligent agent I executes task 3 as role R1, and simultaneously performs an internal self-check on the calculation results of task 3 as role R2; the scheduling intelligent agent S executes task 4 as role R3. Specifically, the cooperation edge is as follows: after completing task 2, the forecasting intelligent agent P triggers the scheduling intelligent agent S to start by transmitting the inflow result through the result transmission cooperation edge; after completing task 3, the irrigation intelligent agent I sends the irrigation area water shortage information to S; after generating the gate opening scheme, S issues instructions to G through the master-slave cooperation edge, and G is responsible for converting the opening value into a specific operation sequence.

[0032] S140. Construct the algorithm model graph.

[0033] Based on the agent collaboration graph, the algorithm model graph nodes represent specific algorithm models or computational tools. In addition to defining the input parameter structure, output parameter structure, and applicable physical object types, each node must also define the applicable problem scenario, such as whether the model is only applicable during the dry season, only applicable during the flood season, or applicable to areas with / without data, to limit the tool's invocation context.

[0034] Edges are divided into two categories: the first category is authorized edges, which connect algorithm models and agents and limit the agent's access rights to call the model; the second category is computation link dependent edges, which connect algorithm models. In addition to specifying the order of calls, they also indicate the combination method, including serial, parallel, and weighted fusion, to support multiple tools to assemble into computation links according to different topologies.

[0035] In one specific embodiment, the algorithm model nodes are defined as follows: M1 is the watershed runoff generation model, applicable to river cross-sections, applicable to data-rich watersheds in humid regions, with rainfall as input and runoff depth as output; M2 is the river confluence model, applicable to river cross-sections, applicable to the dry season of small and medium-sized rivers, with upstream flow as input and downstream flow as output; M3 is the reservoir water balance scheduling model, applicable to reservoirs, applicable to the regular scheduling period, with inflow and water demand as input and discharge and gate opening as output; M4 is also defined as the interval inflow empirical estimation model, and M5 is defined as the inflow peak direct calculation model. The authorization edges are as follows: forecasting agent P is authorized for M1 and M2; scheduling agent S is authorized for M3; irrigation agent I is not authorized for any of the above models. The calculation of the link dependency edge is as follows: a serial dependency edge is established from M1 to M2, and marked as a serial combination method, indicating that it must be run sequentially; a dependency edge is established from M2 and M4 to M3, and marked as a weighted fusion combination method, where the weight of the result of M2 is 0.7 and the weight of the result of M4 is 0.3, indicating that the inflow of M3 is input after weighted combination of the two; the above link dependency edge also specifies that the applicable scenario is the dry season, that is, this combination link is only effective during the dry season, while during the flood season, it switches to another parallel link, that is, M5 directly calculates the inflow flood peak, ensuring that the link selection automatically matches the applicable scenario.

[0036] S150, integrate to obtain a global map of the water network.

[0037] Figure 2 This specification provides a flowchart illustrating the process of constructing a global water network map, such as... Figure 2 As shown, after completing the four sub-graphs, the nodes and edges in each sub-graph are imported into the same graph database. During integration, physical entities serve as the underlying anchor points, and cross-layer associations are completed along multi-layer paths: Starting from any physical entity node, such as reservoir A or river B, the associated business objectives and processes can be located through anchoring relationships, such as water inflow prediction, water supply assessment, and scheduling decisions; from the business process node, the agent role and corresponding agent entity undertaking the task can be found through execution relationships, such as a prediction agent, irrigation agent, or scheduling agent; from the agent entity, the set of algorithm models it is authorized to invoke can be determined through authorization relationships, such as prediction models, optimization models, or computing tools; based on this, further, according to the current problem scenario, such as dry season or flood season, the applicable computing links and combinations are selected from the set of algorithm models. Through the complete tracing path of "physical entity → business task → agent → algorithm model," the four-layer sub-graphs are uniformly associated into a global water network graph, realizing full-link query and standardized execution for specific business problems.

[0038] S200: Obtain the water network business analysis request, identify the target water network entity and target business type corresponding to the water network business analysis request, and extract the business process extension direction and hop range that match the target business type from the preset strategy library.

[0039] For example, the user-proposed water network business problem is mapped to the global water network graph. Constraints are generated through problem analysis, and then the global graph is trimmed according to the constraints to generate a local four-layer sub-graph structure tailored to the current problem. Specifically, this includes the following steps:

[0040] S210, Analysis of Water Network Business Issues.

[0041] Named entity recognition model is used to identify entities in water network business issues. The physical entity names in the problem description are extracted and retrieved in the water network global graph node index by string matching or word vector similarity calculation. The matched water network entity nodes are used as target water network entities.

[0042] A text classification model is used to analyze the intent of the questions, outputting category labels for the business processing type. These category labels are then matched with a pre-defined business type-spatial range mapping table to determine the business processing type. A pre-defined water network business intent database contains various business intents and their corresponding standard processing types. For example, "predicting the inflow of a reservoir over the next three days" matches the "inflow prediction" type in the intent database; "assessing whether the current available water supply in an irrigation district meets crop water requirements" matches the "water supply assessment" type; and "formulating the opening plan for a gate over the next 24 hours" matches the "dispatch decision" type.

[0043] Based on the scope requirements of the business processing type and the target water network entity, the spatial impact range of the water network business problem is inferred. For example, when the business processing type is "inflow prediction," its scope requirement typically includes the upstream catchment area of ​​the target cross section; if the target water network entity is cross section D, the spatial impact range is inferred to be the entire catchment area above cross section D. When the business processing type is "dispatch decision," its scope requirement typically includes the upstream inflow area of ​​the target project, the project itself, and the downstream water receiving area; if the target water network entity is gate C, the spatial impact range is inferred to be all project objects from the upstream river channel B to the downstream irrigation area E of gate C.

[0044] S220, Generate constraints.

[0045] The parsing results are transformed into physical topology constraints and business transit constraints, and then the physical topology constraints and business transit constraints are further structured into a complete rule set that can drive graph traversal. This rule set includes the five-dimensional constraints of the scope template and specific extension strategy constraints.

[0046] S221. Map the spatial influence range to the physical topology level of the global water network map to generate physical topology constraints. This mapping specifically includes:

[0047] Based on the type and spatial influence range of the target water network entity, the permitted upstream and downstream connection relationships and control relationships are determined as the allowed extended upstream and downstream connection and control relationships between water network entities. For example, if the target entity is gate C and its spatial influence range is from the upstream river channel to the downstream irrigation area, then the permitted relationships include the upstream inflow relationship from river B to reservoir A, the downstream outflow relationship from reservoir A to gate C, the control relationship from gate C to section D, and the water supply service relationship from section D to irrigation area E; while other relationship types such as branching relationships and inclusion relationships are not allowed.

[0048] The spatial geographic boundary or the pre-defined engineering level boundary corresponding to the spatial impact range is used as the spatial area or spatial level range of the water network entities that can be included. For example, when the spatial impact range is the catchment area above section D, the spatial area that can be included is all objects within the watershed boundary of that catchment area; when the spatial impact range is irrigation area E, the spatial level range that can be included is irrigation area E and the water source projects it directly depends on, without extending to more fine-grained objects such as crop planting areas inside the irrigation area.

[0049] Based on the preset graph traversal hop count threshold corresponding to the business processing type, this hop count threshold is used as the upper limit of the number of layers traced outward from the target water network entity. For example, when the business processing type is water inflow prediction, the preset hop count threshold is 3 hops, that is, no more than 3 layers of relationship edges are traced upstream from the target section D; when the business processing type is scheduling decision, the preset hop count threshold is 5 hops, that is, no more than 5 layers of relationship edges are traced upstream and downstream from the target gate C simultaneously.

[0050] S222. Method for generating service transit constraints: Service processing types are mapped to the service dependency hierarchy of the global water network graph to generate service transit constraints. This mapping specifically includes:

[0051] A preset business process template is matched based on the business processing type. The preset business process template defines the standard business steps required for that business processing type and their execution order. For example, the business process template for the water inflow forecasting type is defined as: rainfall runoff calculation → river confluence calculation → cross-sectional water inflow forecast. The business process template for the dispatching decision type is defined as: parallel execution of water inflow forecasting and water supply assessment → dispatching scheme generation → scheme verification, where the scheme verification step is limited by ecological baseflow constraints.

[0052] Extract the preceding and following process relationships, input-output relationships, and parallel collaboration relationships between business steps from the business process template, as the process relationships that can be extended between business steps. For example, extract the following from the scheduling decision template: water inflow forecasting and water supply assessment have a parallel collaboration relationship, both pointing to the preceding and following process relationship of scheduling scheme generation, and scheduling scheme generation pointing to the preceding and following process relationship of scheme verification.

[0053] The business process types contained in the business process template are used as node type filtering conditions during the traversal process. This node type filtering condition is used to retain only business process nodes that match the business processing type during the traversal. For example, the node filtering condition for the water inflow forecast type is to retain only three types of task nodes: rainfall runoff calculation, river confluence calculation, and cross-sectional water inflow forecast. Other task nodes, such as water supply assessment and scheduling decisions, are not included in the traversal scope.

[0054] The maximum execution link depth of business steps in the statistical business process template is used as the maximum length for tracing forward or backward in the business step chain. For example, if the maximum execution link depth of the water inflow forecast template is 3 layers, then the maximum tracing length of the business step chain is 3; if the maximum execution link depth of the "scheduling decision" template is 4 layers, then the maximum tracing length is 4.

[0055] S223, Generation scope dimension and extension strategy constraints.

[0056] Based on physical topology constraints and business transmission constraints, the constraints are further structured into five-dimensional constraints of scope templates, and expansion strategy rules that can drive graph traversal are generated, including: (1) Spatial scope constraints: the set of relationship types and spatial ranges that can be expanded as determined in S221 are used to determine the expansion direction and coverage of the physical entity network during the pruning process. (2) Object scope constraints: the spatial hierarchy ranges that are determined in S221 are used to control whether to drill down to the engineering level, equipment level or finer-grained objects during the pruning process. (3) Business scope constraints: the process relationship types and node type filtering conditions that can be expanded as determined in S222 are used to constrain the expansion range in the business relationship network. (4) Temporal scope constraints: the implicit time attributes are extracted from the business processing types. For example, water inflow forecast corresponds to future forecast, water supply assessment corresponds to the current state or short-term forecast, and scheduling decision corresponds to the future execution period. (5) Decision scope constraints: the implicit decision hierarchy attributes are extracted from the business processing types. For example, water inflow forecasting corresponds to the analysis level, water supply assessment corresponds to the assessment level, and scheduling decision-making corresponds to the scheduling level.

[0057] The above five-dimensional constraints are transformed into execution constraints for specific graph traversal. Each extension strategy constraint defines the complete rules for traversing along graph relations starting from the target node, including the starting node type, allowed relation types and relation directions, hop count range, target node type constraints, and stopping conditions.

[0058] For example, for the water inflow prediction type problem, the extended strategy constraint is defined as follows: starting from section D, trace back along the upstream water inflow relationship. The allowed relationship types are upstream water inflow relationship and branch relationship. The maximum number of hops is 3. The target node type is limited to river section and reservoir. The stopping condition is reaching the watershed boundary of the catchment area above section D determined by S221. At the same time, the business task traces along the water inflow prediction template chain determined by S222 for no more than 3 layers.

[0059] For example, for scheduling decision-type problems, the extended strategy constraint is defined as follows: starting from gate C, it extends bidirectionally upstream and downstream. Allowed relationship types include upstream water inflow relationship, downstream water release relationship, control action relationship, and water supply service relationship. The maximum number of hops is 5. The target node type is limited to reservoir, river channel, cross section, and irrigation area. The stopping condition is reaching the spatial boundary of the upstream river channel B to the downstream irrigation area E determined by S221. At the same time, the business task traceability along the scheduling decision template chain determined by S222 does not exceed 4 layers.

[0060] S300. In the global business graph of the water network, starting from the business links directly associated with the target water network entity, the path is extended along the business process extension direction. Business links that exceed the hop count range during the path extension process are filtered out, as well as water network entities and intelligent agents that have anchoring or execution relationships with the filtered business links. Candidate business paths formed by the remaining business links are obtained. The candidate business paths are closed and verified. The graph formed by the candidate business paths that pass the closure verification and their associated water network entities and intelligent agents is used as a local graph.

[0061] Under the combined effect of generated physical topology constraints, service delivery constraints, and extension strategy rules, a constraint-based pruning is performed on the global water network map. Closure verification ensures the integrity of the pruning result, ultimately generating a local map tailored to the current problem. Specifically, this includes the following sub-steps:

[0062] S310, Constraint clipping.

[0063] The pruning process starts with the target water network entity as the traversal starting point and uses a breadth-first search algorithm to expand the traversal along the edges of the global water network graph. The traversal strictly adheres to the expansion strategy constraints generated by S223, including the starting node type, allowed relationship types and directions, hop count range, target node type constraints, and stopping conditions.

[0064] At the physical entity level, traversal starts from the target water network entity and extends only along the allowed upstream and downstream connections and control relationships defined in S221. When a traversed physical entity node exceeds the spatial area range, spatial hierarchy range, or hop count limit defined in S221, the node and its associated edges are removed, terminating further extension on that path. At the business task level, traversal extends synchronously along business process relationships, retaining only the allowed process relationship types defined in S222, including pre- and post-dependencies, input / output transmission, and parallel collaboration relationships. When a traversed business task node type does not meet the node type filtering conditions defined in S222, the node is removed; when the business task chain tracing depth exceeds the maximum execution link depth defined in S222, extension in that direction terminates. When a physical entity node is removed due to exceeding physical topology constraints, all business task nodes anchored to that physical entity, as well as the agent nodes executing those tasks, are also removed. When a business task node is removed due to exceeding business transmission constraints, the agent nodes executing that task, as well as the algorithm model nodes authorized to that agent, are also removed.

[0065] For example, consider a user's objective of assessing whether the current available water supply in irrigation district E meets the target of a water supply guarantee rate of no less than 95%. The target water network entity is irrigation district E. The constraint parameters generated by S200 include: allowed extended relationship types are water supply service relationship, control action relationship, and downstream water release relationship; the maximum number of hops is 3; the business node type filtering condition limits the retention to only three types of task nodes: water supply assessment, reservoir inflow calculation, and inflow prediction; and the maximum business traceability length is 3 layers.

[0066] During the pruning process, starting from irrigation district E, the process traces backward along the water supply service relationship to section D, then backward along the control relationship to gate C, and finally backward along the downstream water release relationship to reservoir A, with a maximum of 3 hops. During the expansion process, branching and inclusion relationships, as well as other relationship types, are not allowed; therefore, nodes connected to these relationships are not included. At the business level, the node type filtering condition limits the traversal to retaining only three types of task nodes: water supply assessment, reservoir inflow calculation, and inflow prediction. Other task nodes, such as scheduling decisions, are filtered during the traversal and are not included in the local graph.

[0067] S320, Closure Verification and Local Graph Generation.

[0068] After trimming, closure verification is performed. Closure verification includes three rules: (1) Path integrity verification. In the trimmed local graph, there must be a complete connected path from the target water network entity to all affected end objects to ensure that there are no broken links. If a broken link is found, a layer of relation edges is extended outward along the direction of the missing path to fill in the intermediate nodes until the path is complete. (2) Business task integrity verification. The trimmed local graph must contain all standard business link nodes and their process relationships defined in the business process template in S222 to ensure that there are no missing tasks. If a missing task is found, the missing task node and its predecessor or successor task are included outward along the business dependency edge until the business chain is complete. (3) Cross-layer association integrity verification. There should be no business task nodes that are not anchored to any physical entity in the trimming result, nor should there be any agent nodes that are not associated with any business task nodes. If the above isolated nodes are found, the completion mechanism is triggered: unanchored business task nodes are expanded outward one layer along the anchoring relationship to include the physical entity nodes to which they should be anchored; intelligent agent nodes that are not associated with business tasks are expanded outward one layer along the execution relationship to include the business task nodes to which they should be executed.

[0069] For example, taking the water supply guarantee rate assessment of irrigation district E as an example, a closure check is performed after trimming. First, the path integrity is checked to confirm that the complete path from irrigation district E to reservoir A is: Irrigation district E—water supply service—section D—control function—gate C—downstream water release—reservoir A, with a complete and unbroken path. Second, the business task integrity is checked to confirm that the three task nodes—inflow prediction, reservoir inflow calculation, and water supply assessment—all exist, and the dependency edge—inflow prediction → reservoir inflow calculation → water supply assessment—is complete. Finally, the cross-layer association integrity is checked. If the water supply assessment task node is found not to be anchored to irrigation district E, then one layer is extended outward along the anchoring relationship to complete the anchoring edge; if the inflow prediction task node is missing, then one layer is extended outward along the dependency chain—inflow prediction → reservoir inflow calculation → water supply assessment—to include the inflow prediction node. After the check passes, a local graph for the water supply guarantee rate assessment problem of irrigation district E is generated.

[0070] The physical entity nodes included in this local graph are: Reservoir A, River B, Gate C, Cross-section D, and Irrigation District E; the operational task nodes are: Inflow Prediction, Reservoir Inflow Calculation, and Water Supply Assessment; the agent nodes are: Forecasting Agent P and Irrigation Agent I; and the algorithm model nodes are: Watershed Runoff Generation Model M1 and River Confluence Model M2. The cross-layer anchoring relationships between these nodes are as follows: the Inflow Prediction task is anchored to River B and Cross-section D, executed by Forecasting Agent P, which is authorized to call Watershed Runoff Generation Model M1 and River Confluence Model M2; the Reservoir Inflow Calculation task is anchored to Reservoir A, executed by Forecasting Agent P; and the Water Supply Assessment task is anchored to Irrigation District E, executed by Irrigation Agent I. This local graph maintains complete cross-layer relationships and can be directly used for subsequent agent scheduling and tool invocation steps.

[0071] S400: Based on the intelligent agent in the local graph, execute the business process in the local graph to obtain the analysis results of the water network business analysis request.

[0072] Based on the generated local graph, a temporary collaborative architecture of a moderator-expert multi-agent system is constructed to address the current problem. The execution boundaries of each agent are constrained through a structured skill generation and tool registration mechanism. During execution, anomaly detection triggers expert intervention to ensure stability. Execution experience is then structured and accumulated for continuous optimization. Specifically, the following steps are included:

[0073] S410, A set of skills and tools for building expert agents based on local graphs.

[0074] Based on the agent collaboration graph in the local graph, the set of expert agents corresponding to the current problem is read. Combining the basic description, applicable business type, input and output requirements, usage scenario, and tool information and computational link process relationship between tools of each agent in the algorithm model graph, corresponding skills are dynamically generated for each expert agent.

[0075] This skill is not a pre-defined prompt, but rather a structured capability description dynamically organized from the physical entity nodes, business process nodes, and related algorithmic model capabilities extracted from the local graph based on the current problem. The skill content includes at least the following elements: the agent's scope of responsibility in the current problem, the business processes it can handle, the required input information, the expected output results, the types of algorithmic models that can be invoked, and the constraints for tool invocation. Through this approach, each expert agent acquires a unique skill matching the current problem before execution, ensuring that its subsequent behavior is constrained by the object scope, business scope, and capability boundaries within the local graph.

[0076] After generating skills, based on the binding relationships of the agent in the algorithm model graph, a set of algorithm models authorized to the agent is selected, and the selected tools are registered as the agent's tool set. After registration, each expert agent can call the registered algorithm models within its own permission scope through function calls; its calling boundaries are constrained by the scope of responsibility, input and output requirements, and usage scenarios defined in the corresponding skill of the agent.

[0077] For example, in the local graph corresponding to the water supply guarantee rate assessment problem of irrigation district E, the forecasting agent P is authorized to call the watershed runoff generation model M1 and the river confluence model M2. The skills generated by the forecasting agent P include: its responsibility is to perform inflow prediction and reservoir inflow calculation; the input is rainfall data and the cross-sectional water level-discharge relationship of river B; the output is the inflow process of reservoir A; the available tools are the watershed runoff generation model M1 and the river confluence model M2; the calling constraint is that M1 must be run sequentially before M2. The skills generated by the irrigation agent I include: its responsibility is to perform water supply assessment; the input is the available water supply of reservoir A and the crop water requirement data of irrigation district E; the output is the assessment conclusion of whether the water supply guarantee rate of the irrigation district meets 95%; it has no callable algorithm model tools and works only through built-in empirical formulas.

[0078] S420. Construct a set of skills and tools for the host agent based on local graphs.

[0079] After completing the construction of the expert agent swarm, the skills of the host agent are dynamically constructed based on user questions, physical entity nodes, business process nodes, available agent nodes, and collaborative relationships between agents in the local graph. The host agent does not directly undertake specific business calculations, but acts as the central hub for multi-agent interaction and collaboration, responsible for perceiving the overall business context, grasping task progress, and forming global coordination capabilities.

[0080] The host agent's skills are dynamically generated from structured information about the current problem and the local graph. The skills include at least: the current problem objective, the physical scope defined by the local graph, the business process, the set of candidate expert agents, the collaborative relationships between expert agents, and the applicable scope of each expert agent. Through these skills, the host agent can understand the scope of objects involved in the current problem, the required business processes, the available expert agents, and the collaborative methods between them.

[0081] Based on this, the system selects expert agents from the local graph to participate in the problem-solving process, forming a temporary set of agents specific to the problem. This temporary set of agents is then registered as a toolkit for the host agent. After registration, the host agent can organize and schedule the expert agents as callable objects through function calls, thereby forming a temporary collaborative architecture of host-expert agent group for the current problem.

[0082] For example, in the water supply guarantee rate assessment of irrigation district E, the skill content generated for the host agent includes: the problem objective is to assess whether the water supply of irrigation district E meets the guarantee rate of no less than 95%; the physical scope is reservoir A, river B, gate C, cross section D, and irrigation district E; the business chain is inflow prediction → reservoir inflow calculation → water supply assessment; the candidate expert agents are forecasting agent P and irrigation agent I; the collaborative relationship is that after P completes the reservoir inflow calculation, its result is used as the input for I to perform the water supply assessment. The tool set of the host agent is registered as forecasting agent P and irrigation agent I, and the host can schedule P and I to complete their respective tasks sequentially through function calls.

[0083] S430. Construct an execution anomaly identification and expert intervention trigger mechanism.

[0084] In the process of multi-agent collaborative execution, a water network anomaly identification agent is constructed and bound to the host agent as an additional tool. This agent is used to monitor and determine anomalies in the output results, key intermediate variables and execution process of each expert agent, thereby providing a trigger basis for expert intervention.

[0085] The water network anomaly identification agent uses the local graph corresponding to the current problem as a constraint and the state data generated during the collaborative execution of multiple agents as input to perform structured analysis of the execution process. Specifically, it extracts execution state information from the unified task context maintained by the host agent and constructs a standardized state representation. This state representation includes the target object set, related physical entity nodes and their attributes, business indicator variables, the output results of each expert agent, tool call results, and execution round information. It is organized through a unified object-indicator-time mapping method, enabling the outputs of different agents to be compared and verified in the same semantic space.

[0086] The water network anomaly detection agent determines anomalies using the following four methods: Constraint violation anomaly determination. Based on the business constraint relationship and physical constraint relationship in the local map, the constraint verification of key business indicators is carried out. When there are situations where the water supply exceeds the engineering capacity, the cross-sectional flow is lower than the ecological lower limit, the water volume of upstream and downstream is mismatched, or the existing scheduling rules are violated, it is determined as a constraint violation anomaly.

[0087] Multi-agent result conflict anomaly detection. The output results of different expert agents around the same object or the same business indicator are aligned and compared, and the degree of difference in the results is calculated. When the numerical differences given by multiple agents for the same variable exceed a preset threshold or there is a directional conflict in the choice of solution, it is judged as a multi-agent result conflict anomaly.

[0088] Convergence anomaly detection. Key variables during execution are sequentially tracked, and the results of multiple execution rounds are analyzed for trends. If variables exhibit repeated fluctuations, periodic switching, or no clear convergence trend over multiple consecutive rounds, it is considered a convergence anomaly.

[0089] Anomaly detection based on experience bias. The current execution result is matched with historical valid results in the experience base. When the result deviates significantly from key indicators or solution structure, it is determined to be an anomaly based on experience bias.

[0090] After completing the above anomaly detection, the host agent decides whether to pause the current automated execution process and trigger expert intervention based on the anomaly type and severity. When the anomaly severity is low, automated execution continues or partial corrections are made; when the anomaly severity is high or involves critical hydraulic engineering objects or important business scenarios, the expert intervention process is initiated.

[0091] For example, during the assessment of the water supply guarantee rate in irrigation district E, the forecasting agent P calculated the inflow rate of reservoir A to be 10 cubic meters per second. Based on this, the irrigation agent I assessed the water supply guarantee rate of irrigation district E as 92%, lower than the target of 95%. After capturing this result, the water network anomaly identification agent found that the water supply guarantee rate was lower than the constraint threshold of the business objective T1, and determined it to be a constraint violation anomaly. At the same time, it matched the current result with the experience base and found that in similar scenarios, the guarantee rate could usually be improved by adjusting the opening of gate C to increase the outflow, but the adjustment range was limited by the ecological base flow constraint. The host agent comprehensively determined that this anomaly involved the key hydraulic engineering object gate C and ecological constraints, triggering the expert intervention process.

[0092] S440. Construct a structured problem expression and interactive correction mechanism oriented towards experts.

[0093] After expert intervention is triggered, the host agent reconstructs the current execution state, generates a structured question expression for the expert, and interacts with the expert based on the structured expression, transforming the expert feedback into actionable correction information.

[0094] Specifically, the host agent extracts core information related to the current anomaly from the unified task context, compresses and filters the original execution process, and forms a minimal decision context for experts. This structured problem expression revolves around the anomaly location, including the target object, the key physical entities involved, relevant business indicators, current calculation results, and anomaly judgment results, and marks the source of conflict or non-convergence location, enabling experts to quickly locate the core of the problem.

[0095] During expert interaction, key decision-making elements are revised, including adjusting target priorities, modifying key parameters, screening or ranking candidate solutions, and supplementing or revising constraints. Expert feedback is returned in a structured format. Upon receiving expert feedback, the moderator agent parses it and maps it into executable correction operations within the system. Based on the correction results, the input parameters, constraints, or candidate solutions of the relevant expert agents are updated, and subsequent collaborative computation continues, enabling the system to escape abnormal paths and obtain stable results.

[0096] For example, the structured problem generated by the host agent for experts is expressed as follows: the target object is the water supply guarantee rate of irrigation district E; the key physical entities are gate C and cross-section D; the business indicator is the difference between the current water supply guarantee rate of 92% and the target value of 95%; the anomaly is a constraint violation anomaly, where the water supply guarantee rate is lower than the business target T1; the current scheduling plan involves adjusting the opening of gate C, but is constrained by the ecological base flow of cross-section D not being less than 5 cubic meters per second. Expert feedback suggests adjusting the opening of gate C to increase the discharge flow, but temporarily relaxing the ecological base flow constraint to 4.5 cubic meters per second, and increasing nighttime compensation water release to compensate for the ecological impact. The host agent maps the expert feedback into internal executable operations: modifying the ecological base flow constraint threshold parameter, regenerating the gate C opening plan, and allocating compensation periods. After the update, the forecast agent P and the irrigation agent I are rescheduled to perform calculations, and the evaluation result achieves a water supply guarantee rate of 95.2%, meeting the business target.

[0097] S450. Establish a structured mechanism for the accumulation and reuse of expert experience.

[0098] Figure 3 This specification provides a schematic diagram of a multi-agent collaborative workflow, such as... Figure 3As shown, the multi-agent collaborative architecture uses the host agent as the coordination center, connecting a group of expert agents through scheduling and feedback edges. The expert agent group consists of multiple expert agents, each invoking the underlying algorithm model toolkit via invocation edges. The diagram exemplifies expert agent A invoking tool A, expert agent B invoking both tool A and tool B, and expert agent C invoking tool A, demonstrating that different expert agents have differentiated tool invocation permissions based on authorization relationships, and that tools can be shared among multiple expert agents. Building upon this collaborative architecture, this step further constructs a structured accumulation and reuse mechanism for expert experience.

[0099] The experience-based governance agent performs structured extraction of key data during the execution process. Inputs include the current problem type, target object, local graph snapshot, expert agent input / output records, tool call trajectories, anomaly identification results, and expert intervention records. Execution processes that do not trigger anomalies but meet business constraints and have stable results are also included in the experience-based governance scope. During the structured extraction process, the experience-based governance agent first maps key elements of the execution process to object nodes, business nodes, and constraint relationships in the local graph, determining the object scope and business scenario corresponding to the experience. Then, it extracts key decision-making information, including decision paths, triggering conditions, and execution results.

[0100] During the experience accumulation process, each experience unit includes the problem type, target object characteristics, key structures of the local graph, triggering conditions, decision-making actions, and final results, enabling the experience to establish a clear connection with specific business scenarios. Based on the applicability and stability of the experience, it is classified and managed. Experiences strongly related to specific regions, specific hydraulic engineering objects, or specific management models are classified as individual experiences and bound to the corresponding region, object type, and business scenario; experiences that repeatedly occur in multiple scenarios and have stable execution effects are classified as general experiences and further abstracted into rule-based expressions for cross-scenario reuse.

[0101] In subsequent multi-agent collaborative execution, when the execution result is judged as abnormal by the water network anomaly identification agent, the host agent searches the experience base based on the current problem characteristics and transforms the matching experience into executable constraints or decision preferences to participate in subsequent calculations, thereby correcting and optimizing the abnormal results. During the continuous accumulation of experience, when a certain type of experience demonstrates stable effectiveness in multiple applications, it is elevated from the experience layer to the rule layer and written into the business constraint relationships or scope rules in the four-network structure. This allows it to directly participate in reasoning as a pre-constraint in subsequent calculations, reducing reliance on expert intervention.

[0102] For example, the experience-based governance agent extracts the experience from this execution: the problem type is irrigation district water supply guarantee rate assessment; the target object is irrigation district E; the triggering condition is that the water supply guarantee rate is lower than the target of 95%, and is also subject to ecological baseflow constraints; the decision action is for experts to relax the ecological baseflow constraints and set up nighttime compensatory water release; the final result is that the water supply guarantee rate is increased to 95.2%. This experience unit is stored in the experience database, categorized as individual experience, and bound to irrigation district E and its watershed. Subsequently, when processing the water supply assessment problem of irrigation district E, the host agent prioritizes retrieving this experience, transforms the strategy of relaxing the ecological baseflow constraints and adding nighttime compensatory water release into a candidate solution for subsequent calculations, thereby reducing the number of expert interventions. When this strategy has proven stable and effective in multiple irrigation districts or multiple assessments, the experience-based governance agent elevates it from the experience layer to the rule layer, writes it into the business constraints of irrigation district water supply assessment, and directly participates in inference as one of the subsequent default scheduling rules.

[0103] Figure 4 This is the second flowchart illustrating a business analysis method for the collaboration of water network maps and intelligent agents, as provided in this specification. Figure 4 As shown, this process begins with user input, proceeds through problem analysis and object identification, local graph generation, multi-agent collaborative execution, anomaly detection and branch processing, and finally outputs results or triggers expert intervention for correction and experience accumulation, forming a complete closed-loop processing chain. Based on the above process architecture, this method achieves beneficial effects in the following three aspects:

[0104] First, by constructing a unified representation system for the global water network graph, the system achieves unified modeling and relational representation of physical entities, business processes, intelligent agents, and algorithm models within the water network. The unified global structure provides a unified underlying semantic foundation for the physical network, business network, intelligent agent network, and tool network. Based on this, after the user inputs a question, the system uses a large language model to analyze the question and identify the object, mapping the natural language question to specific nodes and relationships in the global graph, thus determining the analysis target and scope. This mechanism enables the system to automatically identify the analysis object and business links, establishing a structured data foundation for subsequent local graph generation and multi-agent collaborative execution, thereby improving the organizational capacity and analysis efficiency for complex water network problems.

[0105] Second, through a local graph generation mechanism and a multi-agent collaborative execution mechanism, the system achieves capability boundary constraints on the analysis scope and on-demand organization of multiple agents. After problem analysis and object identification are completed, the system enters the local graph generation stage. Based on the scope template, extension strategy rules, and closure verification, a local graph oriented towards the current problem is cropped from the global water network graph, completing the convergence of physical scope, business links, and execution capabilities. Subsequently, the system enters the multi-agent collaborative execution stage, where the host agent coordinates the overall process, and a group of expert agents execute specific business steps according to their roles. During this process, the skills and tool call relationships of each agent are dynamically constructed based on the local graph, ensuring that each agent has a clear scope of responsibility, business processing capabilities, and tool usage permissions before execution, thereby improving the orderliness and overall controllability of the multi-agent collaborative process.

[0106] Third, a closed-loop design of anomaly identification, expert intervention, and experience accumulation mechanisms enables dynamic verification and continuous optimization of the multi-agent collaborative execution process. An anomaly detection node is set after the multi-agent collaborative execution phase: when the execution result is normal, the result is output directly; when the execution result is determined to be abnormal, an expert intervention correction branch is initiated, where experts correct the deviation. After correction, the system re-enters the multi-agent collaborative execution phase for iterative calculation until the result stabilizes. Simultaneously, each expert intervention correction triggers experience accumulation and rule updates, transforming expert knowledge into reusable experience units or rule constraints, and writing them back to the unified global structure of the four networks. This allows the system to automatically invoke this experience for correction in subsequent similar problems without repeatedly triggering expert intervention. This closed-loop mechanism enables the system to maintain stable operation in complex water network scenarios, continuously reducing reliance on manual intervention and improving the overall stability, reliability, and adaptability of the system.

[0107] The following describes the business analysis device for water network map and intelligent agent collaboration provided by the present invention. The business analysis device for water network map and intelligent agent collaboration described below and the business analysis method for water network map and intelligent agent collaboration described above can be referred to in correspondence.

[0108] Figure 5 This is a schematic diagram of the structure of a business analysis device for water network mapping and intelligent agent collaboration provided by the present invention. For example, please refer to [link to schematic diagram]. Figure 5 As shown, the business analysis device that coordinates the water network map with intelligent agents may include: The global water network graph construction module is used to construct a global water network business graph with water network entities, business processes, and intelligent agents as nodes, and the anchoring relationship between water network entities and business processes, the process relationship between business processes, and the execution relationship between intelligent agents and business processes as edges.

[0109] The business problem analysis module is used to obtain water network business analysis requests, identify the target water network entity and target business type corresponding to the water network business analysis request, and extract the business process extension direction and hop range that match the target business type from the preset strategy library.

[0110] The local graph construction module is used to expand the path in the global water network business graph, starting from the business links directly associated with the target water network entity and extending along the business process. Business links that exceed the hop count range during the path expansion process are filtered out, as well as water network entities and agents that have anchoring or execution relationships with the filtered business links. Candidate business paths formed by the remaining business links are obtained, and the candidate business paths are closed-loop verified. The graph formed by the candidate business paths that pass the closure verification and their associated water network entities and agents is used as the local graph.

[0111] The analysis module is used to execute business processes in the local graph based on the agents in the local graph, and obtain the analysis results of the water network business analysis request.

[0112] Specific limitations regarding the business analysis device for water network mapping and intelligent agent collaboration can be found in the limitations of water network business analysis above, and will not be repeated here. Each module in the aforementioned business analysis device for water network mapping and intelligent agent collaboration can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0113] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The provided water network map and intelligent agent collaborative business analysis method.

[0114] This instruction manual also provides Figure 6 The schematic diagram of the computer device shown is as follows: Figure 6 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 The provided water network map and intelligent agent collaborative business analysis method.

[0115] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0116] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A business analysis method for the collaboration of water network maps and intelligent agents, characterized in that, include: Using water network entities, business processes, and intelligent agents as nodes, and the anchoring relationship between water network entities and business processes, the process relationship between business processes, and the execution relationship between intelligent agents and business processes as edges, a global business graph of the water network is constructed. Obtain water network business analysis requests, identify the target water network entity and target business type corresponding to the water network business analysis requests, and extract the business process expansion direction and hop range that match the target business type from the preset strategy library; In the global business graph of the water network, starting from the business links directly associated with the target water network entity, the path is extended along the business process extension direction. Business links that exceed the hop count range during the path extension process are filtered out, as well as water network entities and intelligent agents that have anchoring or execution relationships with the filtered business links. Candidate business paths formed by the remaining business links are obtained. The candidate business paths are closed and verified. The graph formed by the candidate business paths that pass the closure verification and their associated water network entities and intelligent agents is used as a local graph. Based on the intelligent agents in the local graph, the business processes in the local graph are executed to obtain the analysis results of the water network business analysis request.

2. The business analysis method for water network mapping and intelligent agent collaboration as described in claim 1, characterized in that, The step of extracting the business process extension direction and hop range that match the target business type from the preset strategy library includes: Match the target business type with multiple business process templates in the preset strategy library, and obtain the target business process template corresponding to the target business type based on the matching results; the business process template at least defines the expansion direction and hop range of the business process corresponding to the business type. The extension direction defined in the target business process template is used as the business process extension direction. The extension direction includes tracing upstream along the process relationship, extending downstream, or bidirectional extension.

3. The business analysis method for water network mapping and intelligent agent collaboration as described in claim 1, characterized in that, The path expansion, starting from the business links directly associated with the target water network entity and extending along the business process direction, includes: Starting with the target water network entity, the first round of expansion is carried out along the anchoring relationship to obtain all business link nodes directly related to the target water network entity; Starting with all the business process nodes obtained, a second round of expansion is carried out synchronously along the process relationship and in the direction of business process expansion. During the expansion process, each business process node extends downstream or upstream along its respective process relationship.

4. The business analysis method for water network mapping and intelligent agent collaboration as described in claim 1, characterized in that, Business segments that exceed the hop count limit during path expansion will be filtered, as well as water network entities and intelligent agents that have anchoring or execution relationships with the filtered business segments, including: During the path expansion process, record the current hop count of each business node from the starting point; When the current hop count of any business process node exceeds the hop count range, the business process node is marked as an out-of-range node and removed. Using the removed business process node as the trigger point, search in reverse along the anchoring relationship for water network entity nodes that have an anchoring relationship with the removed business process node and remove them; Using the removed business process node as the trigger point, search backwards along the execution relationship to find the intelligent agent node that has an execution relationship with the removed business process node and remove it.

5. The business analysis method for water network mapping and intelligent agent collaboration as described in claim 1, characterized in that, The closure verification of candidate service paths includes: Traverse each business link node in the candidate business path and check whether the business link node is connected to the corresponding water network entity node through anchoring relationship and connected to the corresponding intelligent agent node through execution relationship in the candidate business path. If all business process nodes in the candidate business path meet the above conditions, then the candidate business path is determined to have passed the closure check. If any business link node in the candidate business path does not meet the above conditions, the candidate business path is determined to have failed the closure check.

6. The business analysis method for water network mapping and intelligent agent collaboration as described in claim 1, characterized in that, The identification of the target water network entity and target service type corresponding to the water network service analysis request includes: Entity identification is performed on water network business analysis requests to obtain the physical entity names in the water network business analysis requests. Based on the physical entity names, a search and matching is performed in the global water network business map, and the matched water network entity nodes are used as target water network entities. The intent analysis of water network business analysis requests is performed, and the results of the intent analysis are matched with a preset set of business types to determine the target business type.

7. The business analysis method for water network mapping and intelligent agent collaboration as described in claim 1, characterized in that, The node also includes an algorithm model node, and the edge also includes the authorization relationship between the algorithm model and the agent; The construction of the global water network business map includes: A physical entity graph is constructed using water network entities as nodes and the upstream and downstream connections and control relationships between water network entities as edges. Based on the physical entity map, the business process is linked to the corresponding water network entity through the anchoring relationship between the business process and the water network entity, and a business relationship map is constructed by combining the process relationship between the business processes. Based on the business relationship graph, the intelligent agents are associated with the corresponding business processes through the execution relationship between the intelligent agents and the business processes, and the intelligent agent relationship graph is constructed by combining the collaborative relationship between the intelligent agents. Based on the agent relationship graph, the scope of the agent's algorithm model call is limited by the authorization relationship between the algorithm model and the agent, and a tool graph is constructed by combining the call relationship between the algorithm models. By integrating the physical entity graph, business relationship graph, agent relationship graph, and tool graph, a global business graph of the water network is obtained.

8. The business analysis method for water network mapping and intelligent agent collaboration as described in claim 7, characterized in that, The process of an agent based on a local graph executing business processes within that local graph to obtain analysis results for a water network business analysis request includes: Based on the execution relationship between agent nodes and business process nodes in the local graph, the expert agents participating in each business process in the business path are identified, and the algorithm models in the local graph that have authorization relationships with agent nodes are registered as the computing tools of the expert agents. Construct a host agent, which determines the order of business processing based on the process relationships between business links in the local graph, and schedules the corresponding expert agents according to the order of business processing. The expert agent responds to the scheduling instructions of the host agent, calls the registered computing tools to execute the corresponding business steps, and obtains the sub-results; The host agent collects the sub-results output by each expert agent and merges them to generate the analysis results for the water network business analysis request.

9. A business analysis device for the collaboration of water network maps and intelligent agents, characterized in that, include: The global water network graph construction module is used to construct a global water network business graph with water network entities, business processes, and intelligent agents as nodes, and the anchoring relationship between water network entities and business processes, the process relationship between business processes, and the execution relationship between intelligent agents and business processes as edges. The business problem analysis module is used to obtain water network business analysis requests, identify the target water network entity and target business type corresponding to the water network business analysis request, and extract the business process extension direction and hop range that match the target business type from the preset strategy library; The local graph construction module is used to expand the path in the global water network business graph, starting from the business links directly associated with the target water network entity and extending along the business process. It filters out business links that exceed the hop count range during the path expansion process, and filters out water network entities and intelligent agents that have anchoring or execution relationships with the filtered business links. It obtains candidate business paths formed by the remaining business links, performs closure verification on the candidate business paths, and uses the graph formed by the candidate business paths that pass the closure verification and their associated water network entities and intelligent agents as the local graph. The analysis module is used to execute business processes in the local graph based on the agents in the local graph, and obtain the analysis results of the water network business analysis request.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the business analysis method for water network mapping and intelligent agent collaboration as described in any one of claims 1 to 8.

Images