Method and system for calculating farmland water storage capacity in chaohu basin based on ai runoff-flow coupling simulation

Abstract

This invention provides a method and system for calculating farmland runoff storage capacity in the Chaohu Lake Basin based on AI-based runoff-generator coupling simulation. It relates to the fields of hydrological and water resources simulation and engineering scheduling. The method includes: acquiring underlying surface and hydrological data of farmland and multi-level ditch-pond and wetland storage systems in the Chaohu Lake Basin; constructing a runoff-generator coupling model constrained by both AI algorithms and hydrophysical mechanisms to characterize the runoff process from farmland to ditches and wetlands grid by grid; inputting real-time soil moisture, water level, and flow monitoring data into the runoff-generator coupling model to dynamically simulate the runoff process after heavy rainfall hourly, obtaining a non-constant runoff curve reflecting the peak runoff and the runoff process; and dynamically identifying multiple feature control points with precise spatiotemporal coordinates from the non-constant runoff curve. This invention enables refined simulation of the runoff process in the polder area and reasonable and effective allocation of storage capacity.

Description

Technical Field

[0001] This invention relates to the field of hydrological and water resources simulation and engineering scheduling technology, and in particular to a method and system for calculating the storage capacity of farmland runoff in the Chaohu Lake Basin based on AI runoff-generation coupling simulation. Background Technology

[0002] The Chaohu Lake basin has a flat terrain and a dense river network. It is widely distributed with a multi-level drainage and storage system of farmland, ditches, ponds, and wetlands. It is a key unit for reducing non-point source pollution and ensuring flood control safety in the polder area during the rainstorm season.

[0003] Current calculations of farmland drainage storage capacity mostly rely on traditional lumped hydrological models or semi-distributed models, which generally face the following technical defects: it is difficult to perform grid-by-grid, non-constant, and nonlinear fine-grained dynamic simulations of the drainage process under physical mechanism constraints, and it is also difficult to accurately depict the spatiotemporal evolution of water volume from fields to ditches and wetlands.

[0004] These limitations mean that existing methods can only estimate drainage load using generalized flow processes and empirical coefficients, making it difficult to capture key spatiotemporal characteristics such as peak drainage, inflection points, and velocity gradients after heavy rainfall. This leads to a mismatch between the allocation of storage capacity and the actual drainage process. For example, in the actual scheduling of polder areas in the Chaohu Lake basin, insufficient or redundant storage capacity is common. This makes it difficult to effectively intercept peak flows and ensure flood control safety in polder areas during periods of heavy rainfall, and also makes it difficult to stably retain nitrogen and phosphorus nutrients during the critical drainage period, significantly reducing the effectiveness of non-point source pollution reduction. Moreover, the models lack a closed-loop correction mechanism driven by real-time monitoring data. The deviation in drainage simulation accumulates over time, making it difficult to dynamically optimize the timing of gate and pump opening and closing and the water storage depth. This results in low operating efficiency of the storage system, making it difficult to adapt to the needs of refined water environment management and coordinated flood control and drainage control in the Chaohu Lake basin. Summary of the Invention

[0005] This invention provides a method and system for calculating the storage capacity of farmland drainage in the Chaohu Lake Basin based on AI runoff-generation coupling simulation, which can realize refined simulation of drainage process in polder areas and reasonable and efficient allocation of storage capacity.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for calculating the storage capacity of farmland runoff in the Chaohu Lake basin based on AI runoff-generation coupling simulation, the method comprising: Acquire underlying surface and hydrological data of farmland and multi-level ditch and wetland regulation and storage system in the polder area of ​​Chaohu Lake Basin, and construct a runoff-runoff coupling model constrained by both AI algorithm and hydrophysical mechanism to characterize the water receding process from field to ditch and wetland on a grid-by-grid basis. Real-time soil moisture, water level and flow monitoring data are input into the runoff-runoff coupling model to dynamically simulate the hourly drainage process after heavy rain, and obtain a non-constant drainage curve that reflects the peak of farmland runoff and the drainage process. Dynamically identify multiple feature control points with precise spatiotemporal coordinates from non-constant receding water curves; Using the aforementioned characteristic control points as nodes, the phase space of the receding water time series is reconstructed to construct a hydrodynamic manifold characterizing the nonlinear evolution path of the receding water process. Information geometry analysis is performed on the hydrodynamic manifold to extract the norm of its geodesic curvature tensor and the spectral entropy of the Fisher information matrix, and then fused to obtain the drainage morphology coefficient that characterizes the nonlinear evolution of farmland drainage. Based on the non-constant receding water curve and receding water morphology coefficient, combined with the real-time monitoring of the current water storage status and physical regulation capacity of ponds and wetlands at all levels, the regulation storage capacity is dynamically allocated in time periods to generate a regulation operation plan that includes the opening and closing sequence of gate pumps and water storage depth. The regulation and storage operation plan is fed back to the runoff-generated runoff coupling model for verification. The deviation of the simulated receding water peak is corrected online by real-time monitoring data. After closed-loop correction, an optimized regulation and storage capacity configuration plan that meets the requirements of flood control and non-point source pollution reduction in the polder area is obtained.

[0007] Secondly, a system for calculating the storage capacity of farmland runoff in the Chaohu Lake basin based on AI-based runoff-generation coupling simulation includes: The data acquisition module is used to acquire underlying surface and hydrological data of farmland and multi-level ditch and wetland regulation and storage systems in the Chaohu Lake basin. The coupled model construction module is used to construct a runoff-runoff coupling model constrained by both AI algorithms and hydrophysical mechanisms, which is used to characterize the drainage process from field plots to ditches and wetlands on a grid-by-grid basis. The drainage process simulation module is used to input real-time soil moisture, water level and flow monitoring data into the runoff-gene coupling model to perform hourly dynamic simulation of the drainage process after heavy rain, and obtain a non-constant drainage curve that reflects the peak of farmland runoff and the drainage process. The feature point recognition module is used to dynamically identify multiple feature control points with precise spatiotemporal coordinates from non-constant receding water curves. The manifold construction and analysis module is used to reconstruct the phase space of the drainage time series with the feature control points as nodes, and construct a hydrodynamic manifold that represents the nonlinear evolution path of the drainage process; perform information geometry analysis on the hydrodynamic manifold, extract the norm of its geodesic curvature tensor and the spectral entropy of the Fisher information matrix, and fuse them to obtain the drainage morphology coefficients that represent the nonlinear evolution characteristics of farmland drainage. The regulation and storage scheme generation module is used to dynamically allocate the regulation and storage capacity in time periods based on the non-constant water receding curve and water receding morphology coefficient, combined with the real-time monitoring of the current water storage status and physical regulation and storage capacity of ponds and wetlands at all levels, and generate a regulation and storage operation scheme that includes the gate pump opening and closing sequence and water storage depth. The closed-loop correction module is used to feed back the regulation and storage operation plan to the runoff-generated runoff coupling model for verification. It corrects the deviation of the simulated receding water peak online through real-time monitoring data. After closed-loop correction, an optimized regulation and storage capacity configuration plan that meets the requirements of flood control and non-point source pollution reduction in the polder area is obtained.

[0008] The above-described solution of the present invention has at least the following beneficial effects: Because it employs a runoff-generation coupled model with dual constraints of AI algorithms and hydrophysical mechanisms, and simulates the recession process in a refined grid-by-grid manner, it overcomes the technical problem that traditional models cannot accurately depict the non-constant and nonlinear spatiotemporal evolution of recession, thus effectively improving the simulation accuracy of the recession process. Because it uses feature control point identification, phase space reconstruction, and information geometric analysis techniques to quantify the nonlinear characteristics of recession, it overcomes the technical problem that existing methods cannot accurately capture key spatiotemporal characteristics of recession and have large deviations in recession load estimation, thus providing a reasonable basis for the dynamic allocation of storage capacity. Because it constructs a closed-loop mechanism of simulation, scheduling, feedback, and correction, and combines real-time monitoring data to correct model parameters online, it overcomes the technical problems of mismatch between storage schemes and actual recession processes and the lack of dynamic optimization mechanisms, thus achieving a reasonable and efficient allocation of storage capacity, taking into account both flood control safety in the Chaohu Lake basin and reduction of non-point source pollution in farmland, and improving the coordinated regulation and control capabilities of the water system in the polder area. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the method for calculating the storage capacity of farmland runoff in the Chaohu Lake Basin based on AI runoff-generation coupling simulation, as provided in an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of a system for calculating the storage capacity of farmland runoff in the Chaohu Lake Basin based on AI runoff-generation coupling simulation, provided by an embodiment of the present invention. Detailed Implementation

[0011] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0012] like Figure 1As shown, embodiments of the present invention propose a method for calculating the storage capacity of farmland runoff in the Chaohu Lake basin based on AI runoff-generation coupling simulation. The method includes the following steps: Step 1: Obtain the underlying surface and hydrological data of farmland and multi-level ditch and wetland regulation and storage system in the polder area of ​​Chaohu Lake Basin, and construct a runoff-runoff coupling model constrained by both AI algorithm and hydrophysical mechanism to characterize the water receding process from field to ditch and wetland grid by grid. Step 2: Input the real-time soil moisture, water level and flow monitoring data into the runoff-runoff coupling model to perform hourly dynamic simulation of the drainage process after the rainstorm, and obtain a non-constant drainage curve that reflects the peak of farmland runoff and the drainage process. Step 3: Dynamically identify multiple feature control points with precise spatiotemporal coordinates from the non-constant receding water curve; Step 4: Using the feature control points as nodes, the phase space of the receding water time series is reconstructed to construct a hydrodynamic manifold that characterizes the nonlinear evolution path of the receding water process. Step 5: Perform information geometry analysis on the hydrodynamic manifold, extract the norm of its geodesic curvature tensor and the spectral entropy of the Fisher information matrix, and fuse them to obtain the drainage morphology coefficient that characterizes the nonlinear evolution of farmland drainage. Step 6: Based on the non-constant receding water curve and receding water morphology coefficient, combined with the real-time monitoring of the current water storage status and physical regulation capacity of ponds and wetlands at all levels, dynamically allocate the regulation storage capacity in time periods to generate a regulation operation plan that includes the gate pump opening and closing sequence and water storage depth. Step 7: Feedback the regulation and storage operation plan to the runoff-generated runoff coupling model for verification. The deviation of the simulated receding water peak is corrected online by real-time monitoring data. After closed-loop correction, an optimized regulation and storage capacity configuration plan that meets the requirements of flood control and non-point source pollution reduction in the polder area is obtained.

[0013] In this embodiment of the invention, by constructing a runoff-generation coupling model with dual constraints of AI algorithms and hydrophysical mechanisms, a refined dynamic simulation of the farmland drainage process in the polder area of ​​the Chaohu Lake Basin is achieved, which is grid-by-grid, non-constant, and nonlinear. This model can accurately identify key spatiotemporal characteristics during the drainage process, construct hydrodynamic manifolds, and quantify drainage morphology coefficients, providing a reliable basis for the allocation of storage capacity. This invention can dynamically generate gate and pump opening and closing schemes and water storage scheduling based on the real-time status of storage facilities, and continuously optimize simulation results through closed-loop feedback correction, effectively improving the simulation accuracy of the drainage process and the rationality of storage capacity allocation. While ensuring flood control safety in the polder area, it strengthens the reduction effect of non-point source pollution in farmland and improves the overall operational efficiency and collaborative management level of the multi-level ditch and wetland storage system.

[0014] In a preferred embodiment of the present invention, step 1 above may include: Step 1.1: Collect multi-source data on farmland and the multi-level ditch and wetland regulation system in the polder area of ​​Chaohu Lake Basin, including topographic elevation, soil texture and stratification parameters, crop type and planting system, layout and cross-sectional geometry of ditches and wetlands, historical rainfall and water level-discharge processes. Process the multi-source data to obtain a standardized basic dataset, specifically including: farmland and the multi-level ditch and wetland regulation system within the polder area of ​​Chaohu Lake Basin: the farmland mainly consists of rice, wheat and other major crops, distributed in the flat areas within the polder area, and the fields are interconnected by irrigation ditches, which are the main source of farmland runoff; the multi-level ditch and wetland regulation system is divided into three levels of ditches: irrigation ditches, tributaries, and main ditches, according to hydraulic connectivity, as well as ecological ponds distributed at the nodes of the ditches, which have both water storage and purification functions, and natural wetlands located along the edge of the polder area, which can receive water from the ditches. In this invention, the fields, irrigation ditches, and ecological ponds are considered as follows: Ponds and wetlands form the core of the polder area's four-tiered drainage network, while tributary ditches, main ditches, and secondary ditches supplement the drainage system, creating a complete drainage path that connects fields, ditches, ponds, and wetlands at each level, providing layer-by-layer drainage. Data on topography, soil texture and stratification parameters, crop type and planting system, layout and cross-sectional geometry of ditches, ponds, and wetlands, historical rainfall, and historical water level and flow rate were collected from multiple sources, including on-site monitoring, field surveys, statistical data, remote sensing images, and hydrological station observations. The collected multi-source heterogeneous data underwent standardized processing, including format conversion, unit conversion, identification and removal of obvious anomalies, appropriate interpolation of missing data for consecutive time periods, and precise matching of data from different sources in time and space. This resulted in a standardized basic dataset with consistent spatiotemporal references, reliable numerical values, and direct applicability for model construction and calculation.

[0015] Step 1.2: Based on the standardized basic dataset, the target area is divided into grids according to a four-level drainage network of fields, irrigation ditches, ecological ponds, and wetlands. Each grid is assigned corresponding physical attribute parameters, including roughness coefficient, saturated hydraulic conductivity, field capacity, and initial soil moisture content, thereby constructing a distributed hydrophysical framework. Specifically, this includes: using the standardized basic dataset as support, and combining the spatial distribution characteristics and hydraulic connectivity of the four-level drainage network of fields, irrigation ditches, ecological ponds, and wetlands in the Chaohu Lake basin, a refined spatial grid is performed on the target computational area. The grid division accuracy closely matches the underlying soil of the polder area. To mitigate surface differences, each grid is designed to independently and accurately reflect the hydrological characteristics of the underlying surface at its corresponding location, avoiding the loss of hydrological details due to excessively large grids and the increase in computational redundancy due to excessively small grids. After grid division, each computational grid is assigned key physical property parameters such as roughness coefficient, saturated hydraulic conductivity, field capacity, and initial soil moisture content based on the actual land use type, soil conditions, ditch morphology, and wetland structure of its location. This process constructs a distributed hydrophysical framework that can truly reflect the underlying surface conditions, hydraulic transmission paths, and distribution characteristics of regulation and storage facilities in the Chaohu Lake basin.

[0016] The roughness coefficient is a parameter characterizing the resistance of the underlying surface (such as field soil, ditch walls, pond bottoms, and slopes) to water flow within a grid. It is primarily determined through field measurements, review of relevant research findings on similar underlying surfaces in the Chaohu Lake basin, and hydrological standards. Different underlying surfaces have different values. The roughness coefficient for field soil is determined in conjunction with crop type; the roughness coefficient for ditches is determined in conjunction with ditch material and the smoothness of the inner wall; and the roughness coefficient for wetlands is determined in conjunction with wetland vegetation cover. Saturated hydraulic conductivity refers to the rate at which water permeates the soil when the soil pores are completely saturated with water. It is mainly determined through indoor permeability tests on soil samples from different areas and soil layers in the Chaohu Lake basin, while also considering soil texture, etc. Parameters such as porosity are corrected to ensure they match the actual soil permeability characteristics within the grid. Field water holding capacity refers to the maximum capillary suspended water that the soil can retain, i.e., the water content that the soil can retain after fully absorbing water and removing gravitational water. This is determined by collecting soil samples from different soil layers within the grid, conducting indoor drying and weighing tests, and calibrating with reference to soil type zoning data of the Chaohu Lake Basin polder area. Initial soil moisture content refers to the actual soil moisture content within the grid at the simulated initial moment. Real-time monitoring data is obtained by deploying soil moisture monitoring stations at different locations in the polder area, and spatial interpolation is performed based on the grid spatial location to ensure that the initial soil moisture content of each grid is consistent with the actual field soil moisture state.

[0017] Step 1.3 involves embedding the AI ​​algorithm into a distributed hydrophysical framework, training the AI ​​algorithm using historical receding water process data, constraining the output boundary of the AI ​​algorithm through physical mechanisms, and simultaneously optimizing parameters in the physical framework that are difficult to measure directly using the AI ​​algorithm. After parameter calibration and verification, a dual-constraint runoff-generation coupling model is obtained. Specifically, this includes embedding a data-driven model based on Long Short-Term Memory (LSTM) networks and incorporating physical information as the core of the AI ​​algorithm into the constructed distributed hydrophysical framework to build a complete runoff-generation coupling model. This model architecture originates from traditional LSTM networks and is specifically improved to address the nonlinearity, time-varying nature, and multi-path confluence characteristics of the receding water process in the Chaohu Lake basin. The model is divided into five interconnected functional modules: an input layer, a physical constraint layer, an LSTM core computation layer, a parameter optimization layer, and an output layer. Each module has a clear division of labor and works collaboratively to form a complete coupling model architecture.

[0018] The core function of the input layer is to receive the standardized basic dataset transmitted by the distributed hydrophysical framework, including topographic, soil, crop, ditch and wetland parameters, and historical drainage data. After normalizing the input data, it is synchronously transmitted to the physical constraint layer and the LSTM core computing layer, providing standardized input for model training and computation. The physical constraint layer is the core constraint module of the coupled model. It incorporates the basic laws of hydrological cycle in the Chaohu Lake Basin, the inherent characteristics of water flow transmission, and the specific hydrophysical boundary requirements of the polder area, such as the maximum water storage depth of ditches and wetlands, field runoff threshold, water conveyance capacity limit of ditches, and soil saturation moisture content threshold. On the one hand, it performs physical rationality verification on the input data and removes abnormal data that exceeds the actual hydrological conditions. On the other hand, it generates constraint parameters and transmits them to the LSTM core computing layer and parameter optimization layer to limit the boundary range and trend of the model output, avoiding simulation results that deviate from physical laws from the pure data-driven model.

[0019] The LSTM core computation layer is the core computational module of the model, responsible for capturing long-term temporal dependencies in the drainage process. It receives standardized data from the input layer and constraint parameters from the physical constraint layer. Through the synergistic action of forget gates, input gates, and output gates, it simulates the time-series processes of field runoff generation, canal runoff, and reservoir / wetland regulation, outputting preliminary drainage process simulation data (including runoff and runoff for each grid). Simultaneously, it transmits the simulation data to the parameter optimization layer. The parameter optimization layer connects the LSTM core computation layer with the distributed hydrophysical framework, receiving the preliminary simulation data from the LSTM core computation layer and the constraints from the physical constraint layer. The model incorporates parameters, including hydrological parameters that are difficult to measure directly (such as roughness, saturated hydraulic conductivity, and field capacity) fed back by the distributed hydrophysical framework. Utilizing the model's adaptive learning capabilities, these difficult-to-measure parameters are dynamically optimized and corrected. The corrected parameters are then synchronously fed back to the LSTM core computing layer and the distributed hydrophysical framework, achieving closed-loop parameter optimization. The output layer receives the simulated data from the LSTM core computing layer after parameter correction, performs inverse normalization on the data, and outputs data on the receding process that conforms to actual hydrological units and physical laws. This output data is then fed back to the model training stage for error calculation and parameter adjustment.

[0020] During the model training phase, real farmland drainage data from the past 10 years in the Chaohu Lake basin under different rainfall intensities, crop growth stages, and water storage states (empty, half-full, and full) of ditches and wetlands were used as training samples. Simultaneously, three years of independent rainstorm drainage data were selected as validation samples, and one year of data as test samples for supervised training and iterative learning of the model. In the initial training phase, the multi-source data of the input layer were normalized, and initial model parameters, such as learning rate, number of iterations, physical constraint weights, and the number of neurons in the LSTM network hidden layer, were set. Standardized data was then input into the model, and preliminary simulation results were output through collaborative computation of various modules. After each training cycle, the simulated drainage data of the output layer was compared with the measured drainage data, and the simulation error (including peak flow deviation, drainage duration deviation, and time-period flow deviation) was calculated. Using the backpropagation algorithm, the parameter correction coefficients of the parameter optimization layer, the network weights of the LSTM core computation layer, and the constraint weights of the physical constraint layer were adjusted sequentially, while simultaneously updating the difficult-to-measure parameters in the distributed hydrophysical framework.

[0021] During training, the focus was on practical application scenarios in the Chaohu Lake basin, where concentrated runoff from fields, rapid flow velocity in ditches, and large fluctuations in the storage capacity of ponds and wetlands occur after heavy rains. The boundary conditions of the physical constraint layer and the temporal capture capabilities of the LSTM core computing layer were optimized to avoid situations that do not conform to physical laws, such as negative runoff, drainage velocities exceeding the carrying capacity of ditches, and water depths in ponds and wetlands exceeding actual limits. Through parameter calibration, accuracy verification, and error analysis of multiple independent rainfall and drainage events, the number of model training iterations, learning rate, and physical constraint weights were repeatedly optimized until the model performed well on the test samples. The simulation deviation is controlled within a preset threshold range, ultimately forming a runoff-generation coupled model that is dually constrained by AI algorithms and hydrophysical mechanisms and applicable to the multi-level drainage system of polder fields, ditches, ponds, and wetlands in the Chaohu Lake Basin. This model retains the ability of traditional distributed hydrological models to clearly characterize physical processes, while also possessing the advantages of improved LSTM models in strongly fitting and rapidly predicting complex nonlinear drainage processes. The modules work together to accurately adapt to the actual scenario of the Chaohu Lake Basin polder area, which has a flat terrain, dense river network, and dispersed drainage paths, taking into account both the rationality of physical processes and the fitting accuracy of complex drainage processes.

[0022] Step 1.4: The runoff-generated runoff coupling model is used to simulate the hourly drainage process after a rainstorm for each grid of the four-level drainage network. The runoff and runoff of each grid unit in the time series are output to form a drainage process data volume that describes the process grid by grid. Specifically, the runoff-generated runoff coupling model with dual constraints of AI algorithm and hydrophysical mechanism, which has been constructed, calibrated and verified, is used to carry out hourly continuous dynamic simulation calculations of the entire process of farmland drainage after a rainstorm for each grid unit in the four-level drainage network of fields, ditches, ecological ponds and wetlands. The simulation process strictly follows the hydraulic connectivity of the four-level drainage network and progresses step by step with a uniform 1-hour time step to ensure that the simulation process is completely synchronized with the actual drainage time sequence.

[0023] When the simulation starts, the physical property parameters of each grid in the constructed distributed hydrophysical framework, including roughness coefficient, saturated hydraulic conductivity, field capacity, and initial soil moisture content, as well as processed standardized basic data, including topography, soil, and crops, are used as the initial input conditions for the model. Simultaneously, the initial hydrological conditions after the rainstorm, including initial water level and initial soil moisture, are loaded to determine the start and end times of the simulation. The end time is set when the drainage flow of the grid unit tends to stabilize and reaches the normal hydrological state of the polder area. During the hourly simulation process, each time interval is incremented... The step size, combined with the simulation results of the previous time step, including the runoff, runoff, water level, soil moisture content of each grid, and real-time supplemented rainfall attenuation data, calculates the field runoff (combining soil water storage capacity and crop interception to determine whether the field is producing runoff and the size of the runoff), ditch runoff (calculating water flow transmission based on ditch roughness, cross-sectional size, and upstream and downstream water level difference), and ecological pond and wetland regulation capacity (combining the current water storage status and maximum water storage capacity of ponds and wetlands to calculate regulation capacity and discharge capacity), and updates the hydrological status parameters of each grid synchronously.

[0024] After each time step of simulation is completed, the flow rate and runoff data of the grid cell at the current moment are automatically recorded and output to ensure that the time series data of each grid cell is continuous, complete, and without omissions or interruptions. After all time step simulations are completed, the hourly flow rate and runoff time series results of all grid cells in the four-level drainage network are accurately integrated and summarized according to the grid spatial coordinates to confirm the water transmission relationship between each grid. The complete path of the runoff from the field, through the tiered flow of the irrigation canals, and the layered regulation of the ecological ponds and wetlands is clearly presented. Finally, a grid-by-grid drainage process data body is formed that can completely and meticulously depict the tiered water transmission process from the runoff from the field, the flow of the irrigation canals, to the regulation of the ponds and wetlands, with accurate spatiotemporal information and reliable continuous data.

[0025] In a preferred embodiment of the present invention, step 2 above may include: Step 2.1: Acquire discrete data collected by the real-time monitoring network deployed at various monitoring sections of fields, ditches, ponds, and wetlands, including soil moisture, water level, and flow data at each monitoring point; perform outlier removal and missing value interpolation on the discrete data, and perform spatial interpolation based on the spatial coordinates of each monitoring point to convert the processed discrete data into a continuous field that matches the grid space of the runoff-runoff coupling model, generating a dynamic input field. Specifically, this includes: relying on the real-time monitoring network pre-deployed at key locations in the four-level drainage network of the Chaohu Lake basin polder area, this real-time monitoring network, according to the distribution characteristics of fields, irrigation ditches, ponds, and wetlands, sets up soil moisture monitoring points in each field, and sets up monitoring points at the inlet and outlet of each irrigation ditch, pond, and wetland. The water level and flow monitoring sections collect discrete data generated by each monitoring point in real time. Among them, the field monitoring points mainly collect soil moisture data at different soil depths, the ditch monitoring sections mainly collect cross-sectional water level and instantaneous flow data, and the pond and wetland monitoring sections mainly collect water storage level and inflow and outflow data. The collected discrete monitoring data are systematically preprocessed. First, through time series trend analysis combined with the reasonable range of hydrological parameters in the Chaohu Lake Basin, abnormal data that exceeds the normal fluctuation range are identified and eliminated, including sudden data caused by monitoring equipment failure and extreme data that exceeds the actual hydrological conditions. Then, short-term missing data and intermittent missing data that occur during the monitoring process are reasonably interpolated to ensure the temporal continuity and numerical integrity of each set of monitoring data.

[0026] Based on the calibrated and accurate spatial coordinates of each monitoring point, a spatial interpolation method adapted to the runoff-runoff coupling model grid is adopted. First, the core parameters such as the grid scale and spatial reference of the runoff-runoff coupling model are confirmed. Then, based on the discrete data of each monitoring point and combined with the underlying surface characteristics around the monitoring point, the discrete data is extended to all grids of the model. The interpolation data corresponding to each model grid is calculated one by one to ensure that the interpolation process fits the actual conditions of the topography, soil, and hydrology of the polder area in the Chaohu Lake Basin. Finally, the discrete data scattered at each monitoring point is converted into a continuous data field that is completely consistent with the spatial scale and spatial reference of the runoff-runoff coupling model grid. This ensures that each grid data in the continuous data field can accurately correspond to the actual position of the model grid, and finally generates a dynamic input field that can reflect the initial state of the receding water in the polder area of ​​the Chaohu Lake Basin in real time and accurately.

[0027] Step 2.2: Load the dynamic input field as initial and boundary conditions into the constructed runoff-generation coupling model to update the initial soil moisture content and the initial water level of ditches and wetlands in each grid of the model. Specifically, this includes: loading the dynamic input field as the initial and boundary conditions of the runoff-generation coupling model, where the initial conditions mainly refer to the initial soil moisture content of each grid cell and the initial water level of each level of ditches and wetlands at the start of the simulation, and the boundary conditions mainly refer to the water exchange boundary of the grid cell, the upper limit of water storage and the water conveyance boundary of the ditches and wetlands during the simulation. This is loaded into the completed and parameter-calibrated runoff-generation coupling model, ensuring that the spatial reference and data format of the dynamic input field are completely matched with the model during the loading process; utilizing the dynamic input field... The model uses continuous soil moisture data for all model grids to update the initial soil moisture content of each grid cell. Based on the type of field, ditch, pond, or wetland where the grid is located, the model matches the correspondence between soil moisture data and initial soil moisture content to ensure that the initial moisture state of each grid is consistent with the actual field soil moisture in the Chaohu Lake basin. At the same time, the model uses continuous water level data from monitoring sections of ditches, ponds, and wetlands at all levels in the dynamic input field to update the starting water level of each level of irrigation ditches, ecological ponds, and wetlands. This determines the initial water storage depth and available storage space of ditches, ponds, and wetlands at different levels, accurately defines the initial boundary conditions of the model simulation, and provides accurate and realistic initial parameters to support the hourly dynamic simulation and iterative calculation of the water receding process after heavy rain.

[0028] Step 2.3: Drive the updated runoff-generated runoff coupling model, and combine it with the real-time rainfall data to perform hourly iterative calculations on the post-rainstorm drainage process. Output the runoff and runoff of each grid unit hourly to form an hourly updated dynamic data sequence of water balance. Specifically, this includes: starting and driving the runoff-generated runoff coupling model updated with initial and boundary conditions, and synchronously accessing the rainfall data transmitted from the real-time rainfall monitoring network of the Chaohu Lake Basin polder area. This real-time rainfall monitoring network consists of rainfall monitoring stations deployed in different areas of the polder area, covering all key areas such as fields, ditches, ponds, and wetlands. The monitoring stations are deployed according to the principle of uniform distribution, and the transmitted rainfall data is updated hourly, covering key information such as rainfall intensity and duration in different areas of the polder area. The data is also verified in real time to ensure accuracy and reliability.

[0029] The hydraulic connectivity of the four-level runoff network follows a hierarchical confluence logic of fields, irrigation ditches, ecological ponds, and wetlands. Fields serve as the runoff source, with their outlets connected to irrigation ditches. Runoff from these fields flows into the irrigation ditches, then sequentially into ecological ponds. After receiving runoff from the ditches, excess water from the ecological ponds flows into the wetlands. The wetlands regulate the incoming water before finally discharging or retaining it, forming a complete hydraulic transmission relationship of hierarchical connectivity and layered regulation. The runoff-runoff coupling model, combined with real-time rainfall data and the determined physical property parameters of each grid cell in the model, including roughness coefficient, saturated hydraulic conductivity, and field capacity, is used to assess the impact of heavy rainfall on farmland in the polder area of ​​the Chaohu Lake basin after a rainstorm. The entire process of water receding is carried out through hourly iterative calculations: proceeding at a uniform time step, each iteration takes the real-time rainfall data of the current period as input, combined with the hydrological state parameters of each grid unit obtained from the previous iteration, and follows the hydraulic connectivity of the four-level water receding network. First, the runoff of each field grid is calculated, then the runoff of the ditch grid is calculated based on the hydraulic characteristics of the ditch, and the storage capacity and discharge of ponds and wetlands are calculated in combination with their regulation capacity. The hydrological state parameters such as water level and soil moisture content of each grid unit are updated synchronously. After completing one iteration, the runoff and runoff data of each grid for that period are recorded and output, and then the iteration calculation for the next period begins, and so on.

[0030] The iterative calculation process ensures that water transmission conforms to actual hydraulic laws. After each hour of iterative calculation, the production and runoff data of each grid unit in the four-level drainage network are output in real time. The real-time water level, soil moisture content and other hydrological parameters of each grid unit are recorded simultaneously. Iterative calculation is continued until the drainage flow of each grid unit tends to stabilize and reaches the normal hydrological state of the Chaohu Lake basin polder area. After that, the production and runoff data of all grids in the iterative period are sorted, summarized and verified in chronological order. Abnormal simulation data that occurs during the iteration process are removed. Finally, a dynamic water balance data sequence that is updated hourly, has continuous time sequence, is reliable and covers all grid units is formed.

[0031] Step 2.4: Extract flow and water level data from the field outlet section, the confluence section of the dounong ditch, and the inlet section of the ecological pond from the dynamic water balance data sequence. After wavelet denoising and cubic spline interpolation, generate the non-constant drainage curve. Specifically, this includes: accurately extracting relevant data from three key sections from the generated hourly dynamic water balance data sequence, based on the layout characteristics of the four-level drainage network in the Chaohu Lake basin polder area. Specifically, hourly flow and water level data are extracted from the field outlet section to reflect the drainage status after runoff generation in the field; the dounong ditch... The hourly flow rate and water level data were extracted from the confluence section to reflect the connection status of the ditch confluence. The hourly flow rate and water level data were also extracted from the inlet section of the ecological pond to reflect the initial state of the ditch confluence entering the storage facility. For the flow rate and water level data of each type of key section, targeted data optimization processing was carried out. First, wavelet denoising method was used to remove monitoring noise and interference signals caused by monitoring equipment interference and environmental factors, and retain the true information of the water receding process in the data to ensure the authenticity and smoothness of the processed data.

[0032] The cubic spline interpolation method works as follows: using denoised data points as nodes, a smooth cubic polynomial curve is constructed between adjacent nodes. This cubic polynomial curve is a continuous curve composed of cubic polynomials, with a smooth overall curve and no obvious inflection points, which can closely match the natural changing trends of flow and water level during the receding process, without any abrupt fluctuations. The first derivative reflects the rate of change of the curve at the corresponding node, that is, the instantaneous change rate of flow or water level during the receding process, ensuring that the rate of change of adjacent curves at the nodes is consistent and avoiding abrupt rate changes. The second derivative reflects the curvature of the curve at the corresponding node, ensuring that the curvature trend of adjacent curves at the nodes is consistent and without sudden concave or convex turns. By filling in the small discontinuities in the data and densifying the sparse data points, the flow and water level process data form a continuous and smooth time-series curve, clearly presenting the changing trends of flow and water level during the receding process.

[0033] The non-constant receding water curve is a time-series curve characterizing the entire process of receding water after a rainstorm in the polder area of ​​Chaohu Lake Basin. With time as the horizontal axis and flow rate or water level as the vertical axis, it can accurately present the dynamic changes in the receding water process. The curve starts at the peak of runoff after the rainstorm ends, and then the flow rate or water level gradually decreases over time. However, the rate of decrease is not constant and will show non-linear fluctuations due to changes in runoff from fields, differences in ditch confluence, and the regulation and storage effects of ponds and wetlands. It can clearly reflect the changes in the intensity of receding water at different times, accurately capture key features such as the peak receding water level and the inflection point of decay, and completely record the entire time-series process of receding water from field runoff, ditch confluence to the regulation and storage effects of ponds and wetlands. Finally, it generates a receding water curve that can accurately characterize the non-constant characteristics of the receding water process after a rainstorm in the polder area of ​​Chaohu Lake Basin, and the data is continuous and the trend is clear.

[0034] In a preferred embodiment of the present invention, step 3 above may include: Step 3.1 involves numerically differentiating the generated non-constant receding water curve to calculate its derivatives along the time axis, thereby obtaining the velocity gradient curve and water level change rate curve. Specifically, this includes numerically differentiating the generated non-constant receding water curve of the Chaohu Lake basin polder area after a rainstorm. The core step is calculating the first derivative of the receding water curve along the time axis. Since subsequent steps identify various characteristic points such as the peak instantaneous flow at the field outlet section and the water level inflection point at the confluence of the dikes, only the first derivative is needed; higher derivatives are not required. The velocity gradient curve and water level change rate curve can be further obtained by calculating the first derivative. The specific calculation process uses the first-order forward difference method. This method approximates the instantaneous change rate of the parameters by using the ratio of the difference in hydrological parameters between two adjacent moments to the time difference. The expression for the water level change rate is: The expression for the velocity gradient is: In the formula, This represents the instantaneous rate of change of water level over time, i.e., the rate of change of water level, expressed in units of length / time. The instantaneous rate of change of flow velocity over time, i.e., the flow velocity gradient, is expressed in units of velocity / time.

[0035] in, Indicates the first At that moment, Indicates the relationship with the first The first time interval adjacent to the first The number of times is given by n, where n represents the total number of times along the non-constant receding curve and the time interval between two adjacent times. The time step is completely consistent with the hourly iterative simulation of the receding water process after the rainstorm in step 2.3; Indicates the first The water level value corresponding to the non-constant receding water curve at a given time is derived from the key section water level process data after wavelet denoising and cubic spline interpolation. Indicates the first The water level value corresponding to the non-constant receding water curve at a given time, and Consistent source; Indicates the first The velocity value corresponding to the non-constant receding water curve at any given time is obtained by converting the flow process data of the key section in step 2.4 into the cross-sectional geometric parameters; Indicates the first The velocity value corresponding to the non-constant receding water curve at a given time, and The conversion method is consistent. During the calculation, the time axis of the non-constant receding water curve is used as the reference. The water level and flow velocity data corresponding to each moment are extracted in turn. According to the above calculation method, the water level change rate and flow velocity gradient data of each moment are calculated one by one. After the calculation is completed, the water level change rate data of all moments are sorted in time order and a time series curve is plotted to obtain the water level change rate curve. At the same time, the flow velocity gradient data of all moments are sorted in time order and a time series curve is plotted to finally obtain the flow velocity gradient curve and water level change rate curve of the receding water curve.

[0036] Step 3.2: Based on the velocity gradient curve and water level change rate curve, and combined with preset physical feature thresholds, identify the instantaneous flow peak point at the field outlet section, the water level inflection point at the confluence of ditches, the velocity gradient extreme point at the ecological pond inlet section, and the water level change rate inflection point on the wetland internal water level monitoring vertical line. Specifically, based on the obtained velocity gradient curve and water level change rate curve, and combined with preset physical feature thresholds, identify four types of feature points corresponding to the field outlet section, the ditches confluence section, the ecological pond inlet section, and the wetland internal water level monitoring vertical line. The preset physical feature thresholds are determined based on the actual hydrological conditions, topographic features, and historical monitoring data of the four-level drainage network in the Chaohu Lake basin polder area. They are fixed after multiple verifications and optimizations, and specifically cover the flow peak threshold, water level inflection point threshold, velocity gradient extreme value threshold, and water level change rate inflection point threshold. Each threshold corresponds to a precise numerical range and is adapted to the hydrological characteristics of different sections.

[0037] Specific identification process: For the field outlet section, the velocity gradient curve data corresponding to this section is accurately extracted. Combined with the velocity-flow conversion relationship of this section, the velocity data is converted into flow data. When the flow rate at a certain moment reaches the preset peak flow threshold, and the flow rate in the hour before and the hour after that moment shows a clear trend of first increasing and then decreasing, and the flow difference reaches the set fluctuation standard, the point corresponding to that moment is identified as the instantaneous peak flow point of the field outlet section. This point can reflect the maximum output state of the field runoff. For the confluence section of the irrigation canals, the water level change rate curve data corresponding to this section is extracted. When the water level change rate value at a certain moment reaches the set water level inflection point threshold, and the water level change rate in the hour before and the hour after that moment changes significantly (i.e., from positive to negative or from negative to positive), and the change amplitude meets the preset requirements, the point corresponding to that moment is identified as the water level inflection point of the confluence section of the irrigation canals. This point can reflect the water volume connection and turning point of the canal confluence.

[0038] For the inlet section of the ecological pond, the velocity gradient curve data corresponding to the section is extracted. When the velocity gradient value at a certain moment reaches the set extreme threshold, which includes two intervals: a maximum threshold and a minimum threshold, and the velocity gradient one hour before and one hour after that moment shows an extreme turning trend (i.e., from increasing to decreasing or from decreasing to increasing), and the gradient difference reaches the set standard, the point corresponding to that moment is identified as the extreme point of the velocity gradient at the inlet section of the ecological pond. This point can reflect the abrupt change in velocity of the ditch flowing into the ecological pond. For the water level monitoring vertical line inside the wetland, the water level change rate curve data corresponding to the monitoring vertical line is extracted. When the water level change rate at a certain moment reaches the set inflection point threshold, and the rate of change of the water level change rate before and after that moment changes significantly abruptly, with the abrupt change exceeding the preset fluctuation range, and after verification in conjunction with the wetland's regulation and storage characteristics, the point corresponding to that moment is identified as the inflection point of the water level change rate of the water level monitoring vertical line inside the wetland. This point can reflect the dynamic turning state of the wetland's regulation and storage capacity.

[0039] Step 3.3 involves calibrating the identified instantaneous flow peak points, water level inflection points, velocity gradient extreme points, and water level change rate inflection points according to their spatial coordinates and corresponding times. These are then used as feature control points with precise spatiotemporal coordinates. Specifically, this includes accurately calibrating the identified instantaneous flow peak points, water level inflection points, velocity gradient extreme points, and water level change rate inflection points to ensure the accurate and traceable spatiotemporal information of each feature point. The calibration process strictly adheres to the unified spatial benchmark of the four-level drainage network in the Chaohu Lake basin polder area. This unified spatial benchmark is consistent with the spatial benchmark of the constructed distributed hydrophysical framework and dynamic input field. To ensure the uniformity and relevance of spatial coordinates, the calibration operation is as follows: First, each feature point is checked one by one to confirm its corresponding specific cross-section, i.e., the instantaneous flow peak point corresponds to the field outlet cross-section, the water level inflection point corresponds to the confluence cross-section of the irrigation canal, the velocity gradient extreme point corresponds to the ecological pond inlet cross-section, and the water level change rate inflection point corresponds to the wetland internal water level monitoring vertical line, accurately locating the specific position of each feature point on the corresponding cross-section; then, the specific position of each feature point on the cross-section is accurately measured to obtain accurate spatial coordinates, including longitude and latitude, accurate to 6 decimal places, ensuring that the accuracy of spatial positioning meets the application requirements of subsequent steps.

[0040] The specific time corresponding to each feature point on the generated non-constant receding water curve is recorded synchronously, accurate to the hour, consistent with the time step of the hourly iterative simulation in step 2.3. The spatial coordinates (longitude and latitude) of each feature point are associated with the corresponding time to establish a spatiotemporal association table of feature points, ensuring that each feature point has a unique spatiotemporal identifier and avoiding confusion or mismatch. After calibration, these points with accurate spatiotemporal coordinates and precise correspondence with each section of the four-level receding water network are uniformly used as feature control points. These feature control points can accurately reflect the key turning point of the receding water process after a rainstorm in the polder area of ​​Chaohu Lake, providing accurate and reliable node data support for the temporal phase space reconstruction of receding water and the construction of hydrodynamic manifolds.

[0041] In a preferred embodiment of the present invention, step 4 above may include: Step 4.1: Extract the spatiotemporal coordinates of each feature control point and its corresponding drainage hydraulic parameters to construct a feature state vector set reflecting the key states of the drainage process. Specifically, this includes: extracting the precise spatiotemporal coordinates of each feature control point and the drainage hydraulic parameters that precisely match those coordinates from the calibrated feature control points. The spatiotemporal coordinates specifically include the longitude and latitude of the section where the feature control point is located and the corresponding drainage time. The drainage time is consistent with the time step of the hourly iterative simulation in Step 2.3, accurate to the hour, to ensure that the time sequence information is completely connected with the previous simulation process. The drainage hydraulic parameters specifically include instantaneous flow rate, water level, flow velocity, water level change rate, and flow velocity gradient. All hydraulic parameters are derived from the generated non-constant drainage curve and the calculated flow velocity gradient curve and water level change rate curve. During the extraction process, the spatiotemporal correspondence between the parameters and the feature control points is checked one by one to avoid parameter mismatch or missing parameters, ensuring that the parameter sources are traceable and the values ​​are accurate.

[0042] By combining the spatiotemporal coordinates of the same feature control point with multiple drainage hydraulic parameters in a one-to-one correspondence, a single feature state vector that can completely reflect the key states of the drainage process is formed. This single feature state vector is a structured multidimensional data unit. With the feature control point as the core, the data is systematically integrated according to the logical dimensions of spatial location, time node, and hydraulic parameters. The spatial dimension includes two core parameters, cross-sectional longitude and latitude, which accurately anchor the physical location of the feature control point in the four-level drainage network of the Chaohu Lake basin polder area. The time dimension only includes the drainage time corresponding to the feature control point, which is completely aligned with the time scale of the hourly iterative simulation. The hydraulic parameter dimension covers five core indicators: instantaneous flow, water level, flow velocity, water level change rate, and flow velocity gradient, which comprehensively reflects the hydraulic characteristics of the drainage process at that spatiotemporal node. Each vector can independently and completely represent a key turning point in the drainage process (such as the peak flow at the field outlet, the water level reversal in the ditch, etc.), and the vectors corresponding to different feature control points maintain uniformity in data dimension and format, and have the characteristics of being comparable and integrable.

[0043] The same combination and sorting operation is performed on all feature control points to remove abnormal vectors with missing parameters and mismatched coordinates. Finally, a feature state vector set is formed covering the cross-section of the field outlet, the cross-section of the irrigation ditch, the cross-section of the ecological pond inlet, and the vertical line of the water level monitoring inside the wetland. This feature state vector set can completely retain the key spatiotemporal information and hydraulic change information in the water receding process, providing a standardized and unambiguous data foundation for phase space reconstruction, and ensuring that the reconstruction process can accurately capture the nonlinear evolution characteristics of the water receding process.

[0044] Step 4.2 involves reconstructing the phase space of the feature state vector set. Based on the preset embedding dimension and delay time, the one-dimensional receding water time series data is mapped to a high-dimensional phase space, obtaining a phase space point set representing the evolution trajectory of the receding water process. Specifically, this includes: reconstructing the phase space of the constructed feature state vector set using a delayed coordinate phase space reconstruction method. The core of this method is to map the one-dimensional receding water time series data to a high-dimensional phase space, breaking the limitations of one-dimensional data and clearly presenting the dynamic evolution law and internal correlation of the receding water process; first, determining the preset embedding dimension and delay time. The embedding dimension is determined by combining the characteristics of the receding water time series data from the Chaohu Lake basin polder area using the spurious nearest neighbor method. The specific implementation process of the spurious nearest neighbor method is as follows: first, starting from the smallest embedding... The embedding dimension is gradually increased. For each candidate embedding dimension, the distance between all points in the phase space and their neighbors is calculated at the current dimension. The dimension is then increased by 1, and the distance between these point pairs is recalculated. If the distance between a neighboring point increases significantly after the dimension is increased (exceeding the preset distance threshold, which is determined in combination with the fluctuation characteristics of the receding water data in the Chaohu Lake Basin), then the neighboring point is determined to be a false neighboring point. The embedding dimension is continuously increased until the proportion of false neighbors drops below the preset standard (e.g., below 5%). The dimension at this point is the final determined embedding dimension. After multiple verifications and optimizations, it is ensured that the nonlinear characteristics of the receding water process can be fully captured, avoiding information omission due to too low an embedding dimension and redundant calculation due to too high an embedding dimension.

[0045] The delay time is determined using autocorrelation analysis: First, the similarity of the receding water time series data (selecting instantaneous flow rate or water level) under different time delays is analyzed. By comparing time series data at different time intervals, the correlation between the data and the time delay is quantified. A curve is plotted with time delay as the horizontal axis and data similarity as the vertical axis. The time delay corresponding to when the data similarity decreases to 1 / e (the reciprocal of the natural constant) of the initial state or a preset proportion (such as 50%) is selected as the initial candidate value. Then, it is adapted and adjusted in conjunction with the time step of the hourly iterative simulation in step 2.3 to ensure that the delay time is an integer multiple of 1 hour. The finally determined delay time can reasonably reflect the temporal correlation of the time series data and avoid time series data... The data redundancy and missing key information are addressed. During the reconstruction process, based on the hydraulic parameters of the receding water flow (preferably instantaneous flow or water level as the core time series data) in the feature state vector set, the one-dimensional time series data is mapped point by point to a high-dimensional phase space according to the determined embedding dimension and delay time. Each high-dimensional phase space point corresponds to a continuous key state sequence in the receding water flow process, which can fully reflect the changing trend of the hydraulic parameters of the receding water flow during that period. The entire mapping process ensures the spatiotemporal correlation and data accuracy of each phase space point, and finally obtains a phase space point set that can fully characterize the dynamic evolution trajectory of the receding water flow process, without information redundancy or data mismatch, providing a reliable high-dimensional data foundation for manifold learning and nonlinear dimensionality reduction processing.

[0046] Step 4.3 involves performing manifold learning and nonlinear dimensionality reduction on the phase space point set to extract the low-dimensional smooth manifold structure embedded in the high-dimensional phase space, thus constructing a hydrodynamic manifold that characterizes the inherent nonlinear evolution path of the receding process. Specifically, this includes performing manifold learning and nonlinear dimensionality reduction on the obtained phase space point set using a locally linear embedding method. This method effectively removes redundant information and noise interference in the high-dimensional phase space while preserving the inherent evolutionary laws and key correlations of the receding process, accurately extracting the low-dimensional smooth manifold structure hidden in the high-dimensional phase space, and finally constructing a hydrodynamic manifold that characterizes the inherent nonlinear evolution path of the receding process. First, the phase space point set is preprocessed to remove abnormal phase space points caused by data errors, ensuring the integrity of the point set. To ensure accuracy and reliability, the preprocessing standards are determined by combining the actual hydrological fluctuation range of the Chaohu Lake basin's polder area. By comparing the core hydraulic parameters such as instantaneous flow and water level corresponding to phase space points with reasonable thresholds for actual hydrology, outliers exceeding the range are screened and eliminated to avoid affecting the accuracy of the manifold structure and the reliability of the analysis. Then, the preprocessed phase space point set is dimensionality-reduced using a local linear embedding method: specifically, a reasonable local neighborhood is determined for each phase space point, and the neighborhood range is determined by combining the distribution density of phase space points to ensure that the neighborhood contains a sufficient number of neighboring points with close temporal correlation. Then, a linear relationship is constructed through the point set within the neighborhood to fit the local linear representation of each phase space point, firmly preserving the intrinsic correlation between points and replicating the local evolution characteristics of the receding process.

[0047] Based on the constructed local linear relationship, the high-dimensional phase space point set is mapped to a low-dimensional space, and redundant dimensions unrelated to the recession evolution law in the high-dimensional space are eliminated, forming a smooth, continuous, and unbroken low-dimensional manifold structure, i.e., a hydrodynamic manifold. This hydrodynamic manifold, capable of characterizing the inherent nonlinear evolution path of the recession process, is a low-dimensional smooth structure that closely matches the actual four-level recession network of the Chaohu Lake basin polder area, and its structural morphology highly conforms to the inherent dynamic laws of the recession process. The inherent dynamic laws of the recession process are the inherent correlation and change rules of hydraulic parameters over time during the recession process after a rainstorm in the Chaohu Lake basin polder area, specifically manifested as: following the principles of field runoff generation, ditch confluence, ecological pond regulation, and wetland stabilization. The established four-stage drainage network exhibits a hydraulic connectivity logic where the drainage intensity gradually decreases from the peak runoff in the fields. This decrease is not uniform but is influenced by various factors such as topographic slope, soil infiltration, and the storage capacity of ponds and wetlands, resulting in nonlinear fluctuations. Core hydraulic parameters such as instantaneous flow rate, water level, and velocity are interconnected and change synergistically. For example, after the peak runoff in the fields, the water level in the irrigation ditches rises and the flow velocity increases. Upon entering the ecological ponds, the flow rate slows down and the water level stabilizes due to the storage effect. Finally, the water volume is regulated by the wetlands, reaching a stable state. This dynamic evolution logic of peak, decay, transition, and stability, along with the synergistic changes in the hydraulic parameters of each section, constitutes the inherent dynamic law of the drainage process.

[0048] Hydrodynamic manifolds can fully replicate this pattern. Each manifold point corresponds to a key state in the recession process, and the connections between points accurately reflect the dynamic changes in recession parameters. This preserves the nonlinear fluctuation characteristics of the recession process while eliminating redundant interference, achieving a structured and visual expression of the essential characteristics of recession dynamics. It clearly presents the inherent evolutionary laws of the recession process, providing a reliable manifold structure basis for Lyapunov index spectrum calculation, and ensuring that the index calculation can accurately quantify the nonlinear intensity of the recession process.

[0049] Step 4.4: Calculate the Lyapunov index spectrum of the hydrodynamic manifold to quantify the convergence and divergence characteristics of the receding water trajectory in phase space. This serves as a criterion for verifying the manifold construction quality and the nonlinear intensity of the receding water process. Specifically, this includes: calculating the Lyapunov index spectrum of the constructed hydrodynamic manifold. The core purpose is to quantify the convergence and divergence characteristics of the receding water trajectory in phase space, thereby verifying the construction quality of the hydrodynamic manifold and determining the nonlinear intensity of the receding water process, providing theoretical support for the precise control of the receding water process. The mathematical expression of the Lyapunov index spectrum is: ; In the formula Indicates the first Lyapunov index, The number is exponential, consistent with the embedding dimension m determined in step 4.2; t represents the evolution time of the receding process, which is adapted to the time scale of the hourly iterative simulation in step 2.3. The initial small disturbance of the phase space point is represented by the disturbance range, which is determined in combination with the actual hydrological fluctuation characteristics of the polder area in the Chaohu Lake Basin to ensure that the disturbance conforms to the actual hydrological scenario. Indicates the initial disturbance Evolutionary deviation after evolution time t; The Euclidean norm of a vector in phase space is used to characterize the magnitude of the vector, i.e., the perturbation amplitude.

[0050] During the calculation, the complete Lyapunov index spectrum was obtained step by step, based on the phase space points on the hydrodynamic manifold. Positive Lyapunov indices represent the divergence of the receding water trajectory in phase space; the greater the divergence, the stronger the nonlinear fluctuation intensity of the receding water process. Negative Lyapunov indices represent the convergence of the receding water trajectory in phase space; the more obvious the convergence, the stronger the trend of the receding water process approaching a stable state. Zero indices represent that the trajectory is in a stable evolutionary state. This Lyapunov index spectrum ultimately serves as the core quantitative basis for verifying the quality of the hydrodynamic manifold construction (the rationality and stability of the index spectrum) and judging the nonlinear dynamic intensity of the receding water process in the polder area of ​​Chaohu Lake Basin, ensuring that subsequent receding water process analysis based on the manifold structure has high accuracy and reliability.

[0051] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1: Based on the constructed hydrodynamic manifold, calculate its Riemannian metric tensor to obtain the metric matrix used to measure the local geometric properties of the manifold. Specifically, this includes: based on the constructed hydrodynamic manifold... The calculation of the Riemannian metric tensor is carried out in the local coordinate system of the manifold. The core is to characterize the local geometric properties of the manifold by quantizing the inner product of vectors in the tangent space of the manifold. The specific calculation process is as follows: Select a hydrodynamic manifold. any point on Establish a local coordinate system for this point, and let the local coordinates be... d represents the low-dimensional dimension of the hydrodynamic manifold, consistent with the dimension after dimensionality reduction in step 4.3; the standard inner product operation on the tangent space of the manifold is defined, and the Riemannian metric tensor is defined. The mathematical expression is In the formula The components of the Riemannian metric tensor are represented by the formula: ; This represents the inner product of vectors in the tangent space. Representing phase space points respectively For local coordinates The partial derivatives, Represents the differential of local coordinates. Represents the tensor product.

[0052] During the calculation, the partial derivatives at each local coordinate are first solved point by point, and then the metric tensor components are obtained through inner product operations. Arrange all components in their ordered positions to form a d×d order metric matrix. This metric matrix can accurately reflect the hydrodynamic manifold. The degree of curvature, directional changes, and spatial structure at various local locations provide basic geometric constraints for subsequent geodesic equation solving and curvature calculation, ensuring the accuracy and rationality of subsequent calculations.

[0053] Step 5.2: Based on the metric matrix, solve the geodesic equations on the hydrodynamic manifold, and simultaneously calculate the geodesic curvature tensor. Then, extract the norm of the geodesic curvature tensor as the first geometric feature characterizing the bending intensity of the receding water trajectory. Specifically, this includes: based on the obtained metric matrix... First, calculate the Christofel symbol that is uniquely determined by it. Next, solve the geodesic equation, and finally calculate the geodesic curvature tensor and its norm as the first geometric characteristic quantity. The specific process is as follows: First, calculate the Christofel notation, whose expression is: In the formula The inverse matrix element of the metric matrix G, Let be the first-order partial derivatives of the metric tensor components with respect to their corresponding local coordinates; solve the geodesic equation, its expression is: In the formula Let be the local coordinate components of the geodesic, and s be the arc length parameter along the geodesic. The second covariant derivative along the geodesic is... The first derivatives of the geodesic coordinate components with respect to the arc length are given; after solving, the geodesic curvature tensor is calculated. Then, the norm is obtained by performing modulus calculation on each component. This norm is used to characterize the receding water trajectory in hydrodynamic manifolds. The first geometric characteristic of the degree of bending and nonlinear deformation , recorded as This provides a geometric quantification basis for subsequent feature fusion.

[0054] Step 5.3 involves performing statistical analysis on the point set on the hydrodynamic manifold to estimate its probability density function and obtain the empirical probability distribution on the hydrodynamic manifold. Specifically, this includes: analyzing the obtained hydrodynamic manifold... All phase space point sets (N is the total number of phase space points on the manifold) A global statistical analysis is performed, and the kernel density estimation method is used to smoothly fit the distribution characteristics of the phase space points, accurately estimating their probability density function. The specific calculation process is as follows: a Gaussian kernel function is selected as the basis function for kernel density estimation; the kernel function expression is... In the formula The bandwidth of the kernel function is determined by combining it with the distribution density of manifold points to ensure fitting accuracy; Representing hydrodynamic manifolds any point on, Denotes the first on the manifold Individual points in phase space ( ), Let d represent the Euclidean distance between the two points, and d be the low-dimensional dimension of the manifold (consistent with the local coordinate dimension in step 5.1).

[0055] Probability density function calculated based on kernel function Its expression is as follows: During the calculation, each phase space point on the manifold is substituted one by one. The positions on the manifold are obtained by weighted summation using kernel functions. The corresponding probability density values ​​are integrated according to the manifold spatial order to obtain an empirical probability distribution that reflects the frequency of occurrence, aggregation characteristics, and distribution concentration trend of the receding state. This distribution fully preserves the statistical regularity of the receding process in phase space, providing a reliable statistical basis for the subsequent construction of the Fisher information matrix.

[0056] Step 5.4: Construct the Fisher information matrix based on the empirical probability distribution, and calculate the spectral entropy of the Fisher information matrix as the second statistical feature quantity for measuring the information complexity of the receding process. Specifically, this includes: based on the obtained empirical probability distribution... and its probability density function In hydrodynamic manifold Within the local geometric framework, a Fisher information matrix is ​​constructed, and then the spectral entropy is calculated through matrix eigenvalue decomposition as the second statistical feature. The specific calculation process is as follows: First, the Fisher information matrix I is constructed, and its mathematical expression is: ; In the formula This represents the element in the i-th row and j-th column of the Fisher information matrix. Represents the local coordinates of the manifold (consistent with step 5.1). , Represent the logarithm of the probability density function in local coordinates. The first-order partial derivative, Representing hydrodynamic manifolds Volume element on, where The expression is , Representation of the metric matrix The determinant; the integral range is the entire hydrodynamic manifold. After constructing the Fisher information matrix I, its eigenvalues ​​are decomposed to obtain the eigenvalues. Then, the spectral entropy is calculated based on the eigenvalues. Its mathematical expression can be represented as In the formula Spectral entropy represents the sum of all eigenvalues ​​and is used to normalize individual eigenvalues. The range of values ​​is The spectral entropy is used as the second statistical characteristic to measure the uniformity of information distribution, dynamic complexity, and nonlinear order of the receding process. , recorded as .

[0057] Step 5.5 involves weighting and fusing the first geometric feature quantity and the second statistical feature quantity, and then normalizing them to generate a drainage morphology coefficient that characterizes the nonlinear evolution of farmland drainage. Specifically, this includes: weighting and fusing the first geometric feature quantity... With the second statistical characteristic Weighted fusion and normalization are performed to generate a drainage morphology coefficient characterizing the nonlinear evolution of farmland drainage in the polder areas of the Chaohu Lake Basin. The specific calculation process is as follows: First, a preset weight is determined, which is determined in conjunction with the drainage characteristics of the polder areas in the Chaohu Lake Basin. The first geometric feature quantity is set as follows. The weight is Second statistical characteristic The weight is ,satisfy The weighted fusion formula is: ; the results after weighted fusion Extreme value normalization is performed using a linear normalization method, mapping the values ​​to the [0,1] interval. The normalization formula is as follows: In the formula, C represents the final generated drainage morphology coefficient. This represents the minimum value of the weighted fusion result of all samples. This represents the maximum value of the weighted fusion result of all samples. Through the above calculation, a drainage morphology coefficient C is generated, which can comprehensively reflect the geometric tortuosity, statistical distribution characteristics, and overall nonlinear evolution law of farmland drainage in the polder area of ​​Chaohu Lake Basin in terms of hydrodynamic manifold. This coefficient can be used as a unified quantitative indicator to quantitatively evaluate the complexity, dynamic characteristics, and evolution pattern of the drainage process. Intuitively, the larger the drainage morphology coefficient C, the more drastic the fluctuation of the drainage process in time and the more obvious the nonlinear characteristics, which means that the drainage load prediction and storage capacity allocation need to have a higher dynamic response capability. Conversely, the smaller the coefficient, the smoother the drainage process and the closer it is to linear decay, and the more stable the scheduling decision is, providing a quantitative basis for the precise control and optimization of the subsequent drainage process.

[0058] In a preferred embodiment of the present invention, step 6 above may include: Step 6.1: Based on the pre-determined water level-storage capacity relationship curves of each regulation and storage facility, and combined with the real-time water level data of ponds and wetlands at all levels, calculate the remaining available regulation and storage capacity of each facility at the current moment to obtain the real-time status dataset of the regulation and storage facility group. Specifically, this includes: based on the water level-storage capacity relationship curves of ponds, ecological ponds, and wetlands at all levels obtained through on-site measurements, multi-water level operating condition tests, and parameter calibration, the water level-storage capacity relationship curves reflect the one-to-one correspondence between different water levels and corresponding water storage volumes of each regulation and storage facility. They cover the entire water level range from the lowest operating water level to the maximum allowable water level, accurately characterizing the correlation between water level changes and storage capacity changes, and providing a core basis for storage capacity calculation.

[0059] Combining real-time water level data collected from field outlet sections, the confluence of irrigation ditches, the inlet sections of ecological ponds, and the vertical water level monitoring lines within the wetland, and after outlier removal, time-series completion, and smoothing preprocessing, the time step of the hourly iterative simulation in step 2.3 is used as the time reference. For each storage facility, the occupied storage capacity corresponding to the measured water level at the current moment is calculated. The calculation process is as follows: first, based on the current measured water level, the water storage volume at the corresponding water level on the water level-storage capacity relationship curve of the facility is found. This water storage volume is the occupied storage capacity at the current moment. Then, by subtracting the occupied storage capacity from the total designed storage capacity of each facility, the remaining storage capacity actually available for that facility at the current moment is obtained. At the same time, information such as the remaining available storage capacity, real-time water level, spatial coordinates, design storage capacity parameters, maximum allowable water level, minimum operating water level, and hydraulic connectivity between facilities are uniformly integrated to form a real-time status dataset of the storage facility group covering the entire four-level drainage network of the Chaohu Lake basin, with consistent time sequence and complete information. This provides a real, reliable, and traceable initial status basis for drainage load prediction, storage capacity allocation, and scheduling decisions.

[0060] Step 6.2: Based on the non-constant water discharge curve, extract the predicted water discharge volume for each time period within the scheduling period corresponding to the storage capacity to be allocated. Combine the predicted water discharge volume with the water discharge morphology coefficient to perform nonlinear correction, and obtain a dynamic water discharge load prediction sequence for each time period. Specifically, based on the generated non-constant water discharge curve after wavelet denoising, outlier removal and cubic spline interpolation optimization, according to the scheduling period length set in advance based on the scheduling habits of the polder area and the hydrological change characteristics, extract the basin average water discharge flow and cumulative water discharge volume for each time period within the entire scheduling cycle to form a temporally continuous and uniformly distributed initial water discharge volume prediction sequence. Then, introduce the water discharge morphology coefficient to perform nonlinear correction on the initial water discharge volume prediction sequence.

[0061] The specific correction calculation process is as follows: First, the core function of the receding water morphology coefficient is to quantify the nonlinearity of the receding water process. Its value is positively correlated with the bending intensity of the receding water trajectory and the information complexity. During correction, the receding water morphology coefficient is normalized to ensure that its value matches the magnitude of the initial receding water volume prediction sequence. Then, for each time period in the initial receding water volume prediction sequence, combined with the corresponding receding water evolution stage (peak receding water stage, rapid decay stage, smooth transition stage, stable stage), the normalized receding water morphology coefficient is correlated with the initial predicted receding water volume for the corresponding time period. For the peak receding water period, the peak amplitude is appropriately adjusted according to the magnitude of the receding water morphology coefficient. The larger the coefficient, the larger the peak correction amplitude, which conforms to the nonlinear fluctuation characteristics of the actual receding water peak. For the rapid decay stage, the decay rate is adjusted in conjunction with the receding water morphology coefficient. A larger number indicates stronger nonlinearity in the receding water flow, resulting in more significant correction of the attenuation rate and avoiding deviations caused by excessively fast or slow attenuation under linear assumptions. For the gradual transition phase, the receding water volume at the transition node is fine-tuned based on the receding water morphology coefficient to ensure that the transition conforms to the evolution law of the hydrodynamic manifold, achieving adaptive adjustment of key features of the receding water process. The correction process fully integrates the geometric bending characteristics, information complexity, and nonlinear evolution intensity of the receding water process in the hydrodynamic manifold, enabling the corrected prediction results to better match the dynamic changes in the actual receding water flow after heavy rain in the Chaohu Lake basin. This effectively reduces the prediction deviation caused by traditional linear assumptions. After correction, a dynamic receding water load prediction sequence is formed that changes dynamically over time, is continuously updated in each time period, and is highly matched with the hydraulic characteristics of the four-stage receding water network, providing accurate, stable, and hydrodynamically consistent input loads for time-by-time reservoir capacity allocation.

[0062] Step 6.3: Using the dynamic water discharge load prediction sequence as input and the real-time status dataset of the water storage facility group as initial conditions, the water storage capacity of ponds and wetlands at each level is determined time-by-time according to the multi-objective optimization allocation principle, resulting in a water storage capacity allocation scheme that meets the flood control constraints and pollution reduction objectives of the polder area. Specifically, this includes: using the obtained dynamic water discharge load prediction sequence as the time-by-time input condition and the obtained real-time status dataset of the water storage facility group as initial constraints and boundary conditions, and following the multi-objective optimization allocation principle that balances polder flood control safety, balanced utilization of water storage space, stable facility operation, and non-point source pollution reduction, this multi-objective optimization allocation principle prioritizes safety, coordinated efficiency, and balanced practicality. Specifically, the core objectives are: First, to prioritize flood control safety in the polder area, strictly control water levels at key sections to ensure they do not exceed set thresholds, and eliminate the risk of flooding. Second, to focus on the balanced utilization of storage space, avoiding long-term idleness of some facilities and overloading of others, and maximizing the overall storage efficiency of the storage facility group. Third, to ensure stable facility operation, avoid frequent fluctuations in storage capacity allocation, and reduce losses caused by frequent start-ups and shutdowns of gate pumps. Fourth, to strengthen the coordination of non-point source pollution reduction, prioritize the allocation of storage capacity to facilities with purification functions such as ecological ponds and wetlands, and improve pollution removal efficiency by extending the retention time of receding water. These four objectives are mutually synergistic and prioritize each other, forming a complete optimization guideline.

[0063] Under the strict constraints of meeting the water level control thresholds of key sections in the polder area, the maximum storage capacity of each storage facility, the hydraulic connectivity sequence of the four-stage drainage network, the flow capacity of channels and sections, and the safe operating range of gate pumps, a rolling optimization method based on time periods is adopted. This method uses the time step of the hourly iterative simulation in step 2.3 as the optimization unit. The specific implementation process is as follows: Based on the dynamic drainage load and the real-time status of the storage facility group in the current time period, the storage capacity allocation for this time period is completed, and the real-time water storage status of each level of facility is determined; then, combined with the predicted dynamic drainage load value for the next time period, the allocation result for this time period is slightly corrected to avoid discontinuity in allocation between time periods; after each time period allocation is completed, the status data of the storage facility group is updated in real time, including... The remaining available reservoir capacity and current water level are used as initial constraints for optimization in the next time period. This process is repeated for each time period, with the optimization calculations rolling forward to ensure that the allocation scheme for each time period can adapt to the current water discharge load and facility status. During the allocation process, the rolling optimization logic for each time period is strictly followed, prioritizing the protection of the flood control safety baseline, balancing the water discharge load borne by each facility, avoiding risks such as overload or overflow in local areas or single facilities, and giving priority to utilizing the natural storage space of ecological ponds and wetlands to improve the pollution reduction effect. Finally, the optimal reservoir capacity allocation scheme is obtained that satisfies the flood control safety constraints of the polder area and achieves the synergistic goals of water storage and pollution reduction throughout the entire scheduling cycle, ensuring that the allocation result has engineering feasibility, operational stability, and field applicability.

[0064] Step 6.4: Based on the reservoir capacity allocation scheme, by combining the gate pump opening and closing characteristics and hydraulic transmission time of each storage facility, a storage operation scheme is generated, which includes the gate pump opening and closing sequence and target water depth of each facility. Specifically, this includes: based on the determined optimal reservoir capacity allocation scheme, combining the opening and closing characteristics, rated operating flow, control logic, response delay, regulation accuracy of gate pumps for each level of ponds, ecological ponds, and wetlands, as well as the hydraulic transmission time and water loss along the way between different monitoring sections and storage facilities, and taking the reservoir capacity demand and water level control target for each time period as the core control basis, the starting time, running time, opening and closing sequence, opening control range, and corresponding target water depth of each gate pump facility are reverse-engineered. The reverse-engineering process is as follows: first, confirm the target water depth and the amount of water to be replenished or discharged for each level of storage facility in each time period; combine the current actual water level of the facility and the remaining available reservoir capacity to determine the water regulation task that the gate pump needs to complete in that time period; then, based on the rated operating flow of the gate pumps, The system adjusts the accuracy of water flow by calculating the runtime required to complete the adjustment task. It also considers hydraulic transmission time and water loss along the route to determine the start-up time of the gate pumps, ensuring that water reaches the target facilities on time and accurately meets the reservoir capacity requirements for the specified period. Simultaneously, based on the hydraulic connection sequence of the four-level drainage network and the gate pump control logic, the system determines the opening and closing sequence of the gate pumps at each level to avoid hydraulic conflicts or storage failures due to improper opening and closing sequences. Furthermore, it fine-tunes the start-up time and runtime by incorporating the gate pump response delay, ensuring a smooth and controllable adjustment process. The system breaks down macro-level reservoir capacity allocation instructions layer by layer and precisely converts them into directly executable individual equipment control instructions. This results in a complete storage and regulation operation plan covering the entire polder area and the entire scheduling cycle, including the opening and closing sequence of gate pumps for each storage facility, runtime, control water level range, and target storage depth. This plan can directly guide the automated scheduling and refined operation of the four-level drainage network in the Chaohu Lake basin, achieving safe, efficient, balanced, and coordinated intelligent control of the rainstorm drainage process.

[0065] In a preferred embodiment of the present invention, step 7 above may include: Step 7.1: Load the storage and regulation operation scheme into the constructed runoff-generation coupling model to drive the model to reenact the receding process according to the gate pump opening and closing sequence and target water depth specified in the storage and regulation operation scheme, and obtain the scheme simulated receding curve. Specifically, this includes: loading the generated complete storage and regulation operation scheme into the pre-constructed and calibrated runoff-generation coupling model. The model has incorporated the core physical features of the Chaohu Lake basin polder area, such as topography, soil type, vegetation cover, and hydraulic connectivity of the four-level receding network, and can accurately replicate the entire process of runoff generation and confluence after rainstorm. After loading, the model is controlled to reenact the receding process time by time according to the actual receding sequence, using the gate pump opening and closing sequence, target water depth, and running time specified in the storage and regulation operation scheme as driving conditions.

[0066] The specific reenactment simulation process is as follows: using the hourly iterative simulation time step in step 2.3 as the unified simulation unit, the specific instructions such as the gate pump opening and closing status, opening and closing degree, target water storage depth, and running time of each time period and each storage facility in the storage operation plan are read synchronously. At the same time, the refined physical parameters of the Chaohu Lake basin polder area pre-stored in this model are called, including the topographic slope, soil texture and infiltration coefficient, vegetation coverage and interception capacity of different areas, roughness and cross-sectional dimensions of channels at all levels, and leakage coefficient of storage facilities. After the simulation starts, the field is started. In the block runoff simulation stage, the process of rainwater infiltration and surface runoff generation within the field is simulated by combining the previous heavy rainfall and the initial soil moisture content. The runoff volume, runoff intensity, and runoff initiation time of each field at each time period are accurately calculated. According to the hydraulic connectivity sequence of the four-level drainage network of fields, irrigation ditches, ecological ponds, and wetlands, the confluence process simulation is initiated. Based on parameters such as the channel roughness, flow capacity, and cross-sectional slope of the irrigation ditches, the confluence velocity of the drainage in the channel, the water loss along the way, and the flow distribution at each confluence section are calculated to ensure that the confluence process closely matches the actual hydraulic characteristics.

[0067] Based on the opening and closing sequence of gate pumps and the target water depth of water storage facilities such as ecological ponds and wetlands in the water storage operation plan, the process of water storage, regulation and discharge after the receding water enters the water storage facilities at all levels is simulated. The real-time water level, occupied storage capacity and remaining available storage capacity of each water storage facility are monitored and updated in real time. At the same time, the inflow and outflow of receding water are controlled according to the rated flow and opening degree of the gate pumps to avoid overload, overflow or insufficient water storage of the facilities. During the simulation, the receding water flow and water level change data of all monitoring sections, such as the field outlet, the confluence of the irrigation canal, the inlet and outlet of the ecological pond, and the inside of the wetland, are recorded in real time. The entire process strictly follows the actual receding water logic of runoff generation, confluence, regulation and discharge, and the simulation is carried out in an orderly manner in each time period to ensure that the simulation of each link is highly matched with the instructions of the water storage operation plan and the actual hydrological and physical characteristics of the polder area. Finally, a simulated receding water curve that can comprehensively and realistically reflect the implementation effect of the water storage operation plan is generated.

[0068] Step 7.2 involves comparing the simulated runoff curve with the non-constant runoff curve to calculate the deviation between the simulated runoff curve and the non-constant runoff curve at the peak runoff time and peak flow rate, thus obtaining a peak deviation vector reflecting the simulation accuracy. Specifically, this includes: comprehensively comparing the obtained simulated runoff curve with the non-constant runoff curve after interpolation optimization. The comparison process uses the hourly iterative simulation time step in Step 2.3 as the unified comparison unit. First, the two curves are time-series aligned to ensure that the simulated data and actual predicted data correspond one-to-one. Then, a time-by-time data comparison is carried out, focusing on the core characteristic parameters in the runoff process: peak runoff time and peak flow rate, which are used as key indicators to judge the simulation accuracy. The core characteristic data corresponding to the two curves are extracted respectively. The specific time (accurate to the hour) and the corresponding simulated peak flow rate value are extracted from the simulated runoff curve. The actual predicted peak runoff time and the corresponding predicted peak flow rate value are extracted from the non-constant runoff curve to ensure that the extracted characteristic data are accurate and without deviation.

[0069] Peak time deviation and peak flow deviation are calculated by difference: Peak time deviation is obtained by subtracting the actual predicted peak time from the simulated peak flow time. If the result is positive, it means that the simulated peak occurred later than the actual predicted time; if it is negative, it means that the simulated peak occurred earlier than the actual predicted time. The absolute value reflects the degree of deviation in peak time. Peak flow deviation is obtained by subtracting the actual predicted peak flow from the simulated peak flow value. If the result is positive, it means that the simulated peak flow is greater than the actual predicted peak flow; if it is negative, it means that the simulated peak flow is less than the actual predicted peak flow. The absolute value reflects the degree of deviation in peak flow. After calculation, the peak time deviation and peak flow deviation are integrated in the order of peak time deviation first and peak flow deviation second to form a peak deviation vector that can intuitively reflect the degree of deviation between the simulated drainage curve and the actual drainage pattern. The value of this vector directly corresponds to the simulation accuracy. The closer the vector value is to zero, the higher the simulation accuracy, and vice versa. This provides a precise correction direction and quantitative basis for model parameter calibration.

[0070] Step 7.3: Based on the peak deviation vector, and combined with the real-time monitoring data, the key parameters of the runoff-runoff coupling model are corrected online to update the physical property parameters of each grid in the model, resulting in a parameter-optimized runoff-runoff coupling model. Specifically, based on the obtained peak deviation vector, and combined with the real-time monitoring data of water level and flow rate at various monitoring verticals such as field outlets, confluence sections of irrigation ditches, ecological pond inlets, and wetland interiors, the key parameters of the runoff-runoff coupling model are corrected online. The key parameters to be corrected include physical property parameters closely related to the drainage process, such as soil infiltration coefficient, surface roughness, leakage coefficient of storage facilities, and channel transmission coefficient. These parameters directly affect the model's simulation accuracy of runoff intensity, confluence velocity, and storage effect, and are the core of determining whether the model conforms to the actual drainage law.

[0071] During the calibration process, the peak deviation vector is analyzed to confirm the magnitude and direction of the peak time deviation and peak flow deviation. Then, combined with real-time monitoring data, the values ​​of each parameter are adjusted specifically to avoid blind calibration: If the peak flow deviation is positive, meaning the simulated peak is greater than the actual predicted peak, the soil infiltration coefficient is appropriately increased to reduce surface runoff, while the surface roughness and channel transmission coefficient are fine-tuned to slow down the confluence velocity and reduce the peak flow; if the peak flow deviation is negative, meaning the simulated peak is less than the actual predicted peak, the soil infiltration coefficient is appropriately decreased to increase surface runoff, while the channel transmission coefficient is adjusted simultaneously to accelerate the confluence velocity and increase the peak flow; if the peak time deviation is positive, meaning the simulated peak occurs later than expected, the channel transmission coefficient and surface roughness are adjusted accordingly to accelerate the confluence velocity and increase the peak flow; if the peak time deviation is positive, meaning the simulated peak occurs later than expected, the values ​​of the soil infiltration coefficient and surface runoff are adjusted accordingly to slow down the confluence velocity and reduce the peak flow. Surface roughness accelerates the confluence process, causing the peak to appear earlier. If the peak time deviation is negative, indicating that the simulated peak appears too early, the channel transmission coefficient and surface roughness are increased to slow down the confluence process, causing the peak to appear later. During the calibration process, the physical attribute parameters corresponding to each grid in the model are updated differently according to the geographical and hydrological differences in different areas of the Chaohu Lake basin. This ensures that the parameter adjustments fit the actual topography, soil, vegetation, and water conservancy facilities of the area, avoiding simulation deviations caused by uniform parameter adjustments across the entire region. At the same time, the effect of parameter adjustments is verified in real time to avoid excessive parameter adjustments affecting model stability. Finally, the optimized runoff-generation coupling model is obtained, which effectively improves the simulation accuracy of the recession process and provides a reliable model foundation for iterative optimization.

[0072] Step 7.4: The runoff-generation coupling model with optimized driving parameters is used to re-simulate the drainage process, and the comparison between the simulated drainage curve and the non-constant drainage curve is iteratively executed, along with the online correction of the model parameters, until the peak deviation vector converges within the preset accuracy threshold range. The simulation result that finally meets the deviation requirements is used as the optimized storage capacity configuration scheme. Specifically, this includes: based on the optimized runoff-generation coupling model, the complete storage operation scheme is reloaded, keeping the gate pump opening and closing sequence, target water depth, and running time instructions in the storage operation scheme unchanged, and the optimized model is used to re-simulate the drainage process. The simulation process strictly follows the principles of runoff generation, runoff confluence, and storage. The discharge logic is implemented step by step, with water level and flow data of each monitoring section recorded synchronously for each time period, ultimately yielding a new simulated drainage curve. The complete operation of step 7.2 is repeated: first, the new simulated drainage curve and the optimized non-constant drainage curve are time-series aligned; data is compared time-by-time, extracting the peak runoff time and peak flow of the two curves; a new peak deviation vector is calculated through the difference to confirm the improvement in simulation accuracy after parameter optimization; then, based on the magnitude and direction of the new peak deviation vector, combined with the real-time access to water level and flow data of each monitoring vertical line, the key parameters of the model are further targeted for online correction, fine-tuning parameter values ​​to further reduce simulation deviation.

[0073] The process iteratively executes the loading and storage operation plan, drives model simulation, compares two receding water curves, calculates the peak deviation vector, and corrects model parameters online. Each iteration aims to improve simulation accuracy, gradually optimizing model parameters and simulation results until both components of the peak deviation vector (peak time deviation and peak flow deviation) converge within a preset accuracy threshold range. If the peak deviation fails to converge after exceeding the preset maximum number of iterations (e.g., 10), or if the deviation value shows a divergent trend, the iteration automatically terminates. The simulation result with the smallest peak deviation vector magnitude from historical iterations is then selected as the suboptimal optimized storage capacity configuration, and an early warning is issued to balance algorithm stability and engineering practicality. The preset accuracy threshold is pre-set based on the actual needs of receding water scheduling in the Chaohu Lake basin, model simulation accuracy requirements, and engineering application standards. The preset accuracy threshold for peak time deviation is controlled within ± Within one hour, ensure that the deviation between the simulated peak occurrence time and the actual predicted time is small, and control the preset accuracy threshold of the peak flow deviation within ±5%, ensuring that the deviation between the simulated peak flow and the actual predicted peak flow is within the acceptable range for engineering. This threshold can be slightly adjusted according to the hydrological characteristics and scheduling priorities of different areas of the polder to ensure adaptability to actual application scenarios. At this point, the model simulation results can accurately match the actual pattern of water receding after heavy rain in the polder area of ​​Chaohu Lake, and the simulation accuracy meets the scheduling requirements. Integrate and sort out the regulation and storage operation parameters (sluice gate and pump opening and closing sequence, running time, opening degree control, etc.) corresponding to the simulation results that finally meet the deviation requirements with the reservoir capacity allocation parameters (regulation and storage capacity of facilities at all levels, target water storage depth, etc.) to form a complete optimized regulation and storage capacity configuration scheme. This scheme takes into account flood control safety, regulation and storage balance and pollution reduction, and provides more accurate, reliable and implementable technical support for the water receding scheduling of the polder area of ​​Chaohu Lake, ensuring the scientific and refined nature of water receding scheduling.

[0074] like Figure 2 As shown, embodiments of the present invention also provide a system for calculating the storage capacity of farmland runoff in the Chaohu Lake basin based on AI runoff-generation coupling simulation, including: The data acquisition module is used to acquire underlying surface and hydrological data of farmland and multi-level ditch and wetland regulation and storage systems in the Chaohu Lake basin. The coupled model construction module is used to construct a runoff-runoff coupling model constrained by both AI algorithms and hydrophysical mechanisms, which is used to characterize the drainage process from field plots to ditches and wetlands on a grid-by-grid basis. The drainage process simulation module is used to input real-time soil moisture, water level and flow monitoring data into the runoff-gene coupling model to perform hourly dynamic simulation of the drainage process after heavy rain, and obtain a non-constant drainage curve that reflects the peak of farmland runoff and the drainage process. The feature point recognition module is used to dynamically identify multiple feature control points with precise spatiotemporal coordinates from non-constant receding water curves. The manifold construction and analysis module is used to reconstruct the phase space of the drainage time series with the feature control points as nodes, and construct a hydrodynamic manifold that represents the nonlinear evolution path of the drainage process; perform information geometry analysis on the hydrodynamic manifold, extract the norm of its geodesic curvature tensor and the spectral entropy of the Fisher information matrix, and fuse them to obtain the drainage morphology coefficients that represent the nonlinear evolution characteristics of farmland drainage. The regulation and storage scheme generation module is used to dynamically allocate the regulation and storage capacity in time periods based on the non-constant water receding curve and water receding morphology coefficient, combined with the real-time monitoring of the current water storage status and physical regulation and storage capacity of ponds and wetlands at all levels, and generate a regulation and storage operation scheme that includes the gate pump opening and closing sequence and water storage depth. The closed-loop correction module is used to feed back the regulation and storage operation plan to the runoff-generated runoff coupling model for verification. It corrects the deviation of the simulated receding water peak online through real-time monitoring data. After closed-loop correction, an optimized regulation and storage capacity configuration plan that meets the requirements of flood control and non-point source pollution reduction in the polder area is obtained.

[0075] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0076] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for calculating the storage capacity of farmland runoff in the Chaohu Lake Basin based on AI runoff-generation coupling simulation, characterized in that, The method includes: Acquire underlying surface and hydrological data of farmland and multi-level ditch and wetland regulation and storage system in the polder area of ​​Chaohu Lake Basin, and construct a runoff-runoff coupling model constrained by both AI algorithm and hydrophysical mechanism to characterize the water receding process from field to ditch and wetland on a grid-by-grid basis. Real-time soil moisture, water level and flow monitoring data are input into the runoff-runoff coupling model to dynamically simulate the hourly drainage process after heavy rain, and obtain a non-constant drainage curve that reflects the peak of farmland runoff and the drainage process. Dynamically identify multiple feature control points with precise spatiotemporal coordinates from non-constant receding water curves; The phase space of the receding water time series is reconstructed using the aforementioned characteristic control points as nodes, and a hydrodynamic manifold characterizing the nonlinear evolution path of the receding water process is constructed. Information geometry analysis is performed on the hydrodynamic manifold to extract the norm of its geodesic curvature tensor and the spectral entropy of the Fisher information matrix, and then fused to obtain the drainage morphology coefficient that characterizes the nonlinear evolution of farmland drainage. Based on the non-constant receding water curve and receding water morphology coefficient, combined with the real-time monitoring of the current water storage status and physical regulation capacity of ponds and wetlands at all levels, the regulation storage capacity is dynamically allocated in time periods to generate a regulation operation plan that includes the opening and closing sequence of gate pumps and water storage depth. The regulation and storage operation plan is fed back to the runoff-generated runoff coupling model for verification. The deviation of the simulated receding water peak is corrected online by real-time monitoring data. After closed-loop correction, an optimized regulation and storage capacity configuration plan that meets the requirements of flood control and non-point source pollution reduction in the polder area is obtained.

2. The method for calculating the storage capacity of farmland runoff regulation reservoirs in the Chaohu Lake basin based on AI runoff-generation coupling simulation as described in claim 1, characterized in that, A runoff-runoff coupling model, constrained by both AI algorithms and hydrophysical mechanisms, is constructed to characterize the drainage process from farmland to ditches and wetlands on a grid-by-grid basis, including: Multi-source data on farmland and multi-level ditch and wetland regulation systems in the Chaohu Lake Basin were collected, including topographic elevation, soil texture and stratification parameters, crop type and planting system, layout and cross-sectional geometry of ditch and wetland, historical rainfall and water level and flow process. The multi-source data were processed to obtain a standardized basic dataset. Based on a standardized basic dataset, the target area is divided into grids according to a four-level drainage network of fields, irrigation ditches, ecological ponds, and wetlands. Each grid is assigned corresponding physical attribute parameters, including roughness coefficient, saturated hydraulic conductivity, field capacity, and initial soil moisture content, thereby constructing a distributed hydrophysical framework. The AI ​​algorithm is embedded in a distributed hydrophysical framework, and historical receding water process data is used to train the AI ​​algorithm. The output boundary of the AI ​​algorithm is constrained by physical mechanisms. At the same time, the AI ​​algorithm is used to optimize parameters in the physical framework that are difficult to measure directly. After parameter calibration and verification, a dual-constraint runoff-generation coupling model is obtained. The runoff-generated runoff coupling model is used to simulate the post-rainfall drainage process of each grid of the four-level drainage network hourly, and the runoff and runoff of each grid cell in the time series are output to form a data volume of drainage process characterized grid by grid.

3. The method for calculating the storage capacity of farmland runoff in the Chaohu Lake basin based on AI runoff-generation coupling simulation as described in claim 2, characterized in that, Real-time soil moisture, water level, and flow monitoring data are input into a runoff-gene coupling model to dynamically simulate the hourly recession process after a rainstorm. This yields a non-constant recession curve reflecting the peak of farmland runoff and the recession process, including: Discrete data collected by a real-time monitoring network deployed at various monitoring sections in fields, ditches, ponds, and wetlands is obtained, including soil moisture, water level, and flow data at each monitoring point. Outlier removal and missing value interpolation are performed on the discrete data, and spatial interpolation is performed based on the spatial coordinates of each monitoring point to convert the processed discrete data into a continuous field that matches the grid space of the runoff-runoff coupling model, thereby generating a dynamic input field. The dynamic input field is loaded as the initial and boundary conditions into the constructed runoff-generation coupled model to update the initial soil moisture content and the initial water level of ditches and wetlands in each grid of the model. The updated runoff-gene coupling model is driven to perform hourly iterative calculations on the post-rainfall drainage process, combined with real-time rainfall data, and outputs the flow rate and runoff of each grid cell hourly, forming an hourly updated dynamic data sequence of water balance. The flow and water level process data of the field outlet section, the confluence section of the irrigation canal, and the inlet section of the ecological pond are extracted from the dynamic water balance data sequence. After wavelet denoising and cubic spline interpolation, the non-constant receding water curve is generated.

4. The method for calculating the storage capacity of farmland runoff regulation reservoirs in the Chaohu Lake basin based on AI runoff-generation coupling simulation according to claim 3, characterized in that, Multiple feature control points with precise spatiotemporal coordinates are dynamically identified from the non-constant receding water curve, including: The generated non-constant receding water curve is numerically differentiated to calculate the derivatives of the receding water curve on the time axis, thereby obtaining the flow velocity gradient curve and the water level change rate curve of the receding water curve. Based on the velocity gradient curve and the water level change rate curve, combined with the preset physical feature thresholds, the instantaneous flow peak point of the field outlet section, the water level inflection point at the confluence of ditches, the velocity gradient extreme point of the ecological pond inlet section, and the water level change rate inflection point of the water level monitoring vertical line inside the wetland are identified respectively. The identified instantaneous flow peak points, water level inflection points, velocity gradient extreme points, and water level change rate inflection points are calibrated according to the spatial coordinates of their respective cross sections and the corresponding time, and are used as feature control points with precise spatiotemporal coordinates.

5. The method for calculating the storage capacity of farmland runoff regulation reservoirs in the Chaohu Lake basin based on AI runoff-generation coupling simulation according to claim 4, characterized in that, Using the aforementioned characteristic control points as nodes, the phase space of the receding water time series is reconstructed to construct a hydrodynamic manifold characterizing the nonlinear evolution path of the receding water process, including: Extract the spatiotemporal coordinates of each point and its corresponding hydraulic parameters for water discharge from the feature control points, and construct a feature state vector set that reflects the key states of the water discharge process. The phase space is reconstructed from the feature state vector set. The one-dimensional receding time series data is mapped to a high-dimensional phase space according to the preset embedding dimension and delay time to obtain the phase space point set representing the evolution trajectory of the receding process. Manifold learning and nonlinear dimensionality reduction are performed on the point set in phase space to extract the low-dimensional smooth manifold structure embedded in the high-dimensional phase space, and to construct a hydrodynamic manifold that characterizes the intrinsic nonlinear evolution path of the receding process. The Lyapunov index spectrum of the hydrodynamic manifold is calculated to quantify the convergence and divergence characteristics of the receding water trajectory in phase space, serving as a criterion for verifying the quality of manifold construction and the nonlinear intensity of the receding water process.

6. The method for calculating the storage capacity of farmland runoff regulation reservoirs in the Chaohu Lake basin based on AI runoff-generation coupling simulation according to claim 5, characterized in that, Information geometry analysis is performed on the hydrodynamic manifold to extract the norm of its geodesic curvature tensor and the spectral entropy of its Fisher information matrix. These are then fused to obtain drainage morphology coefficients characterizing the nonlinear evolution of farmland drainage, including: Based on the constructed hydrodynamic manifold, its Riemannian metric tensor is calculated to obtain the metric matrix used to measure the local geometric properties of the manifold; Based on the metric matrix, the geodesic equations on the hydrodynamic manifold are solved, and the geodesic curvature tensor is calculated. Then, the norm of the geodesic curvature tensor is extracted as the first geometric feature quantity characterizing the bending intensity of the receding water trajectory. Statistical analysis is performed on the point set on the hydrodynamic manifold to estimate its probability density function and obtain the empirical probability distribution on the hydrodynamic manifold. Based on the empirical probability distribution, a Fisher information matrix is ​​constructed, and the spectral entropy of the Fisher information matrix is ​​calculated as the second statistical feature quantity for measuring the information complexity of the dewatering process. The first geometric feature and the second statistical feature are weighted and fused, and after normalization, a drainage morphology coefficient is generated to characterize the nonlinear evolution of farmland drainage.

7. The method for calculating the storage capacity of farmland runoff drainage in the Chaohu Lake basin based on AI runoff-generation coupling simulation as described in claim 6, characterized in that, Based on the non-constant receding water curve and receding water morphology coefficient, combined with the real-time monitoring of the current water storage status and physical regulation capacity of ponds and wetlands at all levels, the storage capacity is dynamically allocated on a time-by-time basis, generating a storage operation plan that includes the timing of gate pump opening and closing and water storage depth, including: Based on the pre-determined water level-storage capacity relationship curves of each regulation and storage facility, and combined with the real-time water level data of ponds and wetlands at all levels, the remaining available regulation and storage capacity of each facility at the current moment is calculated, and the real-time status dataset of the regulation and storage facility group is obtained. Based on the non-constant water discharge curve, the predicted water discharge volume for each time period within the scheduling period corresponding to the storage capacity to be allocated is extracted. The predicted water discharge volume is then nonlinearly corrected by the water discharge morphology coefficient to obtain a dynamic water discharge load prediction sequence for each time period. Using the dynamic water discharge load prediction sequence as input and the real-time status dataset of the water storage facility group as initial conditions, the water storage capacity of ponds and wetlands at each level is determined in time period according to the multi-objective optimization allocation principle, so as to obtain a water storage allocation scheme that meets the flood control constraints and pollution reduction objectives of the polder area. Based on the reservoir capacity allocation scheme, a regulation and storage operation scheme is generated by combining the gate pump opening and closing characteristics and hydraulic transmission time of each regulation and storage facility. This scheme includes the gate pump opening and closing sequence of each facility and the target water storage depth.

8. The method for calculating the storage capacity of farmland runoff drainage in the Chaohu Lake basin based on AI runoff-generation coupling simulation according to claim 7, characterized in that, The water storage operation plan was fed back to the runoff-generated runoff coupling model for verification. The deviation of the simulated receding water peak was corrected online using real-time monitoring data. After closed-loop correction, an optimized water storage capacity configuration plan was obtained that meets the requirements of flood control and non-point source pollution reduction in the polder area, including: The aforementioned regulation and storage operation scheme is loaded into the constructed runoff-generation coupling model to drive the model to reenact the water receding process according to the gate pump opening and closing sequence and target water storage depth specified in the regulation and storage operation scheme, thereby obtaining the scheme simulated water receding curve; The simulated drainage curve is compared with the non-constant drainage curve to calculate the deviation between the simulated drainage curve and the non-constant drainage curve at the peak flow time and peak flow rate, thus obtaining the peak deviation vector that reflects the simulation accuracy. Based on the peak deviation vector, the key parameters of the runoff-runoff coupling model are corrected online in combination with the real-time monitoring data to update the physical property parameters of each grid in the model, so as to obtain the optimized runoff-runoff coupling model. The runoff-generation coupling model with optimized driving parameters is used to re-simulate the drainage process. The simulation of the drainage curve is compared with the non-constant drainage curve, and the model parameters are corrected online until the peak deviation vector converges within the preset accuracy threshold range. The simulation result that finally meets the deviation requirements is used as the optimized storage capacity configuration scheme.

9. A system for calculating the storage capacity of farmland runoff regulation in the Chaohu Lake basin based on AI runoff-generation coupling simulation, wherein the system implements the method as described in any one of claims 1 to 8, characterized in that, include: The data acquisition module is used to acquire underlying surface and hydrological data of farmland and multi-level ditch and wetland regulation and storage systems in the Chaohu Lake basin. The coupled model construction module is used to construct a runoff-runoff coupling model constrained by both AI algorithms and hydrophysical mechanisms, which is used to characterize the drainage process from field plots to ditches and wetlands on a grid-by-grid basis. The drainage process simulation module is used to input real-time soil moisture, water level and flow monitoring data into the runoff-gene coupling model to perform hourly dynamic simulation of the drainage process after heavy rain, and obtain a non-constant drainage curve that reflects the peak of farmland runoff and the drainage process. The feature point recognition module is used to dynamically identify multiple feature control points with precise spatiotemporal coordinates from non-constant receding water curves. The manifold construction and analysis module is used to reconstruct the phase space of the drainage time series with the feature control points as nodes, and construct a hydrodynamic manifold that represents the nonlinear evolution path of the drainage process; perform information geometry analysis on the hydrodynamic manifold, extract the norm of its geodesic curvature tensor and the spectral entropy of the Fisher information matrix, and fuse them to obtain the drainage morphology coefficients that represent the nonlinear evolution characteristics of farmland drainage. The regulation and storage scheme generation module is used to dynamically allocate the regulation and storage capacity in time periods based on the non-constant water receding curve and water receding morphology coefficient, combined with the real-time monitoring of the current water storage status and physical regulation and storage capacity of ponds and wetlands at all levels, and generate a regulation and storage operation scheme that includes the gate pump opening and closing sequence and water storage depth. The closed-loop correction module is used to feed back the regulation and storage operation plan to the runoff-generated runoff coupling model for verification. It corrects the deviation of the simulated receding water peak online through real-time monitoring data. After closed-loop correction, an optimized regulation and storage capacity configuration plan that meets the requirements of flood control and non-point source pollution reduction in the polder area is obtained.

Images