Reservoir dispatching rule self-evolution management method and system based on mapping knowledge domain

Abstract

The invention relates to the technical field of water conservancy and hydropower engineering, in particular to a reservoir dispatching rule self-evolution management method and system based on a knowledge graph. According to the scheme, a double-layer self-adaptive closed-loop architecture is formed by constructing a dynamic knowledge graph fusing three subsystems of flood discharge and sediment transportation, ecological environment and social economy, establishing a scene-triggered rule chain nested execution mechanism and deploying a sudden change event instant response layer and an obstacle degree-driven long-term evolution layer. The invention provides a knowledge graph-based reservoir scheduling rule self-evolution management method and system, and solves or at least alleviates the problem that a conventional reservoir scheduling rule is hard to adapt to a complex dynamic environment due to static solidification.

Description

Technical Field

[0001] This invention relates to the field of water conservancy and hydropower engineering technology, and in particular to a self-evolving management method and system for reservoir scheduling rules based on knowledge graphs. Background Technology

[0002] With the deep integration of artificial intelligence technology and water conservancy scheduling systems, intelligent scheduling methods for reservoir groups based on knowledge graphs have become a key approach to improving the comprehensive benefits of river basins and achieving coordinated development of water and sediment resources and ecological economy. For high-sediment-laden rivers like the Yellow River, with its complex hydrological conditions, cascade reservoir groups must simultaneously address three core objectives: smooth flood and sediment transport, ecological health, and socio-economic growth. This poses unprecedented challenges to the scheduling system's perception capabilities, decision-making accuracy, and rule evolution mechanisms.

[0003] Traditional scheduling methods mostly rely on empirical rules or static models. When faced with dynamic situations such as extreme water and sediment events, sudden changes in ecological thresholds, or sudden increases in grid load, it is difficult to adjust strategies in a timely manner, resulting in a significant deterioration in scheduling effectiveness. Therefore, in recent years, the introduction of knowledge graphs to integrate multi-source heterogeneous data, explicitly express the coupling relationships between multi-dimensional objectives, and support intelligent decision generation has become a research hotspot.

[0004] In the existing technology, patent CN118709545B discloses a method and system for regulating the discharge of large reservoir groups. It assists in joint scheduling decisions by constructing a knowledge graph that integrates physical mechanisms and mining hidden relationships between multi-dimensional data such as hydrology, sediment, and engineering operations. This scheme makes the scheduling scheme more scientific and interpretable to a certain extent, especially under normal circumstances, effectively supporting the screening and comparison of non-dominated solution sets. However, its knowledge graph is essentially a static modeling tool, only usable for offline relational reasoning and scheme recommendation, and it does not establish a closed-loop mechanism of "execution-evaluation-feedback-update". Specifically, when problems arise in the actual scheduling effect, such as deviations in sediment discharge or excessive ecological water shortage rates, the system cannot automatically calculate the cause of the deviation, identify the main obstacle indicators, or use the evaluation results to dynamically correct rule parameters. Therefore, in the event of sudden high sediment content events, such as a sudden increase in sediment due to heavy rain, its response time often exceeds 72 hours, much later than the golden decision window (usually less than 2 hours), which greatly weakens the scheduling system's ability to ensure flood control safety and river health.

[0005] Furthermore, patent CN117952370B primarily focuses on optimizing flood diversion strategies in flood storage and detention areas. It generates a knowledge base capable of rapidly comparing multiple solutions by constructing a correlation graph between flood diversion parameters and water level reduction values. While this method improves decision-making efficiency by introducing a graph structure, its graph construction relies entirely on offline training using historical data, lacking a mechanism for dynamic integration with real-time monitoring data. Particularly when the basin state undergoes sudden changes, such as a sediment concentration increase exceeding 25% in a short period, the system still performs global optimization according to a pre-set fixed objective function. It cannot dynamically adjust the decision weights of subsystems such as flood control, ecology, and economy based on the current situation, nor can it compress the non-dominated solution set to reflect actual conditions. Actual test data shows that under such sudden changes, its decision results differ significantly from the real-time optimal solution, reducing flood control efficiency by 22%. More importantly, this method fails to establish an automatic rule evolution mechanism; all strategy optimization relies on manual post-hoc analysis of obstacle indicators and manual modification of the rule base, essentially remaining in a passive response phase without true adaptive evolutionary capability.

[0006] Ultimately, the shortcomings of these existing technologies stem from a misunderstanding of the nature of rules, which remains focused on encoding them as static knowledge rather than recognizing them as dynamic entities capable of sensing the environment, assessing effects, and autonomously evolving. In complex watershed systems, the three subsystems of flood control and sediment transport, ecological environment, and socio-economic development interact in intricate and constantly changing relationships. For example, increasing outflow to improve sediment discharge efficiency might exacerbate downstream water shortages; conversely, increasing water storage at the end of the flood season to boost power generation might reduce flood control capacity during the flood season. This conflict between multiple objectives requires scheduling rules not only to make the best choice at any given moment but also to continuously learn and optimize over long-term operation. However, current knowledge graph applications generally separate decision generation and rule evolution. The former focuses solely on providing immediate solutions, while the latter relies on human intervention, preventing the system from self-improving. Furthermore, the need for rapid response to emergencies and gradual repair of long-term performance degradation renders single-time-scale control architectures insufficient. Focusing solely on rapid response risks local instability; focusing solely on long-term optimization fails to handle sudden crises. Summary of the Invention

[0007] The purpose of this invention is to provide a knowledge graph-based method and system for the self-evolution of reservoir scheduling rules, which can solve or at least alleviate the problem that traditional reservoir scheduling rules are static and rigid and difficult to adapt to complex dynamic environments.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a self-evolving management method for reservoir scheduling rules based on knowledge graphs, comprising:

[0009] S1. Construct a dynamic knowledge graph that integrates the three major subsystems of flood control and sediment transport, ecological environment and socio-economic development;

[0010] S2. Establish a nested execution mechanism for rule chains triggered by scenarios, including defining a scenario feature recognition rule base and pre-setting a main rule chain and sub-rule chains; deploy an instant response layer for mutation events, including setting mutation event monitoring nodes and a weight reset mechanism;

[0011] S3. Implement an obstacle-driven long-term evolution layer, including setting up an obstacle calculation engine in the dynamic knowledge graph, designing a rule evolution feedback mechanism, and realizing the autonomous updating of the rule base.

[0012] S4 achieves a dual-layer adaptive closed-loop evolution of the immediate response layer and the long-term evolution layer through the central node of the rule chain.

[0013] S5. Initiate the multi-objective scheduling optimization process, generate and execute scheduling instructions, input execution feedback data into the long-term evolution layer, and complete the system's self-evolution cycle.

[0014] To further realize the present invention, the following technical solutions may be preferred:

[0015] Preferably, in the dynamic knowledge graph: the flood discharge and sediment transport system includes sediment discharge nodes, floodplain flow nodes, and pre-flood storage nodes, which are respectively associated with the first objective function, empirical mapping relationship, and reservoir storage database in the multi-objective scheduling model; the ecological environment subsystem includes ecological water shortage rate nodes, maximum sediment concentration nodes, and maximum outflow nodes, which are respectively associated with the second objective function, sediment concentration monitoring point, and gate control system in the multi-objective scheduling model; the socio-economic subsystem includes power generation nodes, end-flood storage nodes, and water abandonment nodes, which are respectively associated with the third objective function, hydropower station parameter database, and water resource economic value assessment model in the multi-objective scheduling model; a bidirectional dynamic edge is established between the flood discharge and sediment transport system and the ecological environment subsystem, and the influence weight of sediment discharge on ecological water shortage rate is increased when the floodplain flow is lower than the threshold; a conditional edge is established between the ecological environment subsystem and the socio-economic subsystem, and the correlation strength of power generation on water abandonment is weakened when the ecological water shortage rate exceeds the critical value.

[0016] Preferably, the dynamic knowledge graph includes self-evolving metadata: an obstacle degree calculation sub-node is configured for the evaluation index node; a rule evolution log node is set at the top level of the knowledge graph; and an interface node is configured between the knowledge graph and the improved cumulative prospect theory decision model to convert the non-dominated solution set generated by the multi-objective scheduling algorithm into a decision matrix.

[0017] Preferably, in the scenario-triggered rule chain nested execution mechanism: the scenario feature identification rule base includes flood emergency scenario, ecological crisis scenario and socio-economic peak period scenario nodes;

[0018] Priority determination rules are configured for feature nodes of each scenario type; a main rule chain is preset for each type of feature node, and sub-rule chains are nested within the main rule chain; a rule chain nesting depth limit mechanism is set, and the decompressor is started when the preset number of layers is exceeded.

[0019] Preferably, in the deployment of the real-time response layer for mutation events: mutation event monitoring nodes monitor events such as fluctuations in sediment content, sudden changes in ecological water shortage rate, and sudden changes in power generation load; a dynamic threshold calculation module is configured for mutation events; an impact range evaluator for mutation events is established; an "event-weight" mapping library is constructed in the knowledge graph; and a weight recovery mechanism is set to gradually restore the original weights according to a preset decay curve.

[0020] Preferably, in the long-term evolution layer driven by obstacle degree: the obstacle degree calculation engine periodically analyzes the obstacle contribution rate of each evaluation indicator; establishes an obstacle degree trend prediction module; configures an obstacle degree coupling effect analyzer; and constructs an "obstacle degree-rule" mapping node in the knowledge graph, triggering rule adjustment when the obstacle degree of a specific indicator is higher than a preset threshold for several consecutive days.

[0021] Preferably, in the process of implementing autonomous updates to the rule base: an experience solidification module is set in the rule evolution log node to incorporate rule adjustments that have been verified to be effective multiple times into the standard rule base; a rule conflict detector is configured; and a rule version management system is established to retain historical rule versions and support rollback.

[0022] Preferably, the dual-layer adaptive closed-loop evolution further includes: implementing a parameter hardware rewriting mechanism, establishing a communication connection between the hardware interface node and the event monitor memory, and writing key parameters into the hardware device; setting a hardware parameter security verification mechanism; and periodically initiating a self-test process to evaluate the overall performance of the dual-layer adaptive system.

[0023] Preferably, the process of initiating multi-objective scheduling optimization includes: receiving real-time hydrological data of the watershed and generating an initial parameter set required by the multi-objective scheduling algorithm; running the multi-objective scheduling algorithm to generate a non-dominated solution set and compressing the solution set according to the scenario characteristics; inputting the compressed solution set into the improved cumulative prospect theory decision model; and generating scheduling instructions based on the decision results and issuing them to the control systems of each reservoir.

[0024] A knowledge graph-based self-evolving management system for reservoir scheduling rules, used to implement the above method, the system comprising:

[0025] The dynamic knowledge graph construction module is used to build a dynamic knowledge graph that integrates the three major subsystems;

[0026] The rule chain nested execution engine is used to execute the nested rule chain structure triggered by the execution scenario;

[0027] The mutation event real-time response module is used to monitor mutation events and perform weight resets;

[0028] The long-term evolution management module is used to calculate obstacle degree and drive rule evolution;

[0029] A two-layer adaptive coordinator is used to achieve closed-loop evolution between the immediate response layer and the long-term evolution layer;

[0030] The scheduling optimization and execution module is used to initiate multi-objective scheduling processes and execute instructions.

[0031] The beneficial effects of this invention are:

[0032] This invention constructs a dynamic knowledge graph that integrates three major subsystems: flood control and sediment transport, ecological environment, and socio-economic development. It establishes a scenario-triggered rule chain nested execution mechanism and deploys an instant response layer for sudden events and a long-term evolution layer driven by obstacle degree, forming a two-layer adaptive closed-loop architecture. This enables reservoir scheduling rules to have comprehensive capabilities of real-time perception, dynamic adjustment, and autonomous evolution. Attached Figure Description

[0033] Figure 1 A flowchart illustrating the method of the present invention;

[0034] Figure 2 A schematic diagram illustrating the dynamic evolution of the reservoir scheduling rules of this invention;

[0035] Figure 3 A schematic diagram illustrating the interaction between the scenario-triggered rule chain nested execution mechanism and the two-layer adaptive closed-loop architecture of this invention;

[0036] Figure 4 A schematic diagram illustrating the collaborative operation of the mutation event immediate response layer and the obstacle-driven long-term evolution layer of this invention;

[0037] Figure 5 A schematic diagram of the system structure of the present invention;

[0038] Figure 6 The timing diagram of the reservoir scheduling module of this invention, which classifies responses based on time.

[0039] Figure 7 A schematic diagram illustrating the effect verification of Embodiment 3 of the present invention. Detailed Implementation

[0040] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Example 1

[0043] This embodiment discloses a knowledge graph-based self-evolutionary management method and system for reservoir scheduling rules. Its core lies in constructing a dynamic knowledge graph integrating three subsystems: flood control and sediment transport, ecological environment, and socio-economic factors. Based on this, a scenario-triggered rule chain nested execution mechanism is established. Simultaneously, an immediate response layer for sudden events and a long-term evolution layer driven by obstacle degree are deployed, forming a two-layer adaptive closed-loop architecture. The following is combined with… Figures 1-4 The specific embodiments of the present invention will be described in detail.

[0044] First, a basic framework for a reservoir scheduling knowledge graph is constructed. This framework adopts a three-layer structure of nodes, edges, and metadata. Nodes represent various evaluation indicators or control variables, edges describe the logical relationships or influence paths between nodes, and metadata stores dynamic attributes, historical records, and calculation parameters related to nodes or edges. In the initialization phase, several core evaluation indicator nodes are set up in the flood discharge and sediment transport system, the ecological environment subsystem, and the socio-economic subsystem.

[0045] In the flood control and sediment transport system, nodes for sediment discharge, floodplain flow, and pre-flood storage are set up. The sediment discharge node is directly associated with objective function one in the multi-objective scheduling model. This objective function is defined as the cumulative difference between the outflow and inflow of sediment from each reservoir during the scheduling cycle, used to quantify the effectiveness of sediment scheduling. The floodplain flow node is bound to an empirical mapping relationship, which is constructed based on the 4-year moving average sediment inflow coefficient and measured flow data during the flood season. The current floodplain flow value is generated through a preset empirical formula to assess the flood control capacity of the river channel. The pre-flood storage node is connected to the monthly storage database of each reservoir, acquiring and updating the reservoir's storage status in real time during the pre-flood period, serving as an important boundary condition for scheduling decisions.

[0046] Within the ecological environment subsystem, nodes for ecological water shortage rate, maximum sediment concentration, and maximum outflow are established. The ecological water shortage rate node is associated with objective function two, defined as the ratio of the deviation between the downstream river's ecological water supply flow and the suitable ecological flow. The calculation formula is as follows:

[0047]

[0048] in, To ensure suitable ecological flow, This represents the actual discharge flow. This node is marked as a negative indicator, meaning a higher value indicates poorer ecological protection. During water and sediment regulation, the maximum sediment concentration node is linked to sediment concentration monitoring points at key sections, enabling real-time collection of sediment concentration data in the water flow. The maximum outflow node communicates with the gate opening and closing control system, recording the maximum discharge value during water and sediment regulation, which is useful for assessing the impact of engineering operations on downstream scouring.

[0049] Within the socio-economic subsystem, nodes for power generation, flood storage, and water release are established. The power generation node is linked to a three-dimensional objective function, which is the cumulative total power generation of each reservoir during the scheduling cycle. Its calculation is based on:

[0050]

[0051] in The overall efficiency coefficient of the hydropower station. For time period The water purifier head, The flow rate is used for power generation. The time step is defined as follows. The end-of-flood-season water storage node is linked to the hydropower station output coefficient parameter database, which can be used to assess the potential power generation benefits of water storage at the end of the flood season. The water abandonment node is linked to the water resource economic value assessment model, which quantifies the economic losses caused by water abandonment based on regional water prices, water use structure, and opportunity costs.

[0052] After node initialization, the dynamic connections of the knowledge graph need to be constructed. Two-way dynamic edges are established between the flood discharge and sediment transport system nodes and the ecological environment subsystem nodes. If the value of the flatland flow node consistently falls below a preset threshold, the system will automatically increase the influence weight of the sediment discharge node on the ecological water shortage rate node through a preset nonlinear function. This reflects the squeezing effect of declining flood discharge capacity on ecological water supply space. Conditional edges are established between the ecological environment subsystem nodes and the socio-economic subsystem nodes. When the ecological water shortage rate exceeds a critical value (e.g., 0.5), the system will automatically weaken the correlation strength between the power generation node and the water wastage node to prevent excessive occupation of ecological water use for economic objectives. Simultaneously, feedback edges are established between all evaluation index nodes of the three subsystems and the NSGA optimization model nodes. Deviation data of each index during actual scheduling is promptly fed back to the multi-objective scheduling model, serving as a dynamic constraint parameter in the non-dominated solution set generation process, ensuring that the optimization results remain consistent with the current hydrological situation of the watershed.

[0053] To further support the self-evolution capability of the rules, self-evolving metadata is set in the knowledge graph. Each evaluation metric node is configured with an obstacle degree calculation sub-node, which stores historical obstacle degree data and calculates the current obstacle contribution rate based on the obstacle degree model. The obstacle degree model uses the following formula:

[0054]

[0055] in, For the first The degree of obstruction of the indicator, This is the actual value. For the target ideal value, This is the initial weight for the indicator. The barrier contribution rate is then defined as... This is used to identify the key indicators that have the greatest impact on improving the overall prospect value. A rule evolution log node is set at the top level of the knowledge graph. This node records the triggering reason, parameter changes, execution timestamp, and effect evaluation results for each rule adjustment, forming a traceable evolutionary history chain. In addition, an interface node is configured between the knowledge graph and the improved cumulative prospect theory decision model. This interface node automatically converts the non-dominated solution set generated by the NSGA optimization algorithm into a decision matrix and presets a standardized rule and prospect value calculation parameter library, including reference point settings, value function forms, and weight allocation schemes, providing structured input for subsequent scheme optimization.

[0056] Based on this, a nested execution mechanism of rule chains triggered by scenarios is established. A scenario feature recognition rule base is defined in the knowledge graph, including three core scenario nodes: flood emergency scenarios, ecological crisis scenarios, and peak socio-economic period scenarios. The triggering conditions for flood emergency scenarios are: the floodplain flow is below a threshold for 24 consecutive hours, the pre-flood water storage is above the safety line, and the maximum sediment concentration fluctuation exceeds 25%. The triggering condition for ecological crisis scenarios is: the ecological water shortage rate is greater than 0.5. The triggering condition for peak socio-economic period scenarios is: the power grid load exceeds a preset threshold. Each scenario feature node has priority discrimination rules. When multiple scenarios occur simultaneously, the primary execution scenario is determined according to the priority (flood > ecological > economic) discrimination rules.

[0057] Each scenario feature node has a main rule chain, with the rule chain corresponding to the scenario with the highest priority being the main rule chain. When multiple scenarios occur simultaneously, the quasi-implementation scenario is determined based on the main rule chain. For each scenario feature node, there is a corresponding main rule chain. The main rule chain for the "Emergency Flood Discharge" scenario uses sediment discharge volume as the clustering axis, compressing the non-dominated solution set generated by NSGA into 5 main schemes through K-medoids clustering. The main rule chain can further nest sub-rule chains. The "Sediment Discharge Priority Sub-Rule Chain" triggers two operations: "Gate Coordination Operation Sequence" and "Sediment Scour Effect Prediction." The "Gate Coordination Operation Sequence" generates the opening and closing sequence of multiple gates based on the reservoir group topology to maximize the superposition effect of water flow; the "Sediment Scour Effect Prediction" calls a one-dimensional water and sediment mathematical model to predict the scour and deposition changes of the riverbed in the next 72 hours. To ensure decision-making efficiency, a rule chain nesting depth limit mechanism is set. If the nesting level exceeds three levels, the system will automatically start the unset compressor and use principal component analysis to reduce the dimensionality of the current unset, retaining principal components with a cumulative variance contribution rate of not less than 95%, so as to avoid response delays caused by excessive nesting.

[0058] The dynamic execution flow of the rule chain is automatically controlled by the knowledge graph. Once the triggering conditions of a scene feature are detected, the system activates the starting node of the corresponding main rule chain and locks down interference signals from other scene features during execution to prevent logical conflicts caused by concurrent multi-scene execution. During the execution of the main rule chain, the running status of the sub-rule chains is monitored in real time. If abnormal signals such as operation timing conflicts, resource competition, or physical constraint violations are detected, the execution order of subsequent nodes is dynamically adjusted, or infeasible branches are skipped. After the rule chain execution is completed, a scheduling plan is automatically generated and pushed to the improved cumulative theoretical decision-making model. At the same time, the execution timestamps, key parameters, and intermediate calculation results of each step in the plan are marked, providing a high-quality data foundation for subsequent rule evolution.

[0059] To address sudden disturbances, an immediate response layer for abrupt events was deployed. A monitoring node for abrupt events was established within the knowledge graph, interfacing with the reservoir's real-time monitoring system to focus on three key events: fluctuations in sediment content, sudden changes in ecological water shortage rate, and abrupt changes in power generation load. Each type of abrupt event is equipped with a dynamic threshold calculation module. This module calculates the mean and standard deviation based on a sliding window of data from the past 30 days, setting the dynamic threshold to the mean plus or minus twice the standard deviation to avoid misjudgments due to fixed thresholds under different hydrological conditions. Simultaneously, an impact range assessor for abrupt events was established. Once abrupt events are detected, the system automatically assesses their impact on the three subsystems, generating an event severity score using a weighted summation method.

[0060]

[0061] in, The change in obstacle degree of each subsystem Preset weights.

[0062] An instant response weight reset mechanism was also designed. An "event-weight" mapping library was built into the knowledge graph, storing the correspondence between different mutation event types and the weights of each objective function in the multi-objective scheduling model. For example, when the sediment concentration fluctuates by more than 25%, the weight of the flood discharge dimension will automatically increase by 40%, which is equivalent to multiplying the weight of objective function one by 1.4. This weight reset mechanism is deeply embedded in the rule execution process. Once a mutation event is detected during the rule chain operation, the system will immediately pause the current calculation and call the weight reset module to update the objective function weights of the multi-objective scheduling model. A weight recovery mechanism was also set up. After the mutation event ends, the system will gradually restore the original weights according to the event's duration and impact, following a preset exponential decay curve. The decay function is:

[0063]

[0064] in, For the original weights, For peak weight, The attenuation coefficient is... This represents the number of hours after the event ends, preventing system oscillations caused by sudden changes in weights.

[0065] The response effectiveness is evaluated through a real-time verification module. After the weight reset is executed, the system automatically compares the deviations between the theoretical and actual values ​​of key indicators before and after the adjustment. The established evaluation criteria are: a response is considered effective if the actual flood discharge efficiency increases by at least 15% or the ecological water shortage rate decreases by at least 10%; otherwise, it is marked as an invalid response. The verification results are fed back to the long-term evolution layer in real time, providing immediate data support for rule base updates and forming a "monitoring-response-verification" micro-loop.

[0066] For long-term operation, an obstacle degree calculation node has been added to the knowledge graph. This node drives the obstacle degree calculation engine daily to call the obstacle degree model to calculate the obstacle degree contribution rate of each evaluation indicator, especially those indicators that significantly impact the improvement of the overall prospect value. An obstacle degree trend prediction node has also been added. This node uses the ARIMA time series model to predict the obstacle degree change trend of key indicators in the next seven days based on the obstacle degree data of the previous seven days, thus identifying potential future system risks in advance. An obstacle degree coupling effect analyzer has been added, using the Granger causality test method to discover the coupling effects between indicators. For example, an increase in the ecological water shortage rate will increase the discharge of waste water, meaning that the water shortage rate has a coupling effect on the generation of waste water, and outputs a coupling compensation coefficient to guide subsequent rule adjustments. (Coupling compensation coefficient) Defined as:

[0067]

[0068] Indicators The impact of changes in obstacle degree on indicators The marginal impact.

[0069] A rule evolution feedback mechanism was designed. An "obstacle degree-rule" mapping node was added to the knowledge graph. When the obstacle degree of sediment discharge exceeds the limit for more than three days during a certain period of the flood season, the rule evolution feedback module is driven to automatically adjust the rules. For example, if the sediment discharge obstacle degree exceeds the limit, the system relaxes the upper limit of the pre-flood water storage constraint by 8%, that is, the original constraint... Modified to This modification involves not only the constraint parameters of the multi-objective scheduling model but also the triggering conditions of scene feature nodes and the execution logic of the rule chain. A rule evolution effect tracker is set up. An "obstacle degree-rule" mapping node is added to the long-term runtime layer knowledge graph. When a rule is adjusted, the rule evolution effect tracker records the system's performance after the rule correction, including changes in obstacle degree, decision stability, and execution deviation, forming an evolutionary logic of "adjustment-effect-re-adjustment" to avoid repeated ineffective adjustments.

[0070] To enable autonomous updates to the rule base, an experience solidification module is implemented in the rule evolution log node. When a rule has been repeatedly adjusted and verified to be effective (i.e., five consecutive adjustments with a reduced key indicator obstacle level and no negative cascading effects after each adjustment), the system will fix it in the standard rule base. Before new rules are added to the base, the rule conflict detector uses a combination of logical consistency checks and simulation analysis. If a conflict is found, a rule optimization algorithm is generated to produce a compromise rule. The rule version management system retains all previously effective rules and labels them with different versions (e.g., "only applicable to high sediment content years," "dry season mode"). In special circumstances, rules can be rolled back to previously effective rules.

[0071] The system has adaptively entered a closed-loop evolution on a two-layer architecture. A central node for the rule chain is placed at the core of the knowledge graph, serving as a data exchange node between the two layers. It receives and forwards control signals between the two layers, implementing data flow separation. The immediate layer data flow uses a high-priority channel (such as UDP protocol with QoS marking) to ensure that the time from event detection to weight reset does not exceed 50 milliseconds. The long-term layer data flow uses an encrypted storage channel to ensure that the time from event detection to weight reset does not exceed 50 milliseconds and that historical data is not leaked. A two-layer conflict coordinator is implemented. When commands from the immediate layer and the long-term layer conflict, such as an immediate layer command increasing the flood weight while the long-term layer needs to decrease it for ecological balance, the system executes the immediate layer command according to a preset priority (higher time sensitivity first), suspending the long-term layer command, and then rebalancing after the event ends.

[0072] The method to enhance the physical layer rule solidification capability enables a parameter hardware rewriting mechanism at the physical layer. Hardware interface nodes are added to the knowledge graph, establishing an I²C connection between them and the EEPROM memory of the event monitor, supporting direct rewriting of hardware parameters. When the long-term evolution layer determines that a rule revision remains effective (e.g., after 30 days, the obstacle degree continues to decrease), key parameters in that rule revision (such as the mutation magnitude triggered by a mutation event, weight decay coefficient) are written into the hardware. Hardware parameter safety checks are implemented. Before writing to the hardware, the parameters are verified for rationality, including range checks (whether they fall within the physical boundaries of the device), monotonicity checks (whether they conform to historical trends), and simulation verification (whether they can achieve the expected effect in the digital twin), avoiding system failures caused by writing parameters that exceed limits into the hardware.

[0073] The system completes closed-loop self-evolution. Every 7 days, the system initiates a self-check process to evaluate the overall performance of the two-layer adaptive system. The evaluation generates a comprehensive score, which includes decision-making efficiency (average time from event occurrence to solution generation), rule fitness (the decrease in obstacle severity after rule adjustment), and system stability (the standard deviation of solution fluctuations). Simultaneously, comparative performance tests are conducted. The system is placed under the same historical hydrological scenarios to compare its sensitivity, thereby verifying the system's self-evolutionary advantages. An evolution report is then generated and pushed to the management terminal, thus forming a complete "execution-evaluation-evolution" evidence chain.

[0074] During the system's collaborative operation phase, a multi-objective scheduling and optimization process is initiated. The system receives real-time hydrological data from the watershed, including rainfall, inflow, sediment concentration, and grid load. This data is then preprocessed using a knowledge graph framework to generate the initial parameter set required by the NSGA optimization algorithm. The NSGA-II optimization algorithm is then run to generate a set of non-dominated solutions. Simultaneously, based on the characteristics of the current scenario, the corresponding nested rule chain structure is activated to perform scenario-based compression on the non-dominated solution set. Finally, the compressed representative solution set is input into the improved cumulative prospect theory decision model. The prospect value calculation formula used in this model is as follows:

[0075]

[0076] in, Let be the decision weight function. For value function, For the scheme in the first The standardized values ​​for each objective are then combined with the weight reset results from the immediate response layer to calculate the improved overall prospect value.

[0077] Next, the decision-making plan is implemented and feedback is provided. Based on the improved cumulative prospect theory decision results, specific dispatch instructions (including the opening degree of each reservoir gate, power generation plan, and storage and release strategy) are generated and issued to the control systems of each reservoir. The execution of dispatch instructions is monitored in real time, and the actual values ​​of key indicators (such as actual discharge flow, sediment concentration, power generation, etc.) are recorded and compared with the target values ​​to calculate the execution deviation. The execution deviation data is fed back to the obstacle analysis system, triggering a new round of obstacle calculation and rule evolution assessment.

[0078] Finally, when the system has run for 24 hours or detects a major event (such as a catastrophic flood or a breach of ecological red lines), it will automatically initiate a complete evolutionary cycle, integrating data from the immediate response layer and the long-term evolutionary layer. Based on the evolutionary cycle analysis results, the rule base, scene feature library, and hardware parameters in the knowledge graph are updated, completing the system's self-evolution. The system monitoring status is reset, temporary caches are cleared, and preparations are made for the next round of self-evolutionary cycles, allowing the reservoir scheduling rules to be continuously optimized, achieving coordinated development of multiple objectives related to water and sediment, ecology, and the economy.

[0079] Example 2

[0080] To implement the technical solution of Embodiment 1, this embodiment discloses a reservoir scheduling rule self-evolution management system based on knowledge graphs, referring to... Figure 5 and Figure 6 It includes the following six core functional modules:

[0081] The dynamic knowledge graph construction module is responsible for initializing the core evaluation indicator nodes of the three subsystems: flood control and sediment transport, ecological environment, and socio-economic development. It establishes dynamic connections between nodes and configures self-evolving metadata. This module stores and queries the knowledge graph through a database, supports dynamic updates of node attributes and real-time adjustments of edge weights, and provides fundamental data structure support for the entire system.

[0082] The nested rule chain execution engine is responsible for scene feature recognition, rule chain structure management, and dynamic execution control. This module includes a scene feature recognition unit for detecting three scenarios: flood emergencies, ecological crises, and peak socio-economic periods; a nested rule chain structure management unit for maintaining the nesting relationship between the main and sub-rule chains; and a dynamic execution control unit for monitoring the rule chain execution status and handling anomalies. This engine employs an event-driven architecture, ensuring both high efficiency and reliability in rule chain execution.

[0083] The real-time response module for sudden events is specifically designed to monitor sudden events, reset weights, and verify the effects. It comprises a sudden event monitoring unit, a weight reset logic unit, and a response effect verification unit. The sudden event monitoring unit interfaces with the reservoir's real-time monitoring system to continuously monitor three key events: fluctuations in sediment concentration, sudden changes in ecological water shortage rate, and abrupt changes in power generation load. The weight reset logic unit stores an "event-weight" mapping library and can perform weight adjustments. The response effect verification unit is used to evaluate the actual effect of the weight reset. This module is designed with a high-priority processing flow, ensuring that the entire process from event detection to weight reset is completed within 50 milliseconds.

[0084] The long-term evolution management module comprises an obstacle calculation engine unit, a rule evolution feedback unit, and a rule base self-updating unit. The obstacle calculation engine unit periodically analyzes the obstacle contribution rate of each evaluation metric; the rule evolution feedback unit triggers rule adjustments when the obstacle level of a specific metric consistently exceeds the limit; and the rule base self-updating unit implements rule solidification, conflict detection, and version management. This module operates on a 24-hour cycle, complementing the immediate response module to ensure long-term system performance optimization.

[0085] The two-layer adaptive coordinator's role is to facilitate data exchange, conflict resolution, and closed-loop evolutionary control between the immediate response layer and the long-term evolutionary layer. It contains a central node in the rule chain, which acts as the hub for data flow exchange between the two layers. A data coordination unit separates the two-layer data flows. A conflict resolution unit handles conflicts between the two layers' instructions. A hardware parameter rewriting unit writes verified and valid rule parameters to the hardware device. The two-layer adaptive coordinator is the core of the entire system's "two-layer adaptive" characteristic, enabling the system to find a balance between responding to sudden events and long-term optimization.

[0086] The scheduling optimization and execution module is responsible for initiating the multi-objective scheduling optimization process, generating scheduling instructions, and collecting feedback data. Its multi-objective optimization calculation unit runs the NSGA-II algorithm to generate a set of undominated solutions. The multi-criteria decision-making unit calculates the improved comprehensive prospect value and then selects the optimal solution. The scheduling instruction generation and execution unit transforms the decision results into specific scheduling instructions. The data acquisition and feedback unit monitors the execution status in real time and calculates deviations. This module serves as the interface between the system and the physical reservoir, directly impacting the effectiveness of the scheduling scheme.

[0087] These six functional modules are decoupled through a microservice architecture, and the modules exchange data through API interfaces. The system is deployed in a containerized manner, can run on industrial-grade server clusters, and can also perform system monitoring and analysis. These modules work together to achieve self-evolving management of reservoir scheduling rules, ensuring a multi-dimensional balance between flood control safety, ecological health, and economic development in a complex and ever-changing watershed environment.

[0088] Example 3

[0089] In this embodiment, to verify the technical effect of the present invention, reference is made to... Figure 7 The following reference and comparative tests were conducted.

[0090] A large reservoir group in the Yellow River Basin was selected as the application object, and the system described in this invention was run continuously for 90 days. During this period, the system experienced one emergency flood discharge scenario (the flow rate at the floodplain was below the threshold for 30 consecutive hours) and two sudden events in the ecological water shortage rate. The system successfully triggered the corresponding rule chain, completed weight reset and scheme compression, and ultimately achieved an 18.7% increase in sediment discharge, a 12.3% reduction in the ecological water shortage rate, and a 9.5% reduction in water wastage.

[0091] In the comparative study, a traditional NSGA-II scheduling method combined with static weights was used under the same hydrological conditions. The results showed that sediment discharge increased by only 6.2%, the ecological water shortage rate decreased by 4.1%, but water wastage increased by 3.8%. Furthermore, the strategy could not be adjusted in time when abrupt events occurred, resulting in a serious disconnect between the scheduling scheme and actual needs.

[0092] The table below shows the comparison results of key indicators between the examples and the comparative examples:

[0093] Index Reference Example Results Comparative Example Results Lifting Range Desanding Capacity Improvement Rate 18.7% 6.2% 12.5% Ecological Water Deficiency Rate Reduction Rate 12.3% 4.1% 8.2% Abandoned Water Volume Change Rate -9.5% +3.8% 13.3% Average Decision Response Time 38 milliseconds 120 milliseconds — Effective Number of Rule Adjustments 7 times 0 times —

[0094] The above data shows that by constructing a dynamic knowledge graph and a two-layer adaptive closed-loop architecture, this invention significantly improves the real-time performance, adaptability, and autonomous evolution capability of reservoir scheduling rules, effectively solving the problem of multi-objective coordination.

[0095] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based self-evolutionary management method for reservoir scheduling rules, characterized in that, include: S1. Construct a dynamic knowledge graph that integrates the three major subsystems of flood control and sediment transport, ecological environment and socio-economic development; S2. Establish a nested execution mechanism for rule chains triggered by scenarios, including defining a rule base for scenario feature recognition and pre-setting a main rule chain and sub-rule chains; Deploy a real-time response layer for mutation events, including setting up mutation event monitoring nodes and a weight reset mechanism; S3. Implement an obstacle-driven long-term evolution layer, including setting up an obstacle calculation engine in the dynamic knowledge graph, designing a rule evolution feedback mechanism, and realizing the autonomous updating of the rule base. S4 achieves a dual-layer adaptive closed-loop evolution of the immediate response layer and the long-term evolution layer through the central node of the rule chain. S5. Initiate the multi-objective scheduling optimization process, generate and execute scheduling instructions, input execution feedback data into the long-term evolution layer, and complete the system's self-evolution cycle.

2. The method according to claim 1, characterized in that, In the dynamic knowledge graph: the flood discharge and sediment transport system includes sediment discharge nodes, floodplain flow nodes, and pre-flood storage nodes, which are respectively associated with the first objective function, empirical mapping relationship, and reservoir storage database in the multi-objective scheduling model; the ecological environment subsystem includes ecological water shortage rate nodes, maximum sediment concentration nodes, and maximum outflow nodes, which are respectively associated with the second objective function, sediment concentration monitoring point, and gate control system in the multi-objective scheduling model; the socio-economic subsystem includes power generation nodes, flood-end storage nodes, and water abandonment nodes, which are respectively associated with the third objective function, hydropower station parameter database, and water resource economic value assessment model in the multi-objective scheduling model; a bidirectional dynamic edge is established between the flood discharge and sediment transport system and the ecological environment subsystem, and the influence weight of sediment discharge on ecological water shortage rate is increased when the floodplain flow is below the threshold; a conditional edge is established between the ecological environment subsystem and the socio-economic subsystem, and the correlation strength of power generation on water abandonment is weakened when the ecological water shortage rate exceeds the critical value.

3. The method according to claim 2, characterized in that, The dynamic knowledge graph is configured with self-evolving metadata: an obstacle degree calculation sub-node is configured for the evaluation index node; a rule evolution log node is set at the top level of the knowledge graph; and an interface node is configured between the knowledge graph and the improved cumulative prospect theory decision model to convert the non-dominated solution set generated by the multi-objective scheduling algorithm into a decision matrix.

4. The method according to claim 1, characterized in that, In the nested execution mechanism of the rule chain triggered by the scenario: the scenario feature recognition rule base includes nodes for flood emergency scenario, ecological crisis scenario and socio-economic peak period scenario; Priority determination rules are configured for feature nodes of each scenario type; a main rule chain is preset for each type of feature node, and sub-rule chains are nested within the main rule chain; a rule chain nesting depth limit mechanism is set, and the decompressor is started when the preset number of layers is exceeded.

5. The method according to claim 1, characterized in that, In the deployment of the real-time response layer for mutation events: mutation event monitoring nodes monitor events such as fluctuations in sediment content, sudden changes in ecological water shortage rate, and sudden changes in power generation load; a dynamic threshold calculation module is configured for mutation events; an impact range evaluator for mutation events is established; an "event-weight" mapping library is constructed in the knowledge graph; and a weight recovery mechanism is set to gradually restore the original weights according to a preset decay curve.

6. The method according to claim 1, characterized in that, In the long-term evolution layer driven by the implementation of obstacle degree: the obstacle degree calculation engine periodically analyzes the obstacle contribution rate of each evaluation indicator; establishes an obstacle degree trend prediction module; configures an obstacle degree coupling effect analyzer; and constructs an "obstacle degree-rule" mapping node in the knowledge graph, triggering rule adjustment when the obstacle degree of a specific indicator is higher than a preset threshold for several consecutive days.

7. The method according to claim 6, characterized in that, In the process of achieving autonomous updates to the rule base: an experience solidification module is set up in the rule evolution log node to incorporate rule adjustments that have been verified to be effective multiple times into the standard rule base; a rule conflict detector is configured; and a rule version management system is established to retain historical rule versions and support rollback.

8. The method according to claim 1, characterized in that, The dual-layer adaptive closed-loop evolution also includes: implementing a parameter hardware rewriting mechanism, establishing a communication connection between the hardware interface node and the event monitor memory, and writing key parameters into the hardware device; setting a hardware parameter security verification mechanism; and periodically initiating a self-test process to evaluate the overall performance of the dual-layer adaptive system.

9. The method according to claim 1, characterized in that, The process of initiating multi-objective scheduling optimization includes: receiving real-time hydrological data of the watershed and generating an initial parameter set required by the multi-objective scheduling algorithm; running the multi-objective scheduling algorithm to generate a non-dominated solution set and compressing the solution set according to the scenario characteristics; inputting the compressed solution set into the improved cumulative prospect theory decision model; and generating scheduling instructions based on the decision results and issuing them to the control systems of each reservoir.

10. A knowledge graph-based reservoir scheduling rule self-evolution management system, used to implement the method according to any one of claims 1 to 9, characterized in that, The system includes: The dynamic knowledge graph construction module is used to build a dynamic knowledge graph that integrates the three major subsystems; The rule chain nested execution engine is used to execute the nested rule chain structure triggered by the execution scenario; The mutation event real-time response module is used to monitor mutation events and perform weight resets; The long-term evolution management module is used to calculate obstacle degree and drive rule evolution; A two-layer adaptive coordinator is used to achieve closed-loop evolution between the immediate response layer and the long-term evolution layer; The scheduling optimization and execution module is used to initiate multi-objective scheduling processes and execute instructions.

Images