Intelligent dispatching method and system for cascade hydropower stations based on basin grading early warning

Abstract

This application provides a method and system for intelligent scheduling of cascade hydropower stations based on basin-level early warning, belonging to the field of optimization scheduling technology. The method includes: building a multi-process coupled model of the basin and outputting real-time forecast results; constructing a point-line-area level early warning map; constructing marginal flood control risk functions and marginal power generation benefit functions for each reservoir based on the point-line-area level early warning map and the current operating data of each reservoir; constructing an objective function based on the marginal flood control risk function and marginal power generation benefit function with the goal of maximizing the expected comprehensive efficiency of the cascade; solving for the spatial distribution scheme of cascade risk and benefit; dynamically replacing flood control risks among different reservoirs within a preset risk threshold range according to the spatial distribution scheme of cascade risk and benefit, generating a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir. This solves the technical problems of low efficiency and insufficient accuracy in the scheduling decisions of cascade hydropower stations in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of optimized scheduling technology, and in particular to a method and system for intelligent scheduling of cascade hydropower stations based on basin-level early warning. Background Technology

[0002] With the continuous expansion of the development scale of cascade hydropower stations in the basin, the demand for coordinated scheduling of cascade reservoir groups in terms of flood control safety, power generation efficiency, and ecological protection is becoming increasingly prominent.

[0003] However, traditional scheduling methods usually rely on the independent operating rules of a single reservoir or the experience and judgment of the schedulers, which makes it difficult to effectively cope with the challenges brought about by the uncertainty of water inflow in the basin, the conflict of multiple objectives, and the complex hydraulic coupling relationship between cascade reservoirs.

[0004] Meanwhile, existing early warning information release mechanisms are mostly based on single-point or single-dimensional data, lacking deep integration with watershed geospatial information, resulting in inaccurate risk perception and untimely information transmission, making it difficult to support refined scheduling decisions at the cascade level. Summary of the Invention

[0005] This invention addresses the technical problems in existing technologies, such as the complex hydraulic coupling and multi-objective conflicts in cascade hydropower station groups, the reliance on single reservoir experience rules for scheduling methods, the difficulty in quantifying flood control risks and power generation benefits, and the lack of spatial correlation in early warning information dissemination, which lead to low scheduling decision-making efficiency and insufficient accuracy. The invention provides an intelligent scheduling method and system for cascade hydropower stations based on basin-level early warning.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0007] In a first aspect, the present invention provides a method for intelligent scheduling of cascade hydropower stations based on basin-level early warning, including:

[0008] A watershed multi-process coupled model is constructed, which includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and real-time forecast results are obtained through the output of the watershed multi-process coupled model.

[0009] Based on the spatial topological relationship of the watershed, a three-level early warning topology structure of point-line-area is constructed, and a point-line-area graded early warning map is generated according to the real-time forecast results, the flood control limit water level of each reservoir and the safe discharge of the downstream river channel.

[0010] Based on the point-line-surface graded early warning map and the current operating data of each reservoir, construct the marginal flood control risk function and marginal power generation benefit function of each reservoir;

[0011] With the goal of maximizing the expected comprehensive efficiency of the cascade, an objective function is constructed based on the marginal flood control risk function and the marginal power generation benefit function, and the spatial distribution scheme of cascade risk and benefit is obtained by solving the problem.

[0012] Within a preset risk threshold range, flood control risks are dynamically replaced among different reservoirs based on the tiered risk benefit spatial distribution scheme, generating a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir.

[0013] Secondly, this invention provides an intelligent dispatching system for cascade hydropower stations based on basin-level early warning, comprising:

[0014] The real-time forecast module is used to build a watershed multi-process coupled model that includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and to obtain real-time forecast results through the output of the watershed multi-process coupled model.

[0015] The early warning map construction module is used to construct a three-level early warning topology structure of points, lines, and areas based on the spatial topology of the watershed, and to generate a point-line-area graded early warning map based on the real-time forecast results, the flood control limit water level of each reservoir, and the safe discharge of the downstream river channel.

[0016] The marginal function construction module is used to construct the marginal flood control risk function and marginal power generation benefit function of each reservoir based on the point-line-surface hierarchical early warning map and the current operating data of each reservoir.

[0017] The distribution scheme generation module is used to construct an objective function based on the marginal flood control risk function and the marginal power generation benefit function with the goal of maximizing the expected comprehensive efficiency of the cascade, and solve for the spatial distribution scheme of the cascade risk and benefit.

[0018] The scheduling scheme generation module is used to dynamically replace flood control risks among different reservoirs within a preset risk threshold range based on the cascade risk-benefit spatial distribution scheme, and generate a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir.

[0019] The beneficial effects of this invention are:

[0020] Compared to existing technologies, this application, by constructing a multi-process coupled model of the watershed, can output accurate real-time forecast results, providing a reliable data foundation for subsequent scheduling. Based on the spatial topology of the watershed, a three-level early warning topology structure of points, lines, and surfaces is constructed. Combined with real-time forecast results, flood control limit water levels of each reservoir, and safe discharge capacity of downstream channels, a point-line-surface graded early warning map is generated, achieving precise positioning and visualization of the risk situation. Based on the point-line-surface graded early warning map and the current operating data of each reservoir, marginal flood control risk functions and marginal power generation benefit functions for each reservoir are constructed, transforming the abstract flood control risk and power generation benefit into... Quantifiable marginal indicators lay the foundation for multi-objective collaborative optimization. With the goal of maximizing the expected comprehensive efficiency of the cascade, an objective function including the risk replacement rate is constructed based on the marginal flood control risk function and the marginal power generation benefit function, and the spatial distribution scheme of cascade risk and benefit is obtained by solving it. This realizes the unified quantification and global optimization of flood control risk and power generation benefit. Within the preset risk threshold range, the flood control risk is dynamically replaced among different reservoirs according to the spatial distribution scheme of cascade risk and benefit, generating a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir. The optimization results are transformed into executable and collaborative scheduling instructions.

[0021] Through the above technical solutions, this application constructs a set of intelligent scheduling methods for cascade hydropower stations that integrates accurate forecasting, spatial early warning, quantitative decision-making, and dynamic optimization. It effectively solves the technical problems of difficulty in coordinating risks and benefits, lack of spatial correlation of early warning information, and low decision-making efficiency and insufficient accuracy in traditional scheduling. Under the premise of ensuring absolute safety constraints, it maximizes the comprehensive efficiency of cascade hydropower stations. Attached Figure Description

[0022] Figure 1 A flowchart illustrating the intelligent scheduling method for cascade hydropower stations based on basin-level early warning provided by this invention;

[0023] Figure 2 This is a schematic diagram of the structure of the intelligent dispatching system for cascade hydropower stations based on basin-level early warning provided by the present invention.

[0024] In the attached diagram, the components represented by each number are as follows:

[0025] Real-time forecast module 11, early warning map construction module 12, marginal function construction module 13, distribution scheme generation module 14, scheduling scheme generation module 15. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0028] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0029] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for intelligent scheduling of cascade hydropower stations based on basin-level early warning, including:

[0030] S10: Construct a watershed multi-process coupled model that includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and obtain real-time forecast results through the output of the watershed multi-process coupled model.

[0031] Traditional watershed hydrological forecasting often relies on a single meteorological forecasting model or an independent hydrological model, which has problems such as difficulty in quantifying meteorological forecast uncertainty, sensitivity of hydrological models to rainfall input errors, and insufficient forecast accuracy due to the separation of meteorological and hydrological processes.

[0032] To address the aforementioned issues, this application constructs a watershed multi-process coupled model that includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and obtains real-time forecast results through the output of the watershed multi-process coupled model.

[0033] Specifically, step S10 in the method includes:

[0034] Acquire geospatial data, historical hydrological and meteorological data, and reservoir operating data for the target watershed. The reservoir operating data includes at least the reservoir capacity curve data, discharge capacity data, and unit output characteristic data for each reservoir.

[0035] The geospatial data, the historical hydrological and meteorological data, and the reservoir operating data are input into a pre-trained watershed multi-process coupled model for coupled calculation, and the output is rainfall forecast data and flood forecast data, wherein the flood forecast data includes flood flow data of each cross section;

[0036] The rainfall forecast data and the flood forecast data are output as the real-time forecast results.

[0037] In this embodiment of the application, geospatial data, historical hydrological and meteorological data, and reservoir operating data of the target watershed are first acquired. Specifically, the geospatial data is used to characterize the topography, river features, and engineering distribution of the target watershed, and may include digital elevation model (DEM) data within the watershed (e.g., using DEM data with a resolution of 30m×30m, which can accurately reflect the topographic relief, confluence path, and elevation distribution of the watershed), river cross-section data (e.g., shape parameters, cross-sectional area, riverbed roughness, etc. of each control section of the river), and the location coordinates of the reservoir dam, etc.

[0038] Specifically, historical hydrological and meteorological data can include long-series hourly observation data from the past 10 years or more, including rainfall data, flow data, water level data, etc. Long-series data can effectively cover the characteristics of different hydrological years and different flood events.

[0039] Specifically, reservoir operating data is used to characterize the operational characteristics and constraints of each reservoir. This data can include reservoir capacity curves, discharge capacity data, and generator output characteristic data. For example, reservoir capacity curves can show the water level-capacity relationship (e.g., when the water level is 300.0m, the corresponding reservoir capacity is 1.2 × 10⁻⁶ m). 8 m³; when the water level is 305.0m, the corresponding reservoir capacity is 1.5×10 m³. 8 The data on the reservoir's capacity is presented as follows: The discharge capacity data is the relationship between gate opening and discharge rate (e.g., when the gate opening is 50%, the discharge rate is 80 m³ / s; when the gate opening is 100%, the discharge rate is 150 m³ / s), used to determine the discharge capacity of the reservoir under different flood discharge conditions; The unit output characteristic data is presented as a head-output-flow relationship table (e.g., when the head is 80 m and the flow rate is 50 m³ / s, the unit output is 38000 kW), used to quantify the impact of the reservoir's water storage status on power generation efficiency.

[0040] The aforementioned geospatial data, historical hydrological and meteorological data, and reservoir operational data can all be obtained from the reservoir management department, hydrological and water resources monitoring department, or relevant water conservancy project management unit corresponding to the target watershed.

[0041] Secondly, geospatial data, historical hydrological and meteorological data, and reservoir operating data are input into a pre-trained multi-process coupled model of the watershed for coupled calculations, outputting rainfall forecast data and flood forecast data. The flood forecast data includes flood discharge data at various cross-sections. The pre-trained multi-process coupled model of the watershed has learned the meteorological-hydrological response patterns of the watershed through historical data. After inputting current geospatial data, historical hydrological and meteorological data, and reservoir operating data, coupled calculations can output rainfall forecast data and flood forecast data for a future period (e.g., 24 hours, 72 hours, etc.). The flood forecast data includes flood discharge data at key cross-sections, such as reservoir inflow cross-sections and downstream flood control cross-sections.

[0042] Finally, the rainfall forecast data and flood forecast data are output as real-time forecast results. For example, the real-time forecast results may include hourly rainfall forecast data for the next 72 hours and hourly flood discharge data for each cross section for the next 72 hours, and are output in time series format.

[0043] Furthermore, the construction process of the watershed multi-process coupling model includes:

[0044] Historical forecast data from multiple numerical weather prediction models within the target watershed, corresponding historical measured rainfall data, and historical inflow data from various reservoir sections were collected to form a training sample set.

[0045] An initial watershed multi-process coupled model is constructed, wherein the initial watershed multi-process coupled model includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule;

[0046] The historical forecast data in the training sample set is input into the initial watershed multi-process coupled model. The historical forecast data of multiple numerical weather prediction models are integrated and processed by the meteorological ensemble forecast submodule to generate rainfall forecast scenario data. The rainfall forecast scenario data is then converted into predicted inflow process line data and predicted cross-sectional flood flow data by the hydrological evolution simulation submodule.

[0047] With the goal of minimizing the error between the predicted inflow process data and the corresponding historical inflow data in the training sample set, supervised learning is used to train the initial watershed multi-process coupling model. The network parameters inside the meteorological ensemble forecast submodule and the hydrological evolution simulation submodule are simultaneously optimized and adjusted through the backpropagation algorithm.

[0048] When the error converges to within a preset threshold range, training stops, and the trained multi-process coupling model of the watershed is obtained.

[0049] In this embodiment, historical forecast data from multiple numerical weather prediction models within the target watershed, corresponding historical measured rainfall data, and historical inflow data from each reservoir section are first collected to form a training sample set. Specifically, the multiple numerical weather prediction models can be the European Centre for Medium-Range Weather Forecasts (ECMWF), the US Global Forecast System (GFS), the China Meteorological Administration Global Assimilation Forecast System (CMA-GFS), etc. Historical forecast data from multiple numerical weather prediction models, such as twice-daily forecast data from the past 5 years, are collected. At the same time, historical measured rainfall data for the corresponding time period and historical inflow data from each reservoir section are collected from ground rain gauges, hydrological stations, etc. Then, the three are spatiotemporally matched: for each historical forecast data point, historical measured rainfall data and historical inflow data from each reservoir section at the corresponding location and time period are taken to form a sample. Finally, a large number of samples form the training sample set.

[0050] Secondly, an initial watershed multi-process coupled model is constructed, which includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule. For example, the initial watershed multi-process coupled model can be constructed using the following technical approach: construct the meteorological ensemble forecast submodule and the hydrological evolution simulation submodule separately, and then couple and integrate them to obtain the initial watershed multi-process coupled model.

[0051] For example, the meteorological ensemble forecast submodule can be constructed using an LSTM network with an attention mechanism, mainly consisting of an input layer, an LSTM layer, an attention mechanism layer, and a fully connected output layer. The input layer receives historical forecast data from multiple numerical weather prediction models, with each model's data serving as an independent feature channel input. The LSTM layer extracts time-series features; it can be configured with two LSTM layers, each with 128 or 256 hidden units, using the tanh activation function. Information flow is controlled through forget gates, input gates, and output gates to capture long-term dependencies in rainfall forecasts. The attention mechanism layer can employ a multi-head self-attention mechanism, with four or eight attention heads, each with a 64-dimensional dimension. By weighted aggregation of features from different time steps and models, it automatically focuses on historical information that contributes more to the current forecast, improving sensitivity to extreme rainfall events. The fully connected output layer contains 2-3 fully connected network layers, with the number of neurons decreasing layer by layer (e.g., 256→128→1), using the ReLU activation function, and finally outputs the areal rainfall forecast sequence for future periods.

[0052] For example, the hydrological evolution simulation submodule can be constructed using a Physical Information-Based Neural Network (PINN), mainly composed of an encoder-decoder architecture and a physical constraint loss function. The encoder part can employ a stacked LSTM network or a temporal convolutional network to map rainfall forecast scenario data into high-dimensional hidden states, extracting the spatiotemporal features of the hydrological response. A two-layer bidirectional LSTM with 256 hidden units is used. The decoder part employs a fully connected network or an attention decoder to decode the hidden states into flow process lines for each cross-section. The output layer uses a linear activation function to maintain the physical range of the flow values. The physical constraint embedding method involves adding residual terms from the Saint-Venant equations or the water balance equations to the loss function. Automatic differentiation is used to calculate the derivatives of the predicted flow with respect to time and space, forcing the network output to satisfy physical laws.

[0053] Next, historical forecast data from the training sample set is input into the initial watershed multi-process coupled model. The meteorological ensemble forecast submodule integrates historical forecast data from multiple numerical weather prediction models to generate rainfall forecast scenario data. The hydrological evolution simulation submodule then transforms the rainfall forecast scenario data into predicted inflow process curve data and predicted cross-sectional flood discharge data. Specifically, for each training sample in the training sample set, historical forecast data from multiple models are input into the initial watershed multi-process coupled model. The meteorological ensemble forecast submodule outputs integrated rainfall forecast scenario data, such as 72-hour hourly areal rainfall. Then, the rainfall forecast scenario data is input into the hydrological evolution simulation submodule, which outputs predicted inflow process curve data and predicted cross-sectional flood discharge data, such as 72-hour hourly discharge and flood discharge at each cross-section.

[0054] Furthermore, with the goal of minimizing the error between the predicted inflow process data and the corresponding historical inflow data in the training sample set, supervised learning is used to train the initial watershed multi-process coupled model. The network parameters within the meteorological ensemble forecast submodule and the hydrological evolution simulation submodule are simultaneously optimized and adjusted using the backpropagation algorithm. For example, the initial watershed multi-process coupled model can be trained using the following technical approach: 1. Data partitioning: The training sample set is randomly divided into a training set and a validation set in an 8:2 ratio. The training set is used for optimizing and updating model parameters, while the validation set is used to monitor overfitting during the training process and adjust hyperparameters.

[0055] 2. Defining the Loss Function: Since the hydrological evolution simulation submodule adopts a neural network architecture based on physical information, the loss function consists of two parts: a data term and a physical term. The data term uses mean squared error or mean absolute error to measure the fitting accuracy between the predicted inflow process curve data and the corresponding historical inflow data in the training sample set. The physical term is based on the residuals of the Saint-Venant equations or the water balance equations. It automatically differentiates and calculates the derivatives of the predicted inflow process curve data with respect to time and space, forcing the network output to satisfy hydrological physical laws. For example, the loss function can be expressed as: L = L data +λ×L physics L data To determine the mean square error between the predicted flow rate and the measured flow rate, L physics λ is the mean square value of the residuals of the Saint-Venant equations, and λ is the balance coefficient used to adjust the strength of the physical constraints. It can be optimized based on the performance of the validation set. For better results, λ can be set to 0.1.

[0056] 3. Training Parameter Settings: The Adam optimizer was used, with an initial learning rate of 0.001 and a batch size of 32. To balance data fitting accuracy and physical consistency, a phased training strategy was adopted: The first phase focused on pre-training with data terms, with 50 training epochs, allowing the model to learn the basic mapping relationship from input to output. The second phase introduced physical terms for joint training, with 100 training epochs, ensuring the model satisfies physical laws while fitting the data. Throughout the training process, the learning rate adopted a cosine decay strategy, gradually decreasing from the initial value of 0.001 to 0.0001.

[0057] 4. Perform backpropagation optimization: In each training batch, historical forecast data from the training sample set are input into the initial watershed multi-process coupled model. Forward propagation is used to calculate the predicted inflow process line data. Then, the loss function value is calculated. The gradient of the loss function with respect to each network parameter is calculated through the backpropagation algorithm. The weight parameters inside the meteorological ensemble forecast submodule and the hydrological evolution simulation submodule are updated synchronously. The above process is repeated until the loss function converges or the preset number of training rounds is reached, and the trained watershed multi-process coupled model is obtained.

[0058] Finally, when the error converges to within a preset threshold range, training stops, resulting in a trained multi-process coupled model of the watershed. The preset threshold range can be flexibly set by those skilled in the art based on actual application requirements. For example, based on the accuracy requirements of watershed flood forecasting, the root mean square error on the validation set can be set below a certain empirical value, such as 5%–10% of the historical maximum flow, as the convergence threshold; or an early stopping strategy can be adopted, where convergence is considered achieved when the validation set error no longer decreases after several consecutive training cycles.

[0059] For example, during the training process, the loss function value on the validation set is monitored periodically. Once it falls below a preset threshold or there is no significant improvement for 20 consecutive cycles, the iteration is stopped, the current model structure and parameters are saved, and thus the trained multi-process coupling model of the watershed is obtained.

[0060] In summary, compared to existing technologies, this application constructs a watershed multi-process coupled model that includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and obtains real-time forecast results through the output of the watershed multi-process coupled model. This effectively improves the accuracy and reliability of rainfall and flood forecasts, providing precise input data support for subsequent tiered early warning and optimized scheduling.

[0061] S20: Construct a three-level early warning topology structure of points, lines, and surfaces based on the spatial topological relationship of the watershed, and generate a point-line-surface graded early warning map based on the real-time forecast results, the flood control limit water level of each reservoir, and the safe discharge of the downstream river channel.

[0062] Traditional watershed early warning information is mostly released based on a single point or a single dimension, such as only releasing water level data of a certain reservoir or flow data of a certain section. It lacks a hierarchical visualization and targeted release mechanism that is linked with the geographical space of the watershed and the scheduling objectives, resulting in information overload or the loss of key information, making it difficult to support the spatial coordinated scheduling decision-making of cascade hydropower station groups.

[0063] To address the aforementioned issues, this application constructs a three-tiered early warning topology based on the spatial topological relationships of the watershed, consisting of points, lines, and areas. It then generates a tiered early warning map based on the real-time forecast results, the flood control limits of each reservoir, and the safe discharge capacity of downstream rivers. This spatialized and hierarchical early warning method transforms abstract risk data into an intuitive perception of the watershed's risk situation, providing precise spatiotemporal positioning for subsequent marginal risk quantification analysis and reservoir capacity risk replacement.

[0064] Specifically, step S20 in the method includes:

[0065] Integrate spatial location data of main and tributary rivers, reservoirs and dams, control sections and meteorological and hydrological stations within the basin area, and establish a three-level early warning topology structure of point engineering stations, line river scheduling paths and area basin.

[0066] Rainfall forecast data, flood forecast data, and flood flow data at each cross section are obtained from the real-time forecast results. The rainfall forecast data and the flood forecast data are then mapped to the corresponding point elements, line elements, and area elements in the point-line-area three-level early warning topology according to their spatial locations.

[0067] Based on the flood flow data of each section, combined with the flood control limit water level of each reservoir and the safe discharge of the downstream river channel, the point-level risk level of exceeding the threshold, the line-level risk level of exceeding the threshold of each river section, and the surface-level risk level of exceeding the threshold of each regional basin are calculated.

[0068] The point-level risk level exceeding the threshold is associated with the corresponding point feature, the point-level risk level exceeding the threshold is associated with the corresponding line feature, and the area-level risk level exceeding the threshold is associated with the corresponding area feature. Combined with the rainfall forecast data and the flood forecast data, a point-line-area graded early warning map is generated.

[0069] In this embodiment, spatial location data of main and tributary rivers, reservoirs and dams, control sections, and meteorological and hydrological stations within the watershed are first integrated to establish a three-level early warning topology structure of point-line-area, consisting of point engineering stations, line river scheduling paths, and area watersheds. Specifically, the three-level early warning topology structure includes point elements, line elements, and area elements, corresponding to point engineering stations, line river scheduling paths, and area watersheds, respectively. For example, geographic information system (GIS) technology can be used to integrate various spatial data within a watershed: 1. Establish a point element layer: The point element layer corresponds to point engineering stations, containing unique identifiers and latitude and longitude coordinates for each reservoir dam, meteorological and hydrological station, and control section; 2. Establish a line element layer: The line element layer corresponds to the line river scheduling path, containing the starting point coordinates, ending point coordinates, river length, and river roughness attribute data of the main and tributary channels, and recording the upstream and downstream topological relationships between each river segment; 3. Establish a surface element layer: The surface element layer corresponds to the surface watershed, containing the boundary coordinate data, watershed area, and average slope attribute data of each sub-watershed, and recording the affiliation relationship between each sub-watershed and the corresponding river segment; 4. Establish a point-line association table and a line-surface association table, wherein the point-line association table is used to record the river segment identifier to which each point element belongs, and the line-surface association table is used to record the sub-watershed identifier to which each line element belongs, thereby forming a three-level early warning topology structure consisting of point engineering stations, line river scheduling paths, and surface watersheds.

[0070] Secondly, rainfall forecast data, flood forecast data, and flood discharge data at various cross-sections are obtained from real-time forecast results. These data are then mapped spatially to the corresponding point, line, and area elements in the point-line-area three-level early warning topology. Specifically, rainfall forecast data, flood forecast data, and flood discharge data at various cross-sections are extracted from real-time forecast results. Then, using the established point-line-area three-level early warning topology, the data is mapped spatially: for point elements, the flood discharge data of the corresponding reservoir dam, hydrological station, or control section is directly assigned to the point; for line elements, the discharge data (or interpolated data) of each cross-section on the river segment is assigned to the river segment, and segmented mapping can be performed according to the river segment length; for area elements, the areal rainfall forecast data of the sub-basin is assigned to the sub-basin.

[0071] For example, if the cross-sectional flood discharge data of a reservoir section is a predicted flow of 1200 m³ / s, then this data is mapped to the point feature corresponding to the reservoir; if the cross-sectional flood discharge data of a downstream control section of a river segment is 1500 m³ / s, then this data is mapped to the line feature of that river segment; if the predicted rainfall data of a sub-basin is 80 mm, then this data is mapped to the area feature of that sub-basin. Through this spatial mapping, each point, line, and area feature is associated with the forecast information in the corresponding real-time forecast results.

[0072] Secondly, based on the flood discharge data at each cross-section, combined with the flood control limit water level of each reservoir and the safe discharge of the downstream river channel, the point-level risk level exceeding the threshold, the line-level risk level exceeding the threshold for each river section, and the area-level risk level exceeding the threshold for each regional basin are calculated. Among them, the risk level exceeding the threshold refers to the degree to which the flood discharge data at each cross-section exceeds the safe threshold, representing the urgency of approaching the safety red line.

[0073] Specifically, for the point level, the focus is on either reservoir water level or cross-sectional flow. If considering reservoir water level, the calculation can be combined with the flood control limit water level; if considering cross-sectional flow, the calculation can be directly based on the flow rate. For example, for a specific reservoir cross-section, the point-level exceedance risk level can be defined as: Point-level exceedance risk level = (Q...) _current -Q _limit ) / (Q _max -Q _limit） , where Q _current Q represents the current or forecasted flow rate. _limit To control flooding, the discharge volume corresponding to the water level is Q. _max This is to verify the maximum discharge capacity corresponding to the flood level or the maximum safe discharge capacity of the downstream river channel.

[0074] Specifically, for the line-level, the degree of risk exceeding the threshold can be calculated based on the flow rate at the most dangerous section within the river segment. For example: Line-level risk exceeding the threshold = (Q_segment -Q _safe ) / (Q _max -Q _safe） , where Q _segment Q represents the maximum predicted cross-sectional flow within the river section. _safe Q is the safe discharge capacity for this river section. _max This represents the maximum safe discharge capacity for the river section.

[0075] Specifically, for the surface level, the degree of risk exceeding the threshold can be calculated based on the areal rainfall. For example, the degree of risk exceeding the threshold at the surface level = (P _area -P _threshold ) / (P _extreme -P _threshold） , where P _area To forecast areal rainfall, P _threshold P is the threshold rainfall amount required to cause a disaster (which can be determined based on historical disaster data). _extreme This represents the historical maximum rainfall or the design rainfall value.

[0076] It should be noted that all threshold parameters in the above calculations can be obtained from engineering archives or hydrological manuals. For example, if the safe discharge of a certain river section is 2000 m³ / s, the ultimate safe discharge is 3000 m³ / s, and the predicted maximum cross-sectional flow is 2500 m³ / s, then the risk level of exceeding the threshold at the line level = (2500-2000) / (3000-2000) = 0.5, indicating that the river section has exceeded the 50% margin of the safe discharge.

[0077] Finally, the risk levels exceeding the threshold at the point level are associated with the corresponding point features, the risk levels at the point level with the corresponding line features, and the risk levels exceeding the threshold at the area level with the corresponding area features. These are then combined with rainfall forecast data and flood forecast data to generate a point-line-area tiered early warning map. Specifically, the calculated risk levels exceeding the threshold at the point level are assigned to the corresponding point features, the risk levels exceeding the threshold at the point level are assigned to the corresponding line features, and the risk levels exceeding the threshold at the area level are assigned to the corresponding area features. These spatial features with risk level attributes are then overlaid and integrated with rainfall forecast data and flood forecast data to form a point-line-area tiered early warning map.

[0078] Optionally, the point-line-area graded early warning map can be presented in the form of a geographic information system map, using different colors to represent risk levels: for example, green for risk levels of 0–0.2 (low risk), yellow for 0.2–0.5 (medium risk), orange for 0.5–0.8 (high risk), and red for 0.8–1.0 (extremely high risk). Each point, line, and area in the point-line-area graded early warning map carries a risk level label, and users can click to view detailed forecast data, such as flow processes and rainfall processes.

[0079] For example, in the generated point-line-area hierarchical early warning map, a reservoir point is displayed in orange, and hovering the mouse displays "Point risk 0.65, current flow 2600 m³ / s, flood control limit water level corresponding flow 2000 m³ / s". In this way, the point-line-area hierarchical early warning map can provide dispatchers with an intuitive understanding of the watershed risk situation, supporting subsequent construction of marginal functions and optimized scheduling.

[0080] In summary, compared to existing technologies, this application constructs a three-tiered early warning topology based on the spatial topology of the watershed, namely point-line-area, and generates a tiered early warning map based on the real-time forecast results, the flood control limit water levels of each reservoir, and the safe discharge of downstream channels. This deeply integrates the dispersed real-time forecast results with the spatial location of the watershed, achieving precise positioning and visualization of the risk situation, and providing an intuitive decision-making basis for the intelligent scheduling of cascade reservoirs.

[0081] S30: Based on the point-line-surface graded early warning map and the current operating data of each reservoir, construct the marginal flood control risk function and marginal power generation benefit function of each reservoir.

[0082] In traditional cascade hydropower station scheduling, flood control risks and power generation benefits are mostly assessed qualitatively or calculated using fixed coefficients. This makes it difficult to accurately and quantitatively characterize the risk and benefit increments brought about by reservoir impoundment and discharge actions. Furthermore, it fails to combine the spatial coupling calculations with the three-level early warning results of the entire basin (point-line-area). Consequently, scheduling decisions cannot accurately weigh the marginal risk costs and marginal power generation benefits of individual reservoirs, making it difficult to achieve the global optimal match between flood control safety and power generation benefits.

[0083] To address the aforementioned issues, this application constructs marginal flood control risk functions and marginal power generation benefit functions for each reservoir based on the aforementioned point-line-area hierarchical early warning map and the current operating data of each reservoir.

[0084] Specifically, step S30 in the method includes:

[0085] Extract the point-level over-threshold risk level of each reservoir's location from the point-line-area hierarchical early warning map, as well as the line-level and area-level over-threshold risk levels associated with each reservoir's downstream location.

[0086] Extract the current water level data, available flood control capacity data, inflow data, and downstream river safe flow data of each reservoir from the current operating data of each reservoir;

[0087] Set absolute safety constraints, including that the water level of each reservoir must not exceed the check flood level and the discharge flow of each reservoir must not exceed the ultimate safe discharge capacity of the downstream river channel;

[0088] Within the scope of the absolute safety constraints, for each reservoir, based on the point-level, line-level, and surface-level risk levels exceeding the threshold, as well as the current water level data, a function for the change in the risk level exceeding the threshold caused by a change in unit water storage is constructed, and this function is used as the marginal flood control risk function.

[0089] Based on the unit output characteristic data of each reservoir, the head-output-flow relationship curve of each reservoir is derived. Combining the upstream and downstream relationship of each reservoir in the cascade, for each reservoir, based on the inflow data and the current water level data, a mapping relationship between the change in total power generation revenue of the reservoir and the downstream cascade caused by the change in unit water storage is constructed as the marginal power generation benefit function.

[0090] In this embodiment, the point-level threshold risk level of each reservoir's location, as well as the line-level and area-level threshold risk levels associated with the downstream of each reservoir, are first extracted from the point-line-area hierarchical early warning map. Specifically, for each reservoir in the cascade hydropower station, multi-dimensional risk information related to it is extracted from the point-line-area hierarchical early warning map: First, the point-level threshold risk level corresponding to the reservoir's location is extracted, denoted as R_point_i, which reflects the current threshold risk level faced by the reservoir itself; second, the line elements associated with the downstream river channel of the reservoir are identified, i.e., the downstream river sections directly affected by the reservoir's discharge, and the line-level threshold risk level of these river sections is extracted, with the maximum or average value taken as the downstream line risk R_line_i of the reservoir; then, the area elements associated with the downstream region of the reservoir are identified, i.e., the sub-basins or inundation areas that may be affected by the reservoir's discharge, and the area-level threshold risk level of these area elements is extracted, with the maximum or average value also taken as the downstream area risk R_area_i.

[0091] For example, for a reservoir i, the risk level exceeding the threshold at the point level is extracted from the point-line-area hierarchical early warning map, with R_point_i=0.6. The risk level exceeding the threshold at the line level associated with the downstream is R_line_i=0.4, and the risk level exceeding the threshold at the area level associated with the downstream is R_area_i=0.3. The risk levels of these three dimensions together constitute the comprehensive risk characteristics currently faced by the reservoir.

[0092] Secondly, extract the current water level data, available flood control capacity data, inflow data, and downstream river safe flow data for each reservoir from the current operating data of each reservoir. Specifically, the current operating data is the operational status data obtained through the reservoir real-time monitoring system. It is necessary to extract the current water level data, available flood control capacity data, inflow data, and downstream river safe flow data for each reservoir from this data: the current water level data is the real-time measurement of the upstream water level, usually in meters, and can be directly read from the water level gauge; the available flood control capacity data refers to the storage capacity between the current water level and the flood control limit level that can still be used to impound floodwaters; the inflow data is the amount of water flowing into the reservoir per unit time, which can be obtained through actual measurement at the upstream hydrological station, in cubic meters per second; the downstream river safe flow data refers to the maximum flow rate that the downstream river can safely pass through without overflowing or flooding. This value is an engineering design parameter and can be obtained from the reservoir operation regulations.

[0093] For example, from the current operating data of a reservoir, the current water level data is extracted as 245m; the reservoir capacity corresponding to the flood control limit water level is 1 billion m³, the current reservoir capacity is 800 million m³, then the available flood control capacity data is 200 million m³; the inflow data is 800 m³ / s; and the downstream river safe flow data is 2000 m³ / s.

[0094] Secondly, absolute safety constraints are established. These constraints include ensuring that the water level of each reservoir does not exceed the check flood level and that the discharge flow from each reservoir does not exceed the downstream channel's ultimate safe discharge capacity. Specifically, these absolute safety constraints are inviolable bottom lines in the operation of cascade hydropower stations, and any operation plan must strictly adhere to them. The check flood level is the highest water level a reservoir is allowed to reach when encountering a check flood; exceeding this level will threaten the dam's structural safety. This information can be obtained from the reservoir's design data. The downstream channel's ultimate safe discharge capacity refers to the maximum flow rate that the downstream channel can withstand without breaching the dikes. It is usually slightly larger than the safe discharge capacity, leaving a small margin for overload, but exceeding this capacity poses an extremely high risk. These absolute safety constraints collectively define the safety boundaries of reservoir operation: ensuring that the water level of each reservoir does not exceed the check flood level ensures dam safety, and ensuring that the discharge flow from each reservoir does not exceed the downstream channel's ultimate safe discharge capacity ensures downstream channel safety.

[0095] It should be noted that in the subsequent construction and optimization of the marginal function, the absolute safety constraint will be a prerequisite that must be met, and any state or decision that exceeds the constraint will be deemed invalid.

[0096] Furthermore, within the constraints of absolute safety, for each reservoir, based on the risk levels at the point, line, and area levels, and current water level data, a function representing the change in risk level caused by a unit change in water storage is constructed. This function serves as the marginal flood control risk function. Specifically, the marginal flood control risk function describes the change in risk level caused by each additional unit of water storage ΔV under the current conditions. Since the risk level is a function of water storage V and is influenced by the combined effects of point, line, and area risks, a mapping relationship between the risk level and water storage V is established, and then the derivative is used to obtain the marginal flood control risk function.

[0097] For example, firstly, the comprehensive risk level of a reservoir is defined as the sum of the point-level, line-level, and area-level exceedance risk levels, i.e., comprehensive risk level = point-level exceedance risk level + line-level exceedance risk level + area-level exceedance risk level. Secondly, the relationship between the comprehensive risk level and the water storage capacity V is established: the comprehensive risk level usually increases rapidly with increasing V, and can be fitted as a power function: comprehensive risk level = a·(V / V_limit)^b, where V_limit is the reservoir capacity corresponding to the flood control limit water level, which is a known design parameter; a is a proportionality coefficient, reflecting the risk baseline level; and b is an exponential coefficient, reflecting the sensitivity of risk to the increase of water storage. Parameters a and b can be obtained by fitting historical data, for example, by collecting point-level, line-level, and area-level exceedance risk levels corresponding to different water storage capacities during multiple floods, and using the least squares method for regression analysis. Finally, the derivative of the comprehensive risk level is used to obtain the marginal flood control risk function.

[0098] Finally, based on the unit output characteristic data of each reservoir, the head-output-flow relationship curves of each reservoir were derived. Combining the upstream and downstream relationships of each reservoir in the cascade, for each reservoir, based on the inflow data and current water level data, a mapping relationship was constructed between the change in total power generation revenue of the reservoir and downstream cascade caused by a change in unit water storage, serving as the marginal power generation benefit function. Specifically, the marginal power generation benefit function describes how much increase in power generation revenue is brought about by adding one unit of water storage (e.g., storing an additional 10,000 cubic meters of water) in the current state of the reservoir. This requires comprehensive consideration of the chain reaction of the reservoir and downstream reservoirs: First, the head-output-flow relationship is obtained based on the unit output characteristic data. The power output formula of the hydropower station is: Output (kW) = 9.81 × Unit Efficiency × Power Generation Flow (m³ / s) × Head (m), where the head is the difference between the upstream and downstream water levels, which can be obtained from the current water level through the reservoir capacity curve; the unit efficiency is a constant.

[0099] Specifically, under a given water head, power output is directly proportional to the power generation flow. When a reservoir stores 1 cubic meter more water, it means releasing 1 cubic meter less water. This has two effects: First, there is a loss in power generation during the current period—releasing less water means a reduction in the power generation flow during the current period, resulting in a loss of power generation. This loss can be estimated using the power generation capacity per unit volume of water under the current water head. For example, if the current water head is 80 meters and the unit efficiency is 0.85, then the electrical energy generated per cubic meter of water is: 9.81 × 0.85 × 80 × (1 / 1000) ≈ 0.667 kWh. If the electricity price is 0.4 yuan / kWh, then the power generation value per cubic meter of water is approximately 0.267 yuan. Therefore, releasing 1 cubic meter less water results in a loss of 0.267 yuan. Second, the future power generation gain – the increased water storage will raise the reservoir level, thereby increasing the head of the reservoir in the future. With a higher head, the same flow of water can generate more electricity. This gain requires estimating the time during which the water can be used for power generation. Assuming that this water will be used for power generation in the future, and the average head increases by ΔH, then the gain can be estimated by multiplying ΔH by the total future power generation.

[0100] For example, if an upstream reservoir currently has a head of 80 meters and an electricity price of 0.4 yuan / kWh, the current power generation value per cubic meter of water is calculated to be 0.267 yuan. Considering the head increase benefit, assuming an empirical coefficient of 0.15, the future gain is 0.267 × 0.15 ≈ 0.04 yuan. Considering the downstream impact, assuming a downstream loss coefficient of 0.3, the downstream loss is 0.267 × 0.3 ≈ 0.08 yuan. Therefore, the total marginal power generation benefit = current loss (negative) + future gain (positive) - downstream loss = -0.267 + 0.04 - 0.08 = -0.307 yuan / cubic meter. This negative value indicates that under the current conditions, increasing water storage would actually reduce the total power generation revenue of the cascade, and power generation should be prioritized over water storage. If the calculated value is positive, it indicates that water storage is more beneficial. In this way, a simple marginal power generation benefit function can be constructed, which dynamically adjusts with changes in water level.

[0101] In summary, compared to existing technologies, this application constructs marginal flood control risk functions and marginal power generation benefit functions for each reservoir based on the aforementioned point-line-area hierarchical early warning map and the current operating data of each reservoir. In this way, the point-line-area hierarchical early warning results can be combined with the real-time operating conditions of the reservoirs, achieving a refined, quantitative, and marginal characterization of flood control risk and power generation benefits.

[0102] S40: With the goal of maximizing the expected comprehensive efficiency of the cascade, an objective function is constructed based on the marginal flood control risk function and the marginal power generation benefit function, and the spatial distribution scheme of cascade risk and benefit is obtained by solving the problem.

[0103] When faced with conflicts between multiple objectives such as flood control and power generation, traditional cascade hydropower station scheduling often relies on empirical rules or single-objective optimization. It is difficult to quantitatively assess the marginal substitution relationship between risks and benefits while ensuring absolute safety, resulting in scheduling decisions lacking global optimality and economic rationality.

[0104] To address the aforementioned issues, this application aims to maximize the expected comprehensive effectiveness of the cascade system. It constructs an objective function based on the marginal flood control risk function and the marginal power generation benefit function, and solves for the spatial distribution scheme of the cascade risk and benefit. The marginal flood control risk function and marginal power generation benefit function of each reservoir can be incorporated into a unified optimization system. Under multi-dimensional constraints, the risk quota to be borne by each reservoir and the corresponding storage and release strategies can be obtained.

[0105] Specifically, step S40 in the method includes:

[0106] Based on the marginal flood control risk function and marginal power generation benefit function of each reservoir, an objective function is constructed with the goal of maximizing the expected comprehensive efficiency of the cascade.

[0107] The constraints required to solve the objective function are set, including at least absolute safety constraints, cascade hydraulic coupling constraints, and reservoir operating condition constraints. The absolute safety constraints include that the water level of each reservoir must not exceed the check flood level and the discharge flow of each reservoir must not exceed the downstream channel's limit safe discharge. The cascade hydraulic coupling constraints include water balance constraints between adjacent reservoirs, flood wave propagation time delay constraints, and channel storage capacity constraints. The reservoir operating condition constraints include upper and lower limits of reservoir water level, discharge capacity constraints, upper and lower limits of power output constraints, and discharge flow variation constraints.

[0108] Under the common constraints of the above conditions, the objective function is solved to obtain the risk quota data that each reservoir should bear during the scheduling cycle, as well as the corresponding water storage and water release strategies.

[0109] The risk quota data, the water storage strategy, and the water release strategy are used as the spatial distribution scheme for the tiered risk-benefit profile.

[0110] In this embodiment, an objective function is first constructed based on the marginal flood control risk function and marginal power generation benefit function of each reservoir, aiming to maximize the expected comprehensive efficiency of the cascade hydropower station. The expected comprehensive efficiency of the cascade hydropower station refers to the difference between the sum of the power generation benefits of each reservoir in the cascade hydropower station and the sum of the safety margin losses, reflecting the overall balance between power generation revenue and flood control safety of the cascade hydropower station. Specifically, constructing the objective function requires integrating the marginal flood control risk function and marginal power generation benefit function of each reservoir to form a quantifiable mathematical expression.

[0111] Secondly, constraints are set for solving the objective function. These constraints include at least absolute safety constraints, cascade hydraulic coupling constraints, and constraints for each reservoir's operating conditions. The absolute safety constraints include that the water level of each reservoir must not exceed the check flood level and that the discharge flow from each reservoir must not exceed the downstream river's ultimate safe discharge capacity. The cascade hydraulic coupling constraints include water balance constraints between adjacent reservoirs, flood wave propagation time delay constraints, and river channel storage capacity constraints. The constraints for each reservoir's operating conditions include upper and lower limits for reservoir water level, discharge capacity constraints, upper and lower limits for power output constraints, and outflow variation constraints.

[0112] Specifically, solving the objective function must be done under the premise of satisfying a series of operating conditions. The absolute safety constraints are the bottom line and cannot be broken: the water level of each reservoir must not exceed the check flood level, and the discharge flow of each reservoir must not exceed the ultimate safe discharge capacity of the downstream river channel. The cascade hydraulic coupling constraints describe the hydraulic connection between adjacent reservoirs: the water balance constraint means that for any two adjacent reservoirs, the outflow from the upstream reservoir, after passing through the river channel, together with the inflow in the interval, constitutes the inflow to the downstream reservoir. This must satisfy the water conservation relationship, that is, the inflow to the downstream reservoir equals the outflow from the upstream reservoir after flood evolution plus the inflow in the interval. The flood wave propagation time delay constraint means that the flood wave released from the upstream reservoir needs a certain propagation time to reach the downstream reservoir. Therefore, there is a time delay between the inflow process of the downstream reservoir and the outflow process of the upstream reservoir. This time delay τ depends on the river channel length and the water flow velocity. The river channel storage capacity constraint means that the river channel has a certain regulation and storage capacity during the flood evolution process, which can temporarily store some water. However, there is an upper limit to the river channel storage capacity. Exceeding this upper limit may lead to overflow or levee breach. Therefore, it is necessary to limit the water storage in the river channel to not exceed its maximum allowable storage capacity.

[0113] Specifically, the operating constraints for each reservoir describe the operational limitations of a single reservoir: Reservoir water level upper and lower limits mean that the operating water level must not be lower than the dead water level to ensure basic needs such as power generation and ecology, and must not exceed the normal storage level or flood control limit level; discharge capacity constraints mean that the reservoir's discharge flow is limited by the maximum discharge capacity of the flood discharge facilities, which is usually related to the reservoir water level; power output upper and lower limits mean that the power generation output of the hydropower station is limited by the rated power of the units, and must not be lower than the minimum technical output and must not exceed the installed capacity; outflow variation constraints mean that the variation range of the outflow per unit time must be controlled within the allowable range to avoid sudden changes in downstream river levels that could cause secondary disasters or affect navigation safety.

[0114] Next, under the common constraints, the objective function is solved to obtain the risk quota data that each reservoir should bear during the scheduling cycle, as well as the corresponding water storage and discharge strategies. Specifically, mathematical programming algorithms or intelligent optimization algorithms can be used to solve this problem. During the solution process, the scheduling cycle is divided into several time periods, such as one hour per period. The water level or discharge rate of each reservoir in each time period is used as the decision variable, and the objective function value is maximized while satisfying all constraints.

[0115] For example, a sequential quadratic programming algorithm can be used to solve the problem. Sequential quadratic programming is an effective method for solving nonlinear constrained optimization problems. Its basic idea is to approximate the original nonlinear programming problem into a quadratic programming subproblem in each iteration step, and then approach the optimal solution through iteration. The specific calculation steps are as follows: 1. Divide the scheduling cycle into T time periods, such as T=72, corresponding to 72 hours. Let the decision variables be the water level or discharge of each reservoir in each time period, forming a decision vector. 2. Discretize the absolute safety constraints, the cascade hydraulic coupling constraints, and the operating conditions constraints of each reservoir, and express them in the form of equality constraints and inequality constraints. For example, the upper and lower limits of water level constraints are: dead water level ≤ reservoir operating water level ≤ normal storage water level or flood control limit water level. 3. Iterative solution using a sequential quadratic programming algorithm: At the current iteration point X_k, construct a quadratic programming subproblem to approximate the original problem. The objective function of the subproblem is the second-order Taylor expansion of the original objective function, and the constraints are linear approximations of the original constraints. Solve this quadratic programming subproblem to obtain the search direction d_k and determine the step size α_k. Update the iteration point X_{k+1} = X_k + α_k d_k. Repeat the above process until the convergence condition is met, such as the gradient norm being less than a threshold or the change in adjacent iteration points being less than a threshold, thus obtaining the optimal solution.

[0116] For example, the water storage strategy obtained by the solution is as follows: the water level of reservoir A rises from 250m to 251m in 0-6 hours, remains at 251m in 6-12 hours, and falls from 251m to 249.5m in 12-24 hours; the water discharge strategy is as follows: the discharge flow gradually increases from 100m³ / s to 120m³ / s in 0-6 hours, remains at 120m³ / s in 6-12 hours, and gradually decreases from 120m³ / s to 90m³ / s in 12-24 hours; the risk quota data is 0.6, indicating that the maximum comprehensive risk level that the reservoir can reach is 0.6.

[0117] Alternatively, in addition to sequential quadratic programming, intelligent optimization algorithms such as particle swarm optimization and genetic algorithms can also be used to solve the problem.

[0118] Finally, the risk quota data, water storage strategies, and water release strategies are used as a spatial distribution scheme for cascade risk and benefit. Specifically, the obtained risk quota data, water storage strategies, and water release strategies are organized into a structured scheme document or data file, which serves as the output of the cascade risk and benefit spatial distribution scheme. The cascade risk and benefit spatial distribution scheme describes the overall arrangement of how each reservoir should allocate risk, store water, and release water under the optimization objective, providing a basis for subsequent dynamic replacement and scheduling execution.

[0119] Furthermore, based on the marginal flood control risk function and marginal power generation benefit function of each reservoir, an objective function is constructed with the goal of maximizing the expected comprehensive efficiency of the cascade system, including:

[0120] By integrating the marginal flood control risk function of each reservoir, the total safety margin loss function of each reservoir under different water storage conditions is obtained.

[0121] By integrating the marginal power generation benefit function of each reservoir, the total power generation benefit function of each reservoir under different water storage conditions is obtained.

[0122] The sum of the total power generation benefits of each reservoir and the sum of the total safety margin losses of each reservoir are subtracted from the sum of the total power generation benefits of each reservoir to obtain the expression for the expected comprehensive efficiency of the cascade.

[0123] The objective function is to maximize the expected comprehensive efficiency expression of the ladder.

[0124] In this embodiment of the application, the marginal flood control risk function of each reservoir is first integrated to obtain the total safety margin loss function of each reservoir under different water storage conditions.

[0125] Specifically, the marginal flood control risk function describes the risk increment caused by a unit change in water storage. Integrating this function from the current water storage to the target water storage yields the total accumulated safety margin loss. The total safety margin loss function is defined as follows: The total safety margin loss function reflects the total amount of flood control risk loss accumulated when the reservoir's water storage increases from the current water storage to the target water storage.

[0126] For example, if the marginal flood control risk function of a reservoir is 0.002·V, the current water storage is 1 million m³, and the target water storage is 1.2 million m³, then the total safety margin loss from storing water from 1 million m³ to 1.2 million m³ is = =4.4.

[0127] Secondly, the marginal power generation benefit function of each reservoir is integrated to obtain the total power generation benefit function of each reservoir under different water storage conditions. Specifically, the marginal power generation benefit function describes the increase in power generation revenue caused by a unit change in water storage. Integrating it from the current water storage to the target water storage yields the cumulative total power generation benefit. The total power generation benefit function is defined as follows: The total power generation benefit function reflects the total accumulated power generation revenue when the reservoir's water storage increases from the current storage level to the target storage level.

[0128] For example, if the marginal power generation benefit function of a reservoir is 0.5 - 0.001·V, the current water storage is 1 million m³, and the target water storage is 1.2 million m³, then the total power generation benefit from storing water from 1 million m³ to 1.2 million m³ is = =7.8.

[0129] Next, by subtracting the sum of the total power generation benefit functions of each reservoir from the sum of the total safety margin loss functions of each reservoir, we obtain the expression for the expected comprehensive efficiency of the cascade hydropower station. Specifically, the expected comprehensive efficiency of the cascade hydropower station is defined as the difference between the sum of the total power generation benefit functions of all reservoirs and the sum of the total safety margin loss functions of all reservoirs: Expected Comprehensive Efficiency of the Cascade Hydropower Station == Σ Total Power Generation Benefit Function - Σ Total Safety Margin Loss Function. The expected comprehensive efficiency of the cascade hydropower station unifies power generation revenue and flood control risk into the same dimension, intuitively reflecting the overall operational efficiency of the cascade hydropower station. The higher the expected comprehensive efficiency of the cascade hydropower station, the higher the net benefit obtained under a given water storage condition.

[0130] Finally, the objective function is to maximize the expected comprehensive efficiency of the cascade. Specifically, the optimization objective is to find an optimal set of water storage capacity V or the corresponding water level and discharge process to maximize the expected comprehensive efficiency of the cascade. The objective function can be expressed as: Objective function = Max(ΣTotal power generation benefit function - ΣTotal safety margin loss function). The objective function comprehensively considers the dual objectives of maximizing power generation revenue and minimizing flood control risk, achieving Pareto optimality through the balance between marginal benefits and marginal risks.

[0131] In summary, compared to existing technologies, this application aims to maximize the expected comprehensive efficiency of the cascade reservoirs. It constructs an objective function based on the marginal flood control risk function and the marginal power generation benefit function, and solves for the spatial distribution scheme of cascade risk and benefit. In this way, flood control risk and power generation benefit are quantified uniformly and globally optimized, resulting in risk allocation and storage / release strategies for each reservoir, providing a scientific risk-benefit spatial distribution scheme for cascade reservoir groups.

[0132] S50: Within the preset risk threshold range, the flood control risk is dynamically replaced among different reservoirs according to the tiered risk benefit spatial distribution scheme, generating a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir.

[0133] There is significant spatial heterogeneity in the regulating capacity and power generation head of each reservoir in the cascade reservoir group. Upstream reservoirs usually have large capacity and strong regulating capacity but low power generation head, while downstream reservoirs often have small capacity and weak regulating capacity but high power generation head. This spatial mismatch between "capacity" and "efficiency" makes it impossible for static risk quota schemes to achieve global optimization.

[0134] Under conditions of uncertain water inflow, if downstream reservoirs are forced to lower their water levels to ensure safety, it will result in the loss of power generation benefits from high water levels; while upstream reservoirs, even if they have the capacity to bear more risks, will be unable to exert their regulatory capacity due to adhering to risk quotas.

[0135] Therefore, dynamic replacement must be carried out within the preset risk threshold range. By transferring flood control risks from downstream reservoirs with high power generation efficiency but weak regulation capacity to upstream reservoirs with strong regulation capacity but low power generation opportunity costs, the power generation potential of downstream reservoirs can be released under the premise of ensuring absolute safety, thereby maximizing the overall efficiency of the cascade.

[0136] To address the aforementioned issues, this application dynamically replaces flood control risks among different reservoirs within a preset risk threshold range based on the aforementioned tiered risk benefit spatial distribution scheme, generating a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir.

[0137] Specifically, step S50 in the method includes:

[0138] Based on the absolute safety constraints, the lower limit of the allowable safety margin for each reservoir is set as the risk threshold boundary.

[0139] Risk quota data for each reservoir is extracted from the cascade risk-benefit spatial distribution scheme. Combined with the regulating capacity data and power generation head data of each reservoir, reservoir combinations with risk replacement potential are identified. The reservoir combination includes upstream reservoirs with regulating capacity greater than a preset threshold and downstream reservoirs with power generation head higher than a preset threshold.

[0140] The rainfall forecast data and flood forecast data in the real-time forecast results are obtained. The marginal flood control risk function and marginal power generation benefit function of each reservoir are combined to dynamically calculate the change in the total power generation benefit of the cascade caused by the transfer of a unit amount of flood control risk between different reservoirs under the current forecast conditions. The change in the total power generation benefit of the cascade is used as the risk replacement rate.

[0141] Acquire flood propagation time data between adjacent cascade reservoirs, which are pre-determined based on the spatial topology of the watershed and the hydraulic characteristics of the river channel.

[0142] Based on the risk replacement rate, the current safety margin level of each reservoir, the rainfall forecast data, the flood forecast data, and the flood propagation time data, calculate the dynamic risk replacement amount that takes into account the impact of flood wave propagation time lag, provided that it does not exceed the lower limit of the safety margin of each reservoir.

[0143] Based on the dynamic risk replacement amount, the timing of flood discharge initiation, duration of flood discharge, and flood discharge flow process of each reservoir are optimized to generate the target discharge time history and target water level control trajectory for each reservoir.

[0144] The target discharge time history and the target water level control trajectory of each reservoir are combined according to the time synchronization relationship to form the differentiated joint scheduling scheme.

[0145] In this embodiment, the lower limit of the allowable safety margin for each reservoir is first set as the risk threshold boundary based on absolute safety constraints. Specifically, the absolute safety constraints are inviolable hard bottom lines, including that the water level of each reservoir must not exceed the check flood level and the discharge flow of each reservoir must not exceed the downstream river channel's limit safe discharge. Based on this, the safety margin of each reservoir is defined as 1 - comprehensive risk level. The lower limit of the safety margin is the minimum allowable safety margin, corresponding to the maximum allowable risk level. The lower limit of the safety margin can be set according to factors such as the importance of the reservoir and the level of downstream protection targets. For example, for key flood control reservoirs, the lower limit of the safety margin can be set to 0.2, meaning the comprehensive risk level does not exceed 0.8. The risk threshold boundary defines an inviolable safety red line for subsequent risk replacement; any replacement operation must ensure that the safety margin of each reservoir is not lower than the risk threshold boundary.

[0146] Secondly, risk quota data for each reservoir is extracted from the cascade risk-benefit spatial distribution scheme. Combined with the regulating capacity and power generation head data of each reservoir, reservoir combinations with risk replacement potential are identified. These combinations include upstream reservoirs with regulating capacity exceeding a preset threshold and downstream reservoirs with power generation head exceeding a preset threshold. Regulating capacity data refers to the available storage capacity of a reservoir between its normal storage level and dead storage level, reflecting its regulation capacity. Power generation head data refers to the difference in water levels between the upstream and downstream reservoirs, reflecting the power generation capacity per unit flow rate and is a key indicator for measuring power generation efficiency.

[0147] Specifically, identifying reservoir combinations with risk swapping potential requires a comprehensive consideration of each reservoir's risk allocation, regulation capacity, and power generation efficiency. First, risk allocation data for each reservoir is extracted from the cascade risk-benefit spatial distribution scheme; this represents the comprehensive risk level each reservoir is allocated to bear in the optimized scheme. Second, the regulation capacity data for each reservoir is obtained, for example, upstream reservoir A has a regulation capacity of 1 billion cubic meters, and the power generation head data, for example, downstream reservoir B has a power generation head of 80 meters. Regulation capacity thresholds (e.g., 500 million cubic meters) and power generation head thresholds (e.g., 50 meters) are set, and upstream reservoirs with regulation capacities exceeding the thresholds and downstream reservoirs with power generation heads exceeding the thresholds are selected. Simultaneously, risk allocation data must be considered: if the upstream reservoir has a surplus of risk allocation, while the downstream reservoir's risk allocation is nearing its limit, then this reservoir combination possesses risk swapping potential, allowing some risk to be transferred from downstream to upstream, thereby releasing downstream power generation capacity.

[0148] For example, in a cascade system, upstream reservoir A has a regulating capacity of 1.2 billion cubic meters (greater than the regulating capacity threshold of 500 million cubic meters), a power generation head of 40 meters (lower than the power generation head threshold of 50 meters), a risk quota of 0.7, and a current risk level of 0.5. Downstream reservoir B has a regulating capacity of 200 million cubic meters (less than the regulating capacity threshold), a power generation head of 90 meters (greater than the power generation head threshold), a risk quota of 0.6, and a current risk level approaching 0.55. Since downstream reservoir B has high power generation efficiency but weak regulating capacity and its risk is close to the upper limit, while upstream reservoir A has strong regulating capacity and its risk still has room to decrease, the AB combination is identified as a reservoir combination with risk replacement potential. Subsequent calculations and optimizations of the risk replacement rate will be performed for such combinations.

[0149] Secondly, by acquiring rainfall and flood forecast data from real-time forecast results and combining them with the marginal flood control risk function and marginal power generation benefit function of each reservoir, the change in total cascade power generation benefit caused by transferring a unit amount of flood control risk between different reservoirs under the current forecast conditions is dynamically calculated. This change in total cascade power generation benefit is used as the risk replacement rate. Specifically, the risk replacement rate is a key indicator for measuring the change in total cascade power generation benefit caused by transferring a unit of flood control risk from downstream reservoirs to upstream reservoirs. Its calculation depends on current and future forecast information and the marginal characteristics of each reservoir: First, rainfall and flood forecast data are acquired from real-time forecast results. These data reflect the inflow situation over a future period, providing a dynamic boundary for the benefit assessment of risk replacement. Then, by combining the marginal flood control risk function and marginal power generation benefit function of each reservoir, the power generation benefit lost by the downstream reservoir by reducing a unit of risk (i.e., by increasing discharge and lowering the water level) and the power generation benefit increased by the upstream reservoir by increasing a unit of risk (i.e., by reducing discharge and raising the water level) under the current conditions can be assessed.

[0150] Based on the above assessment, the risk replacement rate can be approximately expressed as: Risk Replacement Rate ≈ (MB_i - MR_i) - (MB_j - MR_j), where MB_i is the marginal power generation benefit function value of the upstream reservoir, representing the incremental power generation revenue that can be brought about by increasing the water storage of the upstream reservoir by one unit; MR_i is the marginal flood control risk function value of the upstream reservoir, representing the incremental risk loss caused by increasing the water storage of the upstream reservoir by one unit; MB_j is the marginal power generation benefit function value of the downstream reservoir, representing the incremental power generation revenue that can be brought about by increasing the water storage of the downstream reservoir by one unit; MR_j is the marginal flood control risk function value of the downstream reservoir, representing the incremental risk loss caused by increasing the water storage of the downstream reservoir by one unit.

[0151] Therefore, MB_i-MR_i represents the net benefit that the upstream reservoir can gain by increasing unit risk (i.e., storing more water), and MB_j-MR_j represents the net benefit lost by the downstream reservoir by reducing unit risk (i.e. storing less water). The difference between the two is the net change in the total power generation benefit of the cascade caused by transferring unit risk from the downstream to the upstream.

[0152] If the risk replacement rate is greater than 0, it indicates that transferring risk from downstream to upstream can bring positive incremental benefits, and the replacement is economically profitable; if the risk replacement rate is less than 0, the replacement will reduce the overall efficiency of the tiered system and should not be implemented.

[0153] It is important to emphasize that the calculation of the risk replacement rate is dynamic. It will be adjusted in real time as the real-time forecast results are updated and the status of each reservoir (such as water level and risk level) changes, so as to ensure that the replacement decision always adapts to the current and future water inflow situation.

[0154] Furthermore, flood propagation time data between adjacent cascade reservoirs is obtained, pre-determined based on the watershed spatial topology and river hydraulic characteristics. Specifically, flood propagation time data is a crucial factor affecting the effectiveness of risk replacement. Adjustments in upstream reservoir discharge require a period of river channel evolution before affecting the inflow to downstream reservoirs; this time delay constitutes the flood propagation time data. This data can be calculated based on hydraulic characteristics such as river length, cross-sectional shape, and roughness, forming an offline lookup table or functional relationship.

[0155] For example, by simulating the evolution of floods at different flow levels using a one-dimensional hydrodynamic model, the propagation time from upstream A to downstream B is found to be approximately 6 hours. This time lag must be considered in risk replacement calculations; otherwise, the timing of the replacement may be inappropriate. For instance, if the upstream increases its discharge, the downstream may only feel the increased flow 6 hours later. If the downstream is already facing flood pressure at this time, the risk may be exacerbated.

[0156] Furthermore, based on the risk replacement rate, the current safety margin level of each reservoir, rainfall forecast data, flood forecast data, and flood propagation time data, the dynamic risk replacement quantity, considering the impact of flood wave propagation time lag, is calculated under the premise of not exceeding the lower limit of the safety margin of each reservoir. Specifically, the dynamic risk replacement quantity refers to the reasonable amount of risk transferred from downstream reservoirs to upstream reservoirs, taking into account time lags and forecast information. This is a multi-time-period optimization problem: the decision variable is the transfer quantity at each time point, and the objective is to maximize the comprehensive efficiency of the cascade after replacement, while ensuring that the safety margin of each reservoir is never lower than the lower limit during and after the replacement process. Due to the existence of time lags, the transfer decision at the current time point will affect the state at future time points. Therefore, it is necessary to solve an optimization problem in a finite time domain at each time point based on the current state and future forecasts to obtain the amount of risk that should be transferred at the current time. The dynamic risk replacement quantity is dynamically updated and will be adjusted in real time as the forecast changes and the state evolves.

[0157] Furthermore, based on the dynamic risk replacement amount, the timing, duration, and flow rate of flood discharge for each reservoir are optimized to generate the target discharge time history and target water level control trajectory for each reservoir. Specifically, the dynamic risk replacement amount is a total quantity concept that needs to be transformed into specific, executable reservoir operation instructions: discharge time history curve (discharge flow rate at each moment) and water level control trajectory curve (water level at each moment).

[0158] Finally, the target discharge time histories and target water level control trajectories of each reservoir are combined according to time synchronization to form a differentiated joint dispatching scheme. Specifically, the dispatching instructions of each reservoir are aligned by time to form a differentiated joint dispatching scheme for the cascade reservoir group. The differentiated joint dispatching scheme is output in the form of tables or data files for execution by the gate control system and the power generation control system.

[0159] The calculation of dynamic risk replacement, taking into account the impact of flood wave propagation time lag, without exceeding the lower limit of the safety margin of each reservoir, includes:

[0160] Based on the flood propagation time data, establish the time-delay response relationship between upstream reservoir discharge and downstream reservoir inflow;

[0161] The time-delay response relationship is coupled with the risk replacement rate to construct a risk replacement benefit function that considers hydraulic time delay;

[0162] With the goal of maximizing the comprehensive efficiency of the cascade reservoirs, under the joint constraints of the risk threshold boundary and the time-delay response relationship, the risk replacement effect of different reservoir combinations under different flood discharge times is iteratively optimized through reinforcement learning algorithm, and the dynamic risk replacement quantity is output.

[0163] In this embodiment, a time-delay response relationship is first established between upstream reservoir discharge and downstream reservoir inflow based on flood propagation time data. Specifically, the time-delay response relationship describes the dynamic correlation between upstream reservoir discharge and downstream reservoir inflow, reflecting the time delay and morphological changes of flood waves as they evolve in the river channel. The purpose of establishing this relationship is to accurately predict the delayed impact of upstream discharge adjustments on the downstream in risk replacement calculations, avoiding inappropriate replacement timing due to neglecting propagation time.

[0164] For example, a simplified method based on flood evolution coefficients can be used to establish the time-delay response relationship: For adjacent upstream reservoir i and downstream reservoir j, let the flood propagation time be τ{ij} (in hours). The flood propagation time can be calculated based on the river length L and the average flow velocity v: τ{ij} = L / v. Based on the propagation time, the time-delay response relationship can be simplified to a translational function with attenuation: Q in j(t) = K{ij}·Qo ut i(t-τ{ij}), where Q in j(t) represents the inflow rate of downstream reservoir j at time t, Qo ut i(t-τ{ij}) represents the outflow from upstream reservoir i at time t-τ{ij}, and K{ij} is the flood peak attenuation coefficient, where 0 < K{ij} ≤ 1, reflecting the reduction of flood peak caused by channel storage. K{ij} can be obtained by fitting historical flood data, for example, by analyzing the ratio of upstream peak discharge to the corresponding downstream peak discharge in multiple floods.

[0165] For example, assume the flood propagation time from upstream reservoir A to downstream reservoir B is τ{AB} = 6 hours, and the attenuation coefficient K{AB} = 0.9. Upstream reservoir A begins discharging 1000 m³ / s at time t=0, continuing for 2 hours. According to the time-delay response relationship, the inflow to downstream reservoir B will begin to be affected at time t=6, with an inflow of 0.9 × 1000 = 900 m³ / s at t=6, 0.9 × 1000 = 900 m³ / s at t=7, and returning to normal at t=8. If the upstream discharge process is more complex (e.g., varies with time), the downstream inflow is the result of multiplying the upstream discharge process by the attenuation coefficient after a 6-hour delay. This simplified method has low computational cost, is easy to implement, and can meet the basic requirements for characterizing time-delay responses in engineering applications.

[0166] Optionally, in scenarios requiring higher accuracy, the Muskingan method or a one-dimensional hydrodynamic model can be used to establish more complex time-delay response relationships, but the basic idea is still to describe the upstream and downstream hydraulic connections through propagation time and attenuation coefficient.

[0167] Secondly, the time-delay response relationship is coupled with the risk replacement rate to construct a risk replacement benefit function that considers hydraulic time delay. Specifically, a multi-period optimization model is constructed to describe the benefit of risk replacement. The scheduling cycle is divided into T periods, for example, one hour per period. The decision variable is the amount of risk transferred from downstream reservoir j to upstream reservoir i in each period, ΔR_{ij}(t). Risk transfer is achieved by adjusting discharge: downstream reservoir j increases discharge ΔQ_out_j(t) to release the risk, while upstream reservoir i decreases discharge ΔQ_out_i(t) to absorb the risk.

[0168] Specifically, considering the time lag effect, the discharge adjustment of upstream reservoir i at time t will affect the inflow of downstream reservoir j at time t+τ (where τ is the flood propagation time), thus affecting the downstream water level and risk status. Therefore, the risk replacement benefit function needs to consider both the upstream benefit change at the current time and the downstream benefit change at future times. The risk replacement benefit function can be expressed as: Risk Replacement Benefit Function = Σ[(MB_i(t)·ΔV_i(t)-MR_i(t)·ΔV_i(t))+(MB_j(t+τ)·ΔV_j(t+τ)-MR_j(t+τ)·ΔV_j(t+τ))], where: MB_i(t) is the marginal power generation benefit function value of upstream reservoir i at time t, representing the incremental power generation revenue brought by the increase of one unit of water storage upstream; ΔV_i(t) is the marginal power generation benefit function value of upstream reservoir i at time t. The change in water storage caused by the adjustment of the discharge flow; MR_i(t) is the marginal flood control risk function value of upstream reservoir i at time t, representing the increment of risk loss caused by the increase of one unit of water storage upstream; MB_j(t+τ) is the marginal power generation benefit function value of downstream reservoir j at time t+τ; ΔV_j(t+τ) is the change in water storage of downstream reservoir j at time t+τ caused by the adjustment of the discharge flow upstream (after time delay propagation); MR_j(t+τ) is the marginal flood control risk function value of downstream reservoir j at time t+τ. In this structure, the first term of the risk substitution benefit function (MB_i(t)·ΔV_i(t)-MR_i(t)·ΔV_i(t)) represents the net benefit gained by the upstream reservoir at time t due to assuming risk (water storage); the second term (MB_j(t+τ)·ΔV_j(t+τ)-MR_j(t+τ)·ΔV_j(t+τ)) represents the net benefit gained by the downstream reservoir at time t+τ due to releasing risk (water discharge). By embedding the time delay τ into the time index of the downstream benefit, dynamic coupling between the upstream current decision and the downstream future response is achieved.

[0169] For example, suppose a cascade system consists of upstream reservoir A and downstream reservoir B, with a flood propagation time τ = 6 hours. At the current time t = 0, based on real-time forecasts and marginal function calculations, the costs for upstream A are: MB_A(0) = 0.15 yuan / m³, MR_A(0) = 0.05 yuan / m³; for downstream B: MB_B(6) = 0.25 yuan / m³, MR_B(6) = 0.10 yuan / m³. Now consider transferring a risk of ΔR = 100,000 m³ from downstream B to upstream A at time t = 0. Through inverse calculation using the marginal risk function, this requires downstream B to increase its discharge by ΔV_B = -80,000 m³ (i.e., release more water), and upstream A to decrease its discharge by ΔV_A = +80,000 m³ (i.e., store more water). Substituting the values ​​into the risk replacement benefit function: Risk Replacement Benefit Function = (0.15 × 8 - 0.05 × 8) + (0.25 × (-8) - 0.10 × (-8)) = -0.4 million yuan. The calculation shows that this replacement will reduce the overall efficiency of the tiered system by 0.4 million yuan, therefore it should not be implemented. If another set of marginal values ​​makes the risk replacement benefit function positive, then the replacement is profitable. Furthermore, by maximizing the risk replacement benefit function, the optimal risk transfer amount for each time period can be determined.

[0170] Finally, aiming to maximize the overall efficiency of the cascade system, and under the joint constraints of risk threshold boundaries and time-delay response relationships, a reinforcement learning algorithm is used to iteratively optimize the risk replacement effect of different reservoir combinations under different flood discharge times, outputting a dynamic risk replacement quantity. Specifically, due to the complex characteristics of cascade reservoir systems, such as multi-agent, time-series decision-making, time-delay effects, and uncertainties, traditional optimization methods are difficult to solve for the optimal risk replacement strategy in real time. Therefore, a reinforcement learning algorithm is used to construct a risk replacement agent, which learns the optimal decision through trial and error with the environment.

[0171] For example, a risk-permutation agent can be constructed using a deep Q-network algorithm: 1. Model Structure and Parameter Configuration: The cascade hydropower station is modeled as a reinforcement learning environment, defining the following core elements: State Space: This includes the current water level, overall risk level, safety margin, inflow, and future rainfall and flood forecast characteristics for each reservoir, such as the cumulative areal rainfall and peak flow in the next 6 hours, as well as the marginal risk function and marginal benefit function values ​​for each reservoir. Action Space: The action space is defined as the amount of risk transferred from downstream reservoir j to upstream reservoir i in each time period, which can be discretized into several levels. Reward Function: The reward function directly reflects the optimization objective and can be designed as the sum of immediate and delayed rewards. The immediate reward is the change in the overall efficiency of the cascade caused by the permutation in the current time period, i.e., the risk permutation benefit function constructed in the previous time period; the delayed reward uses a discount factor to accumulate future benefits. In addition, a penalty term can be set: if the safety margin of a reservoir is lower than the lower limit, a larger negative reward is given to strengthen safety constraints.

[0172] 2. Neural Network Architecture: A dual-network architecture using deep Q-networks is employed: the main network (Q-network) evaluates the Q-value of the current state-action pair. The input layer receives the state vector, and the intermediate layers can use 2-3 fully connected layers with 128 or 256 neurons per layer, using ReLU as the activation function. The output layer has the same number of neurons as the discrete action space and outputs the Q-value corresponding to each action. The target network has the same structure as the main network, but its parameters are periodically copied from the main network, such as updating every 100 steps, and is used to calculate the target Q-value, improving training stability.

[0173] 3. Training Process: A combination of offline training and online fine-tuning is employed. Offline Training Phase: Based on historical hydrological data and scheduling scenarios, a large number of training samples are generated, covering different inflow conditions, initial states, and boundary constraints. In each training round, an action is selected based on the current state, using an ε-greedy strategy, where ε gradually decays from 1.0 to 0.1. The system interacts with the environment to obtain the next state and reward. Small batches (e.g., 32 or 64) are randomly sampled from the experience replay pool for training. The target Q-value is calculated, and the main network parameters are updated to minimize the mean squared error loss. The number of training rounds can be set from 1000 to 5000 until the cumulative reward converges.

[0174] Furthermore, based on dynamic risk replacement, the timing, duration, and flow rate of flood discharge for each reservoir are optimized, generating the target discharge time history and target water level control trajectory for each reservoir, including:

[0175] The dynamic risk replacement amount is decomposed into the flood discharge flow adjustment amount of each reservoir in each time period within the scheduling cycle;

[0176] Based on the adjustment of the flood discharge flow of each reservoir, combined with the current water level data and available flood control capacity data of each reservoir, the timing and duration of flood discharge for each reservoir are deduced in reverse.

[0177] The timing of the flood discharge start, the duration of the flood discharge, and the adjustment amount of the flood discharge flow are input into a pre-trained flood discharge process optimization model. The flood discharge process optimization model generates the flood discharge flow sequence of each reservoir at different time periods with the constraint of minimizing the superposition of downstream flood peaks.

[0178] The discharge flow sequence is integrated in chronological order to obtain the water level change sequence for each reservoir.

[0179] The discharge flow sequence is used as the target discharge time history for each reservoir, and the water level change sequence is used as the target water level control trajectory for each reservoir.

[0180] In this embodiment, the dynamic risk replacement quantity is first decomposed into the flood discharge flow adjustment quantity of each reservoir during each time period within the scheduling cycle. Specifically, the dynamic risk replacement quantity is a total quantity, which needs to be converted into the specific flood discharge flow adjustment quantity of each reservoir.

[0181] For example, linear programming algorithms can be used to optimize the decomposition of flood discharge adjustment: the dynamic risk replacement quantity is transformed into the discharge adjustment quantity of each reservoir at each time period, which is essentially a resource allocation problem with constraints, and is suitable for precise solution using mathematical programming methods.

[0182] For example, for reservoir j downstream that needs to increase its discharge, the following linear programming model can be established: Decision variables: Let x_t be the adjustment amount of the flood discharge of reservoir j in time period t (t=1,2,...,T) (unit: m³ / s), which is a non-negative variable to be solved. Objective function: The objective is to minimize the superimposed impact of flood peaks downstream, that is, to make the discharge process as smooth as possible and avoid concentrated large flows. By solving the linear programming problem, the dynamic risk displacement can be decomposed into the flood discharge adjustment amount of each reservoir in each time period within the scheduling cycle.

[0183] Secondly, based on the adjusted discharge flow of each reservoir, combined with the current water level data and available flood control capacity data, the timing and duration of flood discharge for each reservoir can be deduced. Specifically, given the adjusted discharge flow for each time period, combined with the current water level and available capacity, it is possible to deduce when to start the adjustment and how long the adjustment will take to achieve the desired effect.

[0184] For example, the downstream reservoir, through linear programming, obtains the adjusted discharge flow rates for each time period in the next 6 hours as [40, 60, 80, 80, 60, 40] m³ / s, with a time period length Δt = 3600 seconds. Therefore, the reservoir's discharge initiation time is at the beginning of the first hour, and the duration is 6 hours. The cumulative discharge is the sum of the adjusted flow rates for each time period multiplied by Δt: (40 + 60 + 80 + 80 + 60 + 40) × 3600 = 1.296 × 10⁻⁶ 6 m³. Assume the current available flood control capacity of the reservoir is 2.0 × 10⁻⁶ m³. 6 If the current water level corresponds to a reservoir capacity of m³, and the reservoir capacity corresponding to the current water level minus the cumulative discharge is still higher than the reservoir capacity corresponding to the dead water level, then the plan is feasible. If the cumulative discharge exceeds the available reservoir capacity, then the total discharge volume needs to be reduced, i.e., the risk replacement volume needs to be reduced, or the duration needs to be extended, and the allocation needs to be re-optimized.

[0185] Next, the timing of flood discharge initiation, duration of flood discharge, and adjustment of flood discharge flow are input into a pre-trained flood discharge process optimization model. This model generates flood discharge flow sequences for each reservoir at different time periods, constrained by minimizing the cumulative impact on downstream flood peaks. Specifically, the flood discharge process optimization model is a deep learning-based sequence generation model. Its inputs are the timing of flood discharge initiation, duration of flood discharge, and adjustment of flood discharge flow for each reservoir, and its output is the flood discharge flow sequence for each reservoir in future time periods. This model learns how to adjust the flood discharge process through extensive offline training data to minimize the cumulative effect of multiple reservoir discharges on downstream flood peaks, thereby generating a smooth, staggered flood discharge scheme while meeting total flow constraints.

[0186] For example, the construction process of the discharge process optimization model includes: 1. Data collection and sample generation: collect historical flood scheduling data of the target watershed or generate a large number of training samples through hydraulic model simulation. Each sample includes: the start time of flood discharge of each reservoir in a certain scheduling scenario (such as the starting hour), the duration of flood discharge (number of hours), the total amount of flood discharge flow adjustment (ten thousand cubic meters), and the optimal discharge flow sequence in the corresponding scenario.

[0187] 2. Model Construction: A sequence-to-sequence model based on an attention mechanism is adopted. The model structure is as follows: Encoder: The input layer receives scalar features such as the start-up timing, duration, and total adjustment amount of each reservoir. These features are mapped into high-dimensional vectors through embedding and fully connected layers, and then temporal dependencies are extracted using stacked LSTM or Transformer encoders. Decoder: Using the context vector output by the encoder as the initial state, the decoder progressively decodes and generates the discharge flow values ​​for each future time period. The decoder uses an autoregressive approach, with the output of each time step serving as the input for the next time step. An attention mechanism is also introduced to monitor the encoder's output at relevant times. Output layer uses a linear activation function to ensure the physical range of the flow values. Loss Function: Mean squared error is used to measure the difference between the predicted flow sequence and the optimal flow sequence. A penalty term can be added to encourage smoothness, such as the sum of squares of the flow differences between adjacent time periods, making the sequence generated by the model more consistent with actual scheduling needs.

[0188] 3. Model Training: The dataset was randomly divided into training, validation, and test sets in an 8:1:1 ratio. The Adam optimizer was used with an initial learning rate of 0.001, a batch size of 64, and 200 training epochs. During training, the network parameters were updated via backpropagation, aiming to minimize the loss function. An early stopping mechanism was employed: training was stopped if the validation set loss did not decrease for 20 consecutive epochs to prevent overfitting. After training, the model performance was evaluated on the test set to ensure that the generated leakage sequence closely approximated the optimal sequence.

[0189] Furthermore, by integrating the discharge sequence in chronological order, the water level change sequence for each reservoir is obtained. Specifically, for each reservoir, by integrating its discharge sequence over time, the change in reservoir capacity over time can be calculated. Combined with the reservoir capacity curve (i.e., the correspondence between water level and capacity), the change in reservoir capacity can be converted into a change in water level, thus obtaining the water level control trajectory for future time periods.

[0190] Finally, the discharge flow sequence is used as the target discharge time history for each reservoir, and the water level change sequence is used as the target water level control trajectory for each reservoir. Specifically, the calculated discharge flow sequence is used as the target discharge flow for each reservoir in future time periods, and the water level sequence is used as the target water level. Together, they constitute executable scheduling instructions. These instructions are aligned by time to form the final differentiated joint scheduling scheme, which is then issued to the gate control system and power generation control system of each reservoir for execution.

[0191] In summary, compared to existing technologies, this application dynamically replaces flood control risks among different reservoirs within a preset risk threshold range based on the aforementioned cascade risk-benefit spatial distribution scheme, generating a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir. In this way, through dynamic replacement, optimized risk quotas are transformed into executable and collaborative differentiated scheduling instructions, accurately achieving a spatiotemporal rebalancing of cascade risks and benefits while ensuring absolute safety.

[0192] In summary, the embodiments of this application have at least the following technical effects:

[0193] Compared with existing technologies, this application first constructs a watershed multi-process coupled model that includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and obtains real-time forecast results through the output of the watershed multi-process coupled model, which effectively improves the accuracy and reliability of rainfall and flood forecasts, and provides accurate input data support for subsequent graded early warning and optimized scheduling.

[0194] Secondly, this application constructs a three-level early warning topology structure based on the spatial topology of the watershed, and generates a point-line-area graded early warning map based on the real-time forecast results, the flood control limit water level of each reservoir and the safe discharge of the downstream river channel. This deeply integrates the scattered real-time forecast results with the spatial location of the watershed, realizing the accurate positioning and visualization of the risk situation, and providing an intuitive decision-making basis for the intelligent scheduling of cascade reservoirs.

[0195] Furthermore, this application constructs the marginal flood control risk function and marginal power generation benefit function of each reservoir based on the aforementioned point-line-area hierarchical early warning map and the current operating data of each reservoir. By combining the point-line-area hierarchical early warning results with the real-time operating conditions of the reservoirs, it achieves a refined, quantitative, and marginal characterization of flood control risk and power generation benefits.

[0196] Furthermore, this application aims to maximize the expected comprehensive efficiency of the cascade reservoirs. Based on the marginal flood control risk function and the marginal power generation benefit function, an objective function is constructed, and the spatial distribution scheme of cascade risk and benefit is obtained by solving the problem. This unifies and quantifies flood control risk and power generation benefit and achieves global optimization, providing a scientific risk allocation and storage and release strategy for cascade reservoir groups.

[0197] Finally, within the preset risk threshold range, this application dynamically replaces flood control risks among different reservoirs according to the cascade risk-benefit spatial distribution scheme, generating a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir. Through dynamic replacement, the optimized risk quota is transformed into executable and collaborative differentiated scheduling instructions, accurately achieving the spatiotemporal rebalancing of cascade risks and benefits under the premise of ensuring absolute safety.

[0198] Through the above technical solutions, this application effectively solves the technical problems of difficulty in coordinating risks and benefits, lack of spatial correlation of early warning information, and low decision-making efficiency and insufficient accuracy in traditional scheduling. Under the premise of meeting absolute safety constraints, it maximizes the expected comprehensive efficiency of the cascade.

[0199] Example 2, as Figure 2 As shown, based on the same inventive concept as the intelligent scheduling method for cascade hydropower stations based on basin-level early warning provided in Embodiment 1, this embodiment of the invention also provides an intelligent scheduling system for cascade hydropower stations based on basin-level early warning, including:

[0200] The real-time forecast module 11 is used to build a watershed multi-process coupled model that includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and to obtain real-time forecast results through the output of the watershed multi-process coupled model.

[0201] The early warning map construction module 12 is used to construct a three-level early warning topology structure of points, lines, and surfaces based on the spatial topology of the watershed, and to generate a point-line-surface graded early warning map based on the real-time forecast results, the flood control limit water level of each reservoir, and the safe discharge of the downstream river channel.

[0202] The marginal function construction module 13 is used to construct the marginal flood control risk function and marginal power generation benefit function of each reservoir based on the point-line-surface hierarchical early warning map and the current operating data of each reservoir.

[0203] The distribution scheme generation module 14 is used to construct an objective function based on the marginal flood control risk function and the marginal power generation benefit function with the goal of maximizing the expected comprehensive efficiency of the cascade, and solve for the spatial distribution scheme of the cascade risk and benefit.

[0204] The scheduling scheme generation module 15 is used to dynamically replace flood control risks among different reservoirs within a preset risk threshold range based on the cascade risk benefit spatial distribution scheme, and generate a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir.

[0205] The real-time forecast module 11 is specifically used for:

[0206] Acquire geospatial data, historical hydrological and meteorological data, and reservoir operating data for the target watershed. The reservoir operating data includes at least the reservoir capacity curve data, discharge capacity data, and unit output characteristic data for each reservoir.

[0207] The geospatial data, the historical hydrological and meteorological data, and the reservoir operating data are input into a pre-trained watershed multi-process coupled model for coupled calculation, and the output is rainfall forecast data and flood forecast data, wherein the flood forecast data includes flood flow data of each cross section;

[0208] The rainfall forecast data and the flood forecast data are output as the real-time forecast results.

[0209] Furthermore, the construction process of the watershed multi-process coupling model includes:

[0210] Historical forecast data from multiple numerical weather prediction models within the target watershed, corresponding historical measured rainfall data, and historical inflow data from various reservoir sections were collected to form a training sample set.

[0211] An initial watershed multi-process coupled model is constructed, wherein the initial watershed multi-process coupled model includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule;

[0212] The historical forecast data in the training sample set is input into the initial watershed multi-process coupled model. The historical forecast data of multiple numerical weather prediction models are integrated and processed by the meteorological ensemble forecast submodule to generate rainfall forecast scenario data. The rainfall forecast scenario data is then converted into predicted inflow process line data and predicted cross-sectional flood flow data by the hydrological evolution simulation submodule.

[0213] With the goal of minimizing the error between the predicted inflow process data and the corresponding historical inflow data in the training sample set, supervised learning is used to train the initial watershed multi-process coupling model. The network parameters inside the meteorological ensemble forecast submodule and the hydrological evolution simulation submodule are simultaneously optimized and adjusted through the backpropagation algorithm.

[0214] When the error converges to within a preset threshold range, training stops, and the trained multi-process coupling model of the watershed is obtained.

[0215] The early warning map construction module 12 is specifically used for:

[0216] Integrate spatial location data of main and tributary rivers, reservoirs and dams, control sections and meteorological and hydrological stations within the basin area, and establish a three-level early warning topology structure of point engineering stations, line river scheduling paths and area basin.

[0217] Rainfall forecast data, flood forecast data, and flood flow data at each cross section are obtained from the real-time forecast results. The rainfall forecast data and the flood forecast data are then mapped to the corresponding point elements, line elements, and area elements in the point-line-area three-level early warning topology according to their spatial locations.

[0218] Based on the flood flow data of each section, combined with the flood control limit water level of each reservoir and the safe discharge of the downstream river channel, the point-level risk level of exceeding the threshold, the line-level risk level of exceeding the threshold of each river section, and the surface-level risk level of exceeding the threshold of each regional basin are calculated.

[0219] The point-level risk level exceeding the threshold is associated with the corresponding point feature, the point-level risk level exceeding the threshold is associated with the corresponding line feature, and the area-level risk level exceeding the threshold is associated with the corresponding area feature. Combined with the rainfall forecast data and the flood forecast data, a point-line-area graded early warning map is generated.

[0220] Specifically, the marginal function construction module 13 is used for:

[0221] Extract the point-level over-threshold risk level of each reservoir's location from the point-line-area hierarchical early warning map, as well as the line-level and area-level over-threshold risk levels associated with each reservoir's downstream location.

[0222] Extract the current water level data, available flood control capacity data, inflow data, and downstream river safe flow data of each reservoir from the current operating data of each reservoir;

[0223] Set absolute safety constraints, including that the water level of each reservoir must not exceed the check flood level and the discharge flow of each reservoir must not exceed the ultimate safe discharge capacity of the downstream river channel;

[0224] Within the scope of the absolute safety constraints, for each reservoir, based on the point-level, line-level, and surface-level risk levels exceeding the threshold, as well as the current water level data, a function for the change in the risk level exceeding the threshold caused by a change in unit water storage is constructed, and this function is used as the marginal flood control risk function.

[0225] Based on the unit output characteristic data of each reservoir, the head-output-flow relationship curve of each reservoir is derived. Combining the upstream and downstream relationship of each reservoir in the cascade, for each reservoir, based on the inflow data and the current water level data, a mapping relationship between the change in total power generation revenue of the reservoir and the downstream cascade caused by the change in unit water storage is constructed as the marginal power generation benefit function.

[0226] The distribution scheme generation module 14 is specifically used for:

[0227] Based on the marginal flood control risk function and marginal power generation benefit function of each reservoir, an objective function is constructed with the goal of maximizing the expected comprehensive efficiency of the cascade.

[0228] The constraints required to solve the objective function are set, including at least absolute safety constraints, cascade hydraulic coupling constraints, and reservoir operating condition constraints. The absolute safety constraints include that the water level of each reservoir must not exceed the check flood level and the discharge flow of each reservoir must not exceed the downstream channel's limit safe discharge. The cascade hydraulic coupling constraints include water balance constraints between adjacent reservoirs, flood wave propagation time delay constraints, and channel storage capacity constraints. The reservoir operating condition constraints include upper and lower limits of reservoir water level, discharge capacity constraints, upper and lower limits of power output constraints, and discharge flow variation constraints.

[0229] Under the common constraints of the above conditions, the objective function is solved to obtain the risk quota data that each reservoir should bear during the scheduling cycle, as well as the corresponding water storage and water release strategies.

[0230] The risk quota data, the water storage strategy, and the water release strategy are used as the spatial distribution scheme for the tiered risk-benefit profile.

[0231] Specifically, based on the marginal flood control risk function and marginal power generation benefit function of each reservoir, an objective function is constructed with the goal of maximizing the expected comprehensive efficiency of the cascade system, including:

[0232] By integrating the marginal flood control risk function of each reservoir, the total safety margin loss function of each reservoir under different water storage conditions is obtained.

[0233] By integrating the marginal power generation benefit function of each reservoir, the total power generation benefit function of each reservoir under different water storage conditions is obtained.

[0234] The sum of the total power generation benefits of each reservoir and the sum of the total safety margin losses of each reservoir are subtracted from the sum of the total power generation benefits of each reservoir to obtain the expression for the expected comprehensive efficiency of the cascade.

[0235] The objective function is to maximize the expected comprehensive efficiency expression of the ladder.

[0236] The scheduling scheme generation module 15 is specifically used for:

[0237] Based on the absolute safety constraints, the lower limit of the allowable safety margin for each reservoir is set as the risk threshold boundary.

[0238] Risk quota data for each reservoir is extracted from the cascade risk-benefit spatial distribution scheme. Combined with the regulating capacity data and power generation head data of each reservoir, reservoir combinations with risk replacement potential are identified. The reservoir combination includes upstream reservoirs with regulating capacity greater than a preset threshold and downstream reservoirs with power generation head higher than a preset threshold.

[0239] The rainfall forecast data and flood forecast data in the real-time forecast results are obtained. The marginal flood control risk function and marginal power generation benefit function of each reservoir are combined to dynamically calculate the change in the total power generation benefit of the cascade caused by the transfer of a unit amount of flood control risk between different reservoirs under the current forecast conditions. The change in the total power generation benefit of the cascade is used as the risk replacement rate.

[0240] Acquire flood propagation time data between adjacent cascade reservoirs, which are pre-determined based on the spatial topology of the watershed and the hydraulic characteristics of the river channel.

[0241] Based on the risk replacement rate, the current safety margin level of each reservoir, the rainfall forecast data, the flood forecast data, and the flood propagation time data, calculate the dynamic risk replacement amount that takes into account the impact of flood wave propagation time lag, provided that it does not exceed the lower limit of the safety margin of each reservoir.

[0242] Based on the dynamic risk replacement amount, the timing of flood discharge initiation, duration of flood discharge, and flood discharge flow process of each reservoir are optimized to generate the target discharge time history and target water level control trajectory for each reservoir.

[0243] The target discharge time history and the target water level control trajectory of each reservoir are combined according to the time synchronization relationship to form the differentiated joint scheduling scheme;

[0244] The calculation of dynamic risk replacement, taking into account the impact of flood wave propagation time lag, without exceeding the lower limit of the safety margin of each reservoir, includes:

[0245] Based on the flood propagation time data, establish the time-delay response relationship between upstream reservoir discharge and downstream reservoir inflow;

[0246] The time-delay response relationship is coupled with the risk replacement rate to construct a risk replacement benefit function that considers hydraulic time delay;

[0247] With the goal of maximizing the comprehensive efficiency of the cascade reservoirs, under the joint constraints of the risk threshold boundary and the time-delay response relationship, the risk replacement effect of different reservoir combinations under different flood discharge times is iteratively optimized through reinforcement learning algorithm, and the dynamic risk replacement quantity is output.

[0248] Furthermore, based on the aforementioned dynamic risk replacement amount, the timing of flood discharge initiation, duration, and flow rate of each reservoir are optimized to generate the target discharge time history and target water level control trajectory for each reservoir, including:

[0249] The dynamic risk replacement amount is decomposed into the flood discharge flow adjustment amount of each reservoir in each time period within the scheduling cycle;

[0250] Based on the adjustment of the flood discharge flow of each reservoir, combined with the current water level data and available flood control capacity data of each reservoir, the timing and duration of flood discharge for each reservoir are deduced in reverse.

[0251] The timing of the flood discharge start, the duration of the flood discharge, and the adjustment amount of the flood discharge flow are input into a pre-trained flood discharge process optimization model. The flood discharge process optimization model generates the flood discharge flow sequence of each reservoir at different time periods with the constraint of minimizing the superposition of downstream flood peaks.

[0252] The discharge flow sequence is integrated in chronological order to obtain the water level change sequence for each reservoir.

[0253] The discharge flow sequence is used as the target discharge time history for each reservoir, and the water level change sequence is used as the target water level control trajectory for each reservoir.

[0254] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0255] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0256] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0257] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0258] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0259] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0260] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent scheduling of cascade hydropower stations based on basin-level early warning, characterized in that, The method includes: A watershed multi-process coupled model is constructed, which includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and real-time forecast results are obtained through the output of the watershed multi-process coupled model. Based on the spatial topological relationship of the watershed, a three-level early warning topology structure of point-line-area is constructed, and a point-line-area graded early warning map is generated according to the real-time forecast results, the flood control limit water level of each reservoir and the safe discharge of the downstream river channel. Based on the point-line-surface graded early warning map and the current operating data of each reservoir, construct the marginal flood control risk function and marginal power generation benefit function of each reservoir; With the goal of maximizing the expected comprehensive efficiency of the cascade, an objective function is constructed based on the marginal flood control risk function and the marginal power generation benefit function, and the spatial distribution scheme of cascade risk and benefit is obtained by solving the problem. Within the preset risk threshold range, the flood control risk is dynamically replaced among different reservoirs according to the tiered risk benefit spatial distribution scheme, generating a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir. A watershed multi-process coupled model is constructed, comprising a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule. Real-time forecast results are obtained through the output of this watershed multi-process coupled model, including: Acquire geospatial data, historical hydrological and meteorological data, and reservoir operating data for the target watershed. The reservoir operating data includes at least the reservoir capacity curve data, discharge capacity data, and unit output characteristic data for each reservoir. The geospatial data, the historical hydrological and meteorological data, and the reservoir operating data are input into a pre-trained watershed multi-process coupled model for coupled calculation, and the output is rainfall forecast data and flood forecast data, wherein the flood forecast data includes flood flow data of each cross section; The rainfall forecast data and the flood forecast data are output as the real-time forecast results; The construction process of a watershed multi-process coupling model includes: Historical forecast data from multiple numerical weather prediction models within the target watershed, corresponding historical measured rainfall data, and historical inflow data from various reservoir sections were collected to form a training sample set. An initial watershed multi-process coupled model is constructed, wherein the initial watershed multi-process coupled model includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule; The historical forecast data in the training sample set is input into the initial watershed multi-process coupled model. The historical forecast data of multiple numerical weather prediction models are integrated and processed by the meteorological ensemble forecast submodule to generate rainfall forecast scenario data. The rainfall forecast scenario data is then converted into predicted inflow process line data and predicted cross-sectional flood flow data by the hydrological evolution simulation submodule. With the goal of minimizing the error between the predicted inflow process data and the corresponding historical inflow data in the training sample set, supervised learning is used to train the initial watershed multi-process coupling model. The network parameters inside the meteorological ensemble forecast submodule and the hydrological evolution simulation submodule are simultaneously optimized and adjusted through the backpropagation algorithm. When the error converges to within a preset threshold range, training stops, and the trained multi-process coupling model of the watershed is obtained.

2. The intelligent scheduling method for cascade hydropower stations based on basin-level early warning as described in claim 1, characterized in that, A three-tiered early warning topology structure (point-line-area) is constructed based on the spatial topology of the watershed. A tiered early warning map (point-line-area) is generated based on the real-time forecast results, the flood control limit water levels of each reservoir, and the safe discharge capacity of downstream rivers. This map includes: Integrate spatial location data of main and tributary rivers, reservoirs and dams, control sections and meteorological and hydrological stations within the basin area, and establish a three-level early warning topology structure of point engineering stations, line river scheduling paths and area basin. Rainfall forecast data, flood forecast data, and flood flow data at each cross section are obtained from the real-time forecast results. The rainfall forecast data and the flood forecast data are then mapped to the corresponding point elements, line elements, and area elements in the point-line-area three-level early warning topology according to their spatial locations. Based on the flood flow data of each section, combined with the flood control limit water level of each reservoir and the safe discharge of the downstream river channel, the point-level risk level of exceeding the threshold, the line-level risk level of exceeding the threshold of each river section, and the surface-level risk level of exceeding the threshold of each regional basin are calculated. The point-level risk level exceeding the threshold is associated with the corresponding point feature, the point-level risk level exceeding the threshold is associated with the corresponding line feature, and the area-level risk level exceeding the threshold is associated with the corresponding area feature. Combined with the rainfall forecast data and the flood forecast data, a point-line-area graded early warning map is generated.

3. The intelligent scheduling method for cascade hydropower stations based on basin-level early warning as described in claim 1, characterized in that, Based on the aforementioned point-line-area graded early warning map and the current operating data of each reservoir, the marginal flood control risk function and marginal power generation benefit function of each reservoir are constructed, including: Extract the point-level over-threshold risk level of each reservoir's location from the point-line-area hierarchical early warning map, as well as the line-level and area-level over-threshold risk levels associated with each reservoir's downstream location. Extract the current water level data, available flood control capacity data, inflow data, and downstream river safe flow data of each reservoir from the current operating data of each reservoir; Set absolute safety constraints, including that the water level of each reservoir must not exceed the check flood level and the discharge flow of each reservoir must not exceed the ultimate safe discharge capacity of the downstream river channel; Within the scope of the absolute safety constraints, for each reservoir, based on the point-level, line-level, and surface-level risk levels exceeding the threshold, as well as the current water level data, a function for the change in the risk level exceeding the threshold caused by a change in unit water storage is constructed, and this function is used as the marginal flood control risk function. Based on the unit output characteristic data of each reservoir, the head-output-flow relationship curve of each reservoir is derived. Combining the upstream and downstream relationship of each reservoir in the cascade, for each reservoir, based on the inflow data and the current water level data, a mapping relationship between the change in total power generation revenue of the reservoir and the downstream cascade caused by the change in unit water storage is constructed as the marginal power generation benefit function.

4. The intelligent scheduling method for cascade hydropower stations based on basin-level early warning as described in claim 1, characterized in that, With the goal of maximizing the expected comprehensive efficiency of the cascade, an objective function is constructed based on the marginal flood control risk function and the marginal power generation benefit function, and the spatial distribution scheme of cascade risk and benefit is obtained by solving the problem, including: Based on the marginal flood control risk function and marginal power generation benefit function of each reservoir, an objective function is constructed with the goal of maximizing the expected comprehensive efficiency of the cascade. The constraints required to solve the objective function are set, including at least absolute safety constraints, cascade hydraulic coupling constraints, and reservoir operating condition constraints. The absolute safety constraints include that the water level of each reservoir must not exceed the check flood level and the discharge flow of each reservoir must not exceed the downstream channel's limit safe discharge. The cascade hydraulic coupling constraints include water balance constraints between adjacent reservoirs, flood wave propagation time delay constraints, and channel storage capacity constraints. The reservoir operating condition constraints include upper and lower limits of reservoir water level, discharge capacity constraints, upper and lower limits of power output constraints, and discharge flow variation constraints. Under the common constraints of the above conditions, the objective function is solved to obtain the risk quota data that each reservoir should bear during the scheduling cycle, as well as the corresponding water storage and water release strategies. The risk quota data, the water storage strategy, and the water release strategy are used as the spatial distribution scheme for the tiered risk-benefit profile.

5. The intelligent scheduling method for cascade hydropower stations based on basin-level early warning as described in claim 4, characterized in that, Based on the marginal flood control risk function and marginal power generation benefit function of each reservoir, an objective function is constructed with the goal of maximizing the expected comprehensive efficiency of the cascade system, including: By integrating the marginal flood control risk function of each reservoir, the total safety margin loss function of each reservoir under different water storage conditions is obtained. By integrating the marginal power generation benefit function of each reservoir, the total power generation benefit function of each reservoir under different water storage conditions is obtained. The sum of the total power generation benefits of each reservoir and the sum of the total safety margin loss functions of each reservoir are subtracted from the sum of the total power generation benefits functions of each reservoir to obtain the expression for the expected comprehensive efficiency of the cascade. The objective function is to maximize the expected comprehensive efficiency expression of the ladder.

6. The intelligent scheduling method for cascade hydropower stations based on basin-level early warning as described in claim 1, characterized in that, Within a preset risk threshold range, flood control risks are dynamically shifted among different reservoirs based on the tiered risk-benefit spatial distribution scheme, generating a differentiated joint scheduling scheme that includes the discharge time histories and water level control trajectories of each reservoir, including: Based on the absolute safety constraints, the lower limit of the allowable safety margin for each reservoir is set as the risk threshold boundary. Risk quota data for each reservoir is extracted from the cascade risk-benefit spatial distribution scheme. Combined with the regulating capacity data and power generation head data of each reservoir, reservoir combinations with risk replacement potential are identified. The reservoir combination includes upstream reservoirs with regulating capacity greater than a preset threshold and downstream reservoirs with power generation head higher than a preset threshold. The rainfall forecast data and flood forecast data in the real-time forecast results are obtained. The marginal flood control risk function and marginal power generation benefit function of each reservoir are combined to dynamically calculate the change in the total power generation benefit of the cascade caused by the transfer of a unit amount of flood control risk between different reservoirs under the current forecast conditions. The change in the total power generation benefit of the cascade is used as the risk replacement rate. Acquire flood propagation time data between adjacent cascade reservoirs, which are pre-determined based on the spatial topology of the watershed and the hydraulic characteristics of the river channel. Based on the risk replacement rate, the current safety margin level of each reservoir, the rainfall forecast data, the flood forecast data, and the flood propagation time data, calculate the dynamic risk replacement amount that takes into account the impact of flood wave propagation time lag, provided that it does not exceed the lower limit of the safety margin of each reservoir. Based on the dynamic risk replacement amount, the timing of flood discharge initiation, duration of flood discharge, and flood discharge flow process of each reservoir are optimized to generate the target discharge time history and target water level control trajectory for each reservoir. The target discharge time history and the target water level control trajectory of each reservoir are combined according to the time synchronization relationship to form the differentiated joint scheduling scheme; The calculation of dynamic risk replacement, taking into account the impact of flood wave propagation time lag, without exceeding the lower limit of the safety margin of each reservoir, includes: Based on the flood propagation time data, establish the time-delay response relationship between upstream reservoir discharge and downstream reservoir inflow; The time-delay response relationship is coupled with the risk replacement rate to construct a risk replacement benefit function that considers hydraulic time delay; With the goal of maximizing the comprehensive efficiency of the cascade reservoirs, under the joint constraints of the risk threshold boundary and the time-delay response relationship, the risk replacement effect of different reservoir combinations under different flood discharge times is iteratively optimized through reinforcement learning algorithm, and the dynamic risk replacement quantity is output.

7. The intelligent scheduling method for cascade hydropower stations based on basin-level early warning as described in claim 6, characterized in that, Based on the aforementioned dynamic risk replacement amount, the timing of flood discharge initiation, duration of flood discharge, and flood discharge flow process of each reservoir are optimized to generate the target discharge time history and target water level control trajectory for each reservoir, including: The dynamic risk replacement amount is decomposed into the flood discharge flow adjustment amount of each reservoir in each time period within the scheduling cycle; Based on the adjustment of the flood discharge flow of each reservoir, combined with the current water level data and available flood control capacity data of each reservoir, the timing and duration of flood discharge for each reservoir are deduced in reverse. The timing of the flood discharge start, the duration of the flood discharge, and the adjustment amount of the flood discharge flow are input into a pre-trained flood discharge process optimization model. The flood discharge process optimization model generates the flood discharge flow sequence of each reservoir at different time periods with the constraint of minimizing the superposition of downstream flood peaks. The discharge flow sequence is integrated in chronological order to obtain the water level change sequence for each reservoir. The discharge flow sequence is used as the target discharge time history for each reservoir, and the water level change sequence is used as the target water level control trajectory for each reservoir.

8. A cascade hydropower station intelligent dispatching system based on basin-level early warning, characterized in that, The method for executing the intelligent scheduling method for cascade hydropower stations based on basin-level early warning as described in any one of claims 1-7 includes: The real-time forecast module is used to build a watershed multi-process coupled model that includes a meteorological ensemble forecast submodule and a hydrological evolution simulation submodule, and to obtain real-time forecast results through the output of the watershed multi-process coupled model. The early warning map construction module is used to construct a three-level early warning topology structure of points, lines, and areas based on the spatial topology of the watershed, and to generate a point-line-area graded early warning map based on the real-time forecast results, the flood control limit water level of each reservoir, and the safe discharge of the downstream river channel. The marginal function construction module is used to construct the marginal flood control risk function and marginal power generation benefit function of each reservoir based on the point-line-surface hierarchical early warning map and the current operating data of each reservoir. The distribution scheme generation module is used to construct an objective function based on the marginal flood control risk function and the marginal power generation benefit function with the goal of maximizing the expected comprehensive efficiency of the cascade, and solve for the spatial distribution scheme of the cascade risk and benefit. The scheduling scheme generation module is used to dynamically replace flood control risks among different reservoirs within a preset risk threshold range based on the cascade risk-benefit spatial distribution scheme, and generate a differentiated joint scheduling scheme that includes the discharge time history and water level control trajectory of each reservoir.

Images