Cascade reservoir collaborative flood control scheduling decision-making method coupled with meteorological-hydrological-hydraulic model

Abstract

The invention discloses a cascade reservoir collaborative flood control scheduling decision-making method of a coupling meteorological-hydrological-hydraulic model. The method comprises the steps of multi-source meteorological data fusion and probability forecast generation, dynamic coupling hydrological-hydraulic simulation, risk entropy driven collaborative optimization decision-making, digital twin platform verification and correction, instruction execution and closed loop feedback. Through multi-technology fusion and an intelligent optimization mechanism, the scientificity, timeliness and safety of flood control scheduling are remarkably improved. In the weather forecast stage, multi-source data are integrated to output ensemble rainfall forecast scene data, rainfall input uncertainty is reduced from the source, it is ensured that initial driving precision of flood simulation is improved, and a reliable input basis is provided for subsequent model coupling; in the hydrological-hydraulic dynamic coupling stage, a coarse-fine grid dynamic division strategy is adopted to reduce redundancy calculation, the asynchronous pipeline technology enables the flood routing calculation efficiency to meet the real-time scheduling requirement, and the simulation speed is greatly accelerated.

Description

Technical Field

[0001] This invention relates to a reservoir flood control scheduling decision-making method, specifically a cascade reservoir collaborative flood control scheduling decision-making method coupled with meteorological-hydrological-hydraulic models, belonging to the technical field of cascade reservoir flood control scheduling optimization in water conservancy projects. Background Technology

[0002] Currently, my country's water conservancy and hydropower projects are developing rapidly, with cascade reservoirs forming successively in major river basins. Therefore, flood control management of cascade reservoirs has become a top priority in flood prevention. Cascade reservoir flood control involves a wide area and complex situations. On the one hand, the flood control standards of upstream and downstream water conservancy and hydropower projects within a cascade are not entirely the same, and there are even significant mismatches. If a dam in one reservoir within a cascade fails, it could trigger a chain reaction in downstream reservoirs, with potentially catastrophic consequences. On the other hand, there is a general lack of flood control capacity in river embankments, and some embankments even exhibit flood control risks that are severely out of balance with the rapidly developing national economy of the region.

[0003] Currently, flood control scheduling of cascade reservoirs faces three major technical bottlenecks: First, fragmented models lead to error accumulation. Existing technologies, such as the method and system for joint flood control optimization scheduling of cascade reservoir groups disclosed in CN102817335B, use an improved genetic algorithm to achieve hierarchical scheduling. Although this improves decision-making efficiency by 30% through flood category judgment and optimized model selection, meteorological forecasting, hydrological simulation, and hydraulic evolution still operate independently, lacking a dynamic error propagation and correction mechanism. This results in insufficient accuracy of flood evolution simulation, with flood peak prediction errors generally exceeding 18%. Second, the scheduling objectives are singular. For example, the cascade reservoir optimization scheduling system disclosed in CN118485261A, although integrating multi-source data fusion and decision support systems, has its optimization... The objectives are limited to water level control and power generation efficiency improvement, failing to quantify dam failure risks and downstream socio-economic losses, and neglecting the assessment of inundation losses in high-risk areas, resulting in insufficient disaster resilience of the dispatching scheme.25 Thirdly, there is a delay in real-time response. Traditional methods rely on static rules or human experience for decision-making. For example, the Hongshui River cascade reservoirs need to pre-determine a set of "peak-shaving duration - controlled release timing" schemes, which cannot dynamically respond to the spatiotemporal variability of rainstorms. The peak reduction effect fluctuates by 700~2500 m³ / s. Although the cascade reservoir group flood control dispatching optimization system disclosed in announcement number CN118396806A alleviates the dimensionality problem through cloud models, it does not solve the timeliness bottleneck of multi-model collaborative computing. The dispatching scheme generation time under rainstorm scenarios still exceeds 1 hour.78 Furthermore, although existing coupled models such as CHHMS (Surface Water and Groundwater Coupled Simulation System) achieve hydrological-hydraulic mechanism fusion, they do not integrate probabilistic forecasting and real-time risk decision-making, and their computational efficiency is insufficient to meet the minute-level response requirements. Summary of the Invention

[0004] The purpose of this invention is to provide a method for coordinated flood control scheduling of cascade reservoirs that couples meteorological-hydrological-hydraulic models in order to solve at least one of the above-mentioned technical problems.

[0005] This invention achieves the above objective through the following technical solution: a cascade reservoir collaborative flood control scheduling decision-making method coupled with meteorological-hydrological-hydraulic models, which includes the following steps: Step 1: Multi-source meteorological data fusion and probabilistic forecast generation. Meteorological data from satellite cloud images, radar echoes, and ground meteorological stations are obtained using a mesoscale numerical weather prediction system to generate forecast scenarios for the next 72 hours and output ensemble rainfall forecast scenario data. This reduces the uncertainty of rainfall input from the source, ensures the improvement of the initial driving accuracy of flood simulation, and provides a reliable input basis for subsequent model coupling. Step 2: Dynamically couple hydrological-hydraulic simulation to break down the disconnect between meteorological, hydrological, and hydraulic models. The hydrological-hydraulic simulation includes hydrological and hydraulic phases. In the hydrological phase, multiple runoff sequences are generated by inputting aggregated rainfall forecast scenario data into the HEC-HMS surface runoff model. Parameters are corrected in real time every 30 minutes by triggering the root mean square error between measured and simulated flow rates. In the hydraulic phase, the runoff sequences are input into the HEC-RAS river hydraulic analysis model based on a parallel architecture of a graphics processor, and the flood inundation process is simulated. The model outputs inundation depth, flow velocity, and arrival time, greatly improving the efficiency of flood evolution calculation and meeting the minute-level response requirements in heavy rain scenarios. Step 3: Risk entropy-driven collaborative optimization decision-making to achieve a multi-objective balance between flood control safety and water resource utilization. By calculating the real-time risk entropy of the reservoir, overly conservative scheduling in high-value areas is avoided, and the dynamic safety domain water level of the reservoir is constrained to prevent dam failure risk. The real-time risk entropy is defined as the logarithmic product of the dam failure probability and the downstream impact index. The downstream impact index is the product of population density and the weight of key facilities. The dynamic safety domain water level is the upper limit of the water level dynamically calculated from real-time engineering safety monitoring data. Step 4: Verification and correction of the digital twin platform. Based on the digital twin platform, the scheduling scheme is simulated. Through the simulation of the hydrological and hydraulic stages, and the calculation of the real-time risk entropy and dynamic safety zone water level of the reservoir, the center coordinates of the areas with water depth > 2m marked in the real-time thermal distribution map of the reservoir are identified as the maximum risk points. The accuracy of high-risk area identification is improved in the simulated scheduling scheme. The optimal scheduling scheme is simulated by combining reinforcement learning agent. The flood discharge sequence of the reservoir is adjusted based on the simulated scheduling scheme. Step 5: Command execution and closed-loop feedback. The gate control command with a three-level emergency response mechanism is sent to the automatic control system of the reservoir gate through the industrial Ethernet protocol. The gate opening and closing control is performed according to the corresponding level of the command, and the water level deviation is monitored in real time. When the actual water level deviates from the predicted value by ≥5% for 1 hour or the single deviation is ≥10%, the model parameter recalibration is automatically triggered.

[0006] As a further aspect of the present invention: the parallel architecture of the graphics processor specifically includes: The graphics processor uses a coarse grid division strategy for the main river channel area and dynamically adjusts the grid size according to the complexity of the river channel topography; The graphics processor deploys a dynamic adaptive grid in the floodplain. By monitoring the rate of change of water level and the terrain gradient in real time, when the water level fluctuation exceeds the stability threshold or the terrain undulation increases sharply, the grid densification mechanism is automatically triggered to subdivide the local area into high-precision computing units. The graphics processor uses shared memory to pre-store water depth and flow velocity data of adjacent grids; While the graphics processor calculates the current water flow status, the terrain elevation data for the next time time is preloaded simultaneously through a high-speed bus, realizing parallelization of calculation and transmission, and improving the overall efficiency of flood evolution simulation to the level required for real-time scheduling.

[0007] As a further aspect of the present invention: the parameters for real-time correction are measured flow data, which include, but are not limited to, the state vectors of saturated hydraulic conductivity, number of runoff curves, and channel roughness; and during correction, the state vectors are updated every 30 minutes by integrating measured flow data to ensure that the root mean square error of the runoff simulation after correction is ≤15%.

[0008] As a further aspect of the present invention, the constraints of the dynamic safety zone water level include flood discharge flow ≤ 90% of the design flow and real-time water level ≤ upper limit of the dynamic safety zone. The upper limit of the dynamic safety zone is dynamically calculated based on the monitoring data of the dam stress sensor and the seepage line.

[0009] As a further aspect of the present invention, the construction of the reinforcement learning agent specifically includes: Define a multidimensional state space that integrates reservoir status and inundation map. The multidimensional state space integrates the spatial features of time series data of reservoir group water level, inflow, risk entropy value and 30-meter resolution inundation raster map. The design incorporates continuous action space control to regulate the opening of flood discharge facilities. The action space is defined as the continuous adjustment commands for the opening of each reservoir gate. A strategy evaluation model is constructed, which includes two paths: temporal feature extraction and spatial feature extraction. The extracted temporal path is the reservoir state sequence, and the extracted spatial path is the reservoir inundation map features. The agent is trained based on historical disaster scenarios, using a historical flood scenario dataset.

[0010] As a further aspect of the present invention: when simulating the scheduling plan, a social impact assessment is conducted. The social impact assessment specifically includes: generating a real-time heat map by integrating multi-source population distribution data; calculating the number of affected people by overlaying the flood inundation range; assessing disaster losses according to the inundation depth; and setting a population impact threshold to trigger a manual decision-making confirmation process.

[0011] As a further aspect of this invention: the digital twin platform integrates a three-dimensional flood scene rendered in real time through holographic projection, with the flood depth mapped using a four-color gradient of blue-green-yellow-red, and the flow velocity vector marked with dynamic arrows; the digital twin platform integrates a gesture interaction system that recognizes swiping, grabbing, and double-clicking actions, enabling gate opening adjustment, scene rotation, and scheme confirmation; the digital twin platform integrates a multi-user terminal that supports multi-party collaborative operation, with all user terminals automatically recording parameter modifications and risk assessment results as log reports, the log reports including but not limited to timestamps, operator IDs, and risk entropy change curves.

[0012] As a further aspect of this invention: the execution rules of the three-level emergency response mechanism are as follows: Level I Response: Open all flood diversion gates, and cascade reservoirs discharge floodwater in the topological order from upstream to downstream, with the discharge flow not exceeding 120% of the downstream safe capacity. Simultaneously, emergency management departments will evacuate the population in high-risk areas. Level II Response: Flood discharge tasks are dynamically allocated according to the risk entropy ratio, with high-risk reservoirs bearing the main responsibility for flood discharge; Level III Response: Close unnecessary floodgates, store water according to the reservoir's regulation capacity, limit flood discharge and prioritize water storage, and control the water storage rate according to the reservoir's regulation capacity.

[0013] The beneficial effects of this invention are: 1. This invention significantly improves the scientific rigor, timeliness, and safety of flood control scheduling through multi-technology fusion and intelligent optimization mechanisms. During the weather forecasting phase, it integrates multi-source data output sets of rainfall forecast scenario data, reducing the uncertainty of rainfall input from the source, ensuring improved initial driving accuracy of flood simulation, and providing a reliable input foundation for subsequent model coupling; 2. In the hydrological-hydraulic dynamic coupling stage, this invention achieves a breakthrough improvement in flood simulation efficiency: the coarse-fine grid dynamic partitioning strategy reduces redundant calculations, the asynchronous pipeline technology enables the flood evolution calculation efficiency to meet real-time scheduling requirements, the resolution of inundation depth and flow velocity prediction is improved, and the simulation speed is greatly accelerated. 3. In the optimization decision-making process, this invention utilizes reinforcement learning agents to dynamically adjust the flood discharge sequence: a dual-path network enables precise output of action strategies, and combined with a social impact assessment model, a three-level manual confirmation mechanism is triggered when the affected population is ≥10,000 people, ensuring the reliability of major decisions. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0015] 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.

[0016] Example 1, as Figure 1 As shown, a cascade reservoir collaborative flood control scheduling decision-making method coupled with meteorological-hydrological-hydraulic models is presented. This cascade reservoir collaborative flood control scheduling decision-making method includes the following steps: Step 1: Multi-source meteorological data fusion and probabilistic forecast generation. Meteorological data from satellite cloud images, radar echoes, and ground meteorological stations are obtained using a mesoscale numerical weather prediction system to generate a forecast scenario for the next 72 hours (spatial resolution ≤ 1 km, temporal resolution ≤ 1 hour). The ensemble rainfall forecast scenario data is then output to reduce the uncertainty of rainfall input from the source, ensure the improvement of the initial driving accuracy of flood simulation, and provide a reliable input basis for subsequent model coupling. Step 2: Dynamically couple hydrological-hydraulic simulation, breaking down the disconnect between meteorological, hydrological, and hydraulic models. The hydrological-hydraulic simulation includes hydrological and hydraulic phases. In the hydrological phase, multiple runoff sequences are generated by inputting ensemble rainfall forecast scenario data into the HEC-HMS surface runoff model. Parameters are corrected in real-time every 30 minutes using the root mean square error between measured and simulated flow rates, addressing the flood peak prediction bias caused by traditional calibration lags. In the hydraulic phase, the runoff sequences are input into the HEC-RAS river hydraulic analysis model based on a parallel architecture using a graphics processor, simulating the floodplain inundation process and outputting inundation depth, flow velocity, and arrival time, thus improving the efficiency of flood evolution calculations. This greatly enhances the system's capabilities, meeting minute-level response requirements in heavy rain scenarios. It should be noted that the HEC-HMS surface runoff model is used to simulate and analyze hydrological processes within a watershed, such as rainfall generation, runoff peaks, and flood propagation. The HEC-RAS river hydraulic analysis model is a hydraulic model for one-dimensional steady / unsteady flow, mainly used for open channel flow analysis and the determination of floodplain areas. The results obtained from the model can be used for flood area management and flood safety research and analysis to evaluate the extent and severity of flood inundation areas. The parallel architecture of the graphics processor enables parallel computing of the graphics processor, achieving large-scale data processing and computation through the collaborative work of multiple graphics processor nodes. Step 3: Risk entropy-driven collaborative optimization decision-making to achieve a multi-objective balance between flood control safety and water resource utilization. By calculating the real-time risk entropy of the reservoir, overly conservative scheduling in high-value areas is avoided, and the dynamic safety domain water level of the reservoir is constrained to prevent dam failure risk. The real-time risk entropy is defined as the logarithmic product of the dam failure probability and the downstream impact index. The downstream impact index is the product of population density and the weight of key facilities. The dynamic safety domain water level is the upper limit of the water level dynamically calculated from real-time engineering safety monitoring data. Step 4: Verification and correction of the digital twin platform. Based on the digital twin platform, the scheduling scheme is simulated. Through the simulation of the hydrological and hydraulic stages, and the calculation of the real-time risk entropy and dynamic safety zone water level of the reservoir, the center coordinates of the areas with water depth > 2m marked in the real-time thermal distribution map of the reservoir are identified as the maximum risk points. The accuracy of high-risk area identification is improved in the simulated scheduling scheme. The optimal scheduling scheme is simulated by combining reinforcement learning agents. The flood discharge sequence of the reservoir is adjusted based on the simulated scheduling scheme. The simulation generation time of the scheduling scheme is shortened from hours to minutes by using reinforcement learning agents. Step 5: Command execution and closed-loop feedback. The gate control command with a three-level emergency response mechanism is sent to the automatic control system of the reservoir gate through the industrial Ethernet protocol. The gate opening and closing control is performed according to the corresponding level of the command, and the water level deviation is monitored in real time. When the actual water level deviates from the predicted value by ≥5% for 1 hour or the single deviation is ≥10%, the model parameter recalibration is automatically triggered.

[0017] Example 2, in addition to all the technical features included in Example 1, also includes: a parallel graphics processor architecture specifically comprising: The graphics processor uses a coarse grid division strategy for the main river channel area and dynamically adjusts the grid size according to the complexity of the river channel topography; The graphics processor deploys a dynamic adaptive grid in the floodplain. By monitoring the rate of change of water level and the terrain gradient in real time, when the water level fluctuation exceeds the stability threshold or the terrain undulation increases sharply, the grid densification mechanism is automatically triggered to subdivide the local area into high-precision computing units. The graphics processor uses shared memory to pre-store water depth and flow velocity data of adjacent grids; While the graphics processor calculates the current water flow status, the terrain elevation data for the next time time is preloaded simultaneously through a high-speed bus, realizing parallelization of calculation and transmission, and improving the overall efficiency of flood evolution simulation to the level required for real-time scheduling.

[0018] The parameters for real-time correction are measured flow data, which include, but are not limited to, the state vectors of saturated hydraulic conductivity, number of runoff curves, and channel roughness. During correction, the state vectors are updated every 30 minutes by integrating measured flow data to ensure that the root mean square error of the runoff simulation after correction is ≤15%.

[0019] The constraints on the dynamic safety zone water level include a discharge flow rate ≤ 90% of the design flow rate and a real-time water level ≤ the upper limit of the dynamic safety zone. The upper limit of the dynamic safety zone is dynamically calculated based on the monitoring data from the dam stress sensor and the seepage line.

[0020] The construction of reinforcement learning agents specifically includes: Define a multidimensional state space that integrates reservoir status and inundation map. The multidimensional state space integrates the spatial features of time series data of reservoir group water level, inflow, risk entropy value and 30-meter resolution inundation raster map. The design incorporates continuous action space control to regulate the opening of flood discharge facilities. The action space is defined as the continuous adjustment commands for the opening of each reservoir gate. A strategy evaluation model is constructed, which includes two paths: temporal feature extraction and spatial feature extraction. The extracted temporal path is the reservoir state sequence, and the extracted spatial path is the reservoir inundation map features. The agent is trained based on historical disaster scenarios, using a historical flood scenario dataset.

[0021] When simulating the dispatch plan, a social impact assessment is conducted, which specifically includes: generating a real-time heat map by integrating multi-source population distribution data; calculating the number of affected people by overlaying the flood inundation area; assessing disaster losses according to the inundation depth; and setting a population impact threshold to trigger a manual decision-making confirmation process.

[0022] The digital twin platform integrates a 3D flood scene rendered in real time via holographic projection, with the flood depth mapped using a four-color gradient of blue-green-yellow-red, and the flow velocity vector marked with dynamic arrows. It also integrates a gesture interaction system that recognizes swiping, grasping, and double-tapping actions, enabling gate opening adjustments, scene rotation, and scheme confirmation. Furthermore, the platform integrates multi-user terminals that support collaborative operation, automatically recording parameter modifications and risk assessment results as log reports. These log reports include, but are not limited to, timestamps, operator IDs, and risk entropy change curves.

[0023] The rules for implementing the Level III emergency response mechanism are as follows: Level I Response: Open all flood diversion gates, and cascade reservoirs discharge floodwater in the topological order from upstream to downstream, with the discharge flow not exceeding 120% of the downstream safe capacity. Simultaneously, emergency management departments will evacuate the population in high-risk areas. Level II Response: Flood discharge tasks are dynamically allocated according to the risk entropy ratio, with high-risk reservoirs bearing the main responsibility for flood discharge; Level III Response: Close unnecessary floodgates, store water according to the reservoir's regulation capacity, limit flood discharge and prioritize water storage, and control the water storage rate according to the reservoir's regulation capacity.

[0024] Working principle: Utilizing a mesoscale numerical weather prediction system to acquire satellite cloud images, radar echoes, and meteorological data from ground weather stations, a 72-hour forecast scenario is generated, and ensemble precipitation forecast scenario data is output. In the hydrological stage, the ensemble precipitation forecast scenario data is input into the HEC-HMS surface runoff model to generate multiple runoff sequences. Parameters are corrected in real-time every 30 minutes using the root mean square error between measured and simulated flow rates. In the hydraulic stage, the runoff sequences are input into the HEC-RAS river hydraulic analysis model based on a parallel architecture using a graphics processor, simulating the floodplain inundation process and outputting inundation depth, flow velocity, and arrival time, significantly improving the efficiency of flood evolution calculations. To meet the minute-level response requirements in heavy rain scenarios, the system avoids overly conservative scheduling in high-value areas by calculating the real-time risk entropy of the reservoir and constrains the dynamic safety zone water level of the reservoir to prevent dam failure risk. Through hydrological and hydraulic stage simulations, as well as the calculation of the real-time risk entropy and dynamic safety zone water level of the reservoir, the system identifies the center coordinates of areas with water depth > 2m marked in real-time on the reservoir inundation heat distribution map as the maximum risk points, and deduces the optimal scheduling scheme. Based on the deduced scheduling scheme, the system adjusts the flood discharge sequence of the reservoir and issues gate control commands with a three-level emergency response mechanism to the automated control system of the reservoir gates via the industrial Ethernet protocol. The gates are opened and closed according to the corresponding level of the command.

[0025] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0026] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for coordinated flood control scheduling of cascade reservoirs coupled with meteorological-hydrological-hydraulic models, characterized in that, The method for coordinated flood control scheduling of cascade reservoirs includes the following steps: Step 1: Multi-source meteorological data fusion and probabilistic forecast generation. Meteorological data from satellite cloud images, radar echoes, and ground meteorological stations are obtained using a mesoscale numerical weather prediction system to generate forecast scenarios for the next 72 hours and output ensemble rainfall forecast scenario data. This reduces the uncertainty of rainfall input from the source, ensures the improvement of the initial driving accuracy of flood simulation, and provides a reliable input basis for subsequent model coupling. Step 2: Dynamically couple hydrological-hydraulic simulation to break down the disconnect between meteorological, hydrological, and hydraulic models. The hydrological-hydraulic simulation includes hydrological and hydraulic phases. In the hydrological phase, multiple runoff sequences are generated by inputting aggregated rainfall forecast scenario data into the HEC-HMS surface runoff model. Parameters are corrected in real time every 30 minutes by triggering the root mean square error between measured and simulated flow rates. In the hydraulic phase, the runoff sequences are input into the HEC-RAS river hydraulic analysis model based on a parallel architecture of a graphics processor, and the flood inundation process is simulated. The model outputs inundation depth, flow velocity, and arrival time, greatly improving the efficiency of flood evolution calculation and meeting the minute-level response requirements in heavy rain scenarios. Step 3: Risk entropy-driven collaborative optimization decision-making to achieve a multi-objective balance between flood control safety and water resource utilization. By calculating the real-time risk entropy of the reservoir, overly conservative scheduling in high-value areas is avoided, and the dynamic safety domain water level of the reservoir is constrained to prevent dam failure risk. The real-time risk entropy is defined as the logarithmic product of the dam failure probability and the downstream impact index. The downstream impact index is the product of population density and the weight of key facilities. The dynamic safety domain water level is the upper limit of the water level dynamically calculated from real-time engineering safety monitoring data. Step 4: Verification and correction of the digital twin platform. Based on the digital twin platform, the scheduling scheme is simulated. Through the simulation of the hydrological and hydraulic stages, and the calculation of the real-time risk entropy and dynamic safety zone water level of the reservoir, the center coordinates of the areas with water depth > 2m marked in the real-time thermal distribution map of the reservoir are identified as the maximum risk points. The accuracy of high-risk area identification is improved in the simulated scheduling scheme. The optimal scheduling scheme is simulated by combining reinforcement learning agent. The flood discharge sequence of the reservoir is adjusted based on the simulated scheduling scheme. Step 5: Command execution and closed-loop feedback. The gate control command with a three-level emergency response mechanism is sent to the automatic control system of the reservoir gate through the industrial Ethernet protocol. The gate opening and closing control is performed according to the corresponding level of the command, and the water level deviation is monitored in real time. When the actual water level deviates from the predicted value by ≥5% for 1 hour or the single deviation is ≥10%, the model parameter recalibration is automatically triggered.

2. The method for coordinated flood control scheduling of cascade reservoirs according to claim 1, characterized in that: In step 2, the parallel architecture of the graphics processor specifically includes: The graphics processor uses a coarse grid division strategy for the main river channel area and dynamically adjusts the grid size according to the complexity of the river channel topography; The graphics processor deploys a dynamic adaptive grid in the floodplain. By monitoring the rate of change of water level and the terrain gradient in real time, when the water level fluctuation exceeds the stability threshold or the terrain undulation increases sharply, the grid densification mechanism is automatically triggered to subdivide the local area into high-precision computing units. The graphics processor uses shared memory to pre-store water depth and flow velocity data of adjacent grids; While the graphics processor calculates the current water flow status, the terrain elevation data for the next time time is preloaded simultaneously through a high-speed bus, realizing parallelization of calculation and transmission, and improving the overall efficiency of flood evolution simulation to the level required for real-time scheduling.

3. The method for coordinated flood control scheduling of cascade reservoirs according to claim 1, characterized in that: In step 2, the parameters for real-time correction are measured flow data, which include, but are not limited to, the state vectors of saturated hydraulic conductivity, number of runoff curves, and channel roughness. During correction, the state vectors are updated every 30 minutes by integrating the measured flow data to ensure that the root mean square error of the runoff simulation after correction is ≤15%.

4. The method for coordinated flood control scheduling of cascade reservoirs according to claim 1, characterized in that: In step 3, the constraints on the dynamic safety zone water level include flood discharge flow ≤ 90% of design flow and real-time water level ≤ upper limit of dynamic safety zone. The upper limit of dynamic safety zone is dynamically calculated based on the monitoring data of dam stress sensor and seepage line.

5. The method for coordinated flood control scheduling of cascade reservoirs according to claim 1, characterized in that: In step S4, the construction of the reinforcement learning agent specifically includes: Define a multidimensional state space that integrates reservoir status and inundation map. The multidimensional state space integrates the spatial features of time series data of reservoir group water level, inflow, risk entropy value and 30-meter resolution inundation raster map. The design incorporates continuous action space control to regulate the opening of flood discharge facilities. The action space is defined as the continuous adjustment commands for the opening of each reservoir gate. A strategy evaluation model is constructed, which includes two paths: temporal feature extraction and spatial feature extraction. The extracted temporal path is the reservoir state sequence, and the extracted spatial path is the reservoir inundation map features. The agent is trained based on historical disaster scenarios, using a historical flood scenario dataset.

6. The method for coordinated flood control scheduling of cascade reservoirs according to claim 1, characterized in that: In S4, when simulating the scheduling plan, a social impact assessment is conducted. The social impact assessment specifically includes: generating a real-time heat map by integrating multi-source population distribution data; calculating the number of affected people by overlaying the flood inundation range; assessing disaster losses according to the inundation depth; and setting a population impact threshold to trigger a manual decision confirmation process.

7. The method for coordinated flood control scheduling of cascade reservoirs according to claim 1, characterized in that: In step 4, the digital twin platform integrates a 3D flood scene rendered in real time through holographic projection, with the flood depth mapped using a four-color gradient of blue-green-yellow-red, and the flow velocity vector marked with dynamic arrows; the digital twin platform also integrates a gesture interaction system that recognizes swiping, grabbing, and double-clicking actions, enabling gate opening adjustment, scene rotation, and scheme confirmation; the digital twin platform also integrates a multi-user terminal that supports multi-party collaborative operation, with all user terminals automatically recording parameter modifications and risk assessment results as log reports, including but not limited to timestamps, operator IDs, and risk entropy change curves.

8. The method for coordinated flood control scheduling of cascade reservoirs according to claim 1, characterized in that: In S5, the execution rules of the three-level emergency response mechanism are as follows: Level I Response: Open all flood diversion gates, and cascade reservoirs discharge floodwater in the topological order from upstream to downstream, with the discharge flow not exceeding 120% of the downstream safe capacity. Simultaneously, emergency management departments will evacuate the population in high-risk areas. Level II Response: Flood discharge tasks are dynamically allocated according to the risk entropy ratio, with high-risk reservoirs bearing the main responsibility for flood discharge; Level III Response: Close unnecessary floodgates, store water according to the reservoir's regulation capacity, limit flood discharge and prioritize water storage, and control the water storage rate according to the reservoir's regulation capacity.

Images