Ecological scheduling method for relieving biogenic substance retention of deep and large reservoir

Abstract

The invention discloses an ecological scheduling method for relieving biogenic substance retention of a deep and large reservoir, and belongs to the technical field of water resource management and environmental protection. The method comprises the following steps: 1, collecting deep and large reservoir characteristic parameters and various hydrology and water quality monitoring data; 2, constructing a three-dimensional hydrodynamic force-water quality model of the deep and large reservoir; step 3, based on the three-dimensional hydrodynamic force-water quality model of the deep and large reservoir, constructing a data-driven proxy model in combination with a long-series multi-input and multi-output sequence of the model; and 4, constructing a reservoir optimization scheduling model, and deducing a proper scheduling rule for a decision maker. The hydrodynamic condition is improved through drainage, biogenic substance retention is reduced, the activation conveying effect is achieved through the reservoir area water temperature layering condition, downstream conveying of effective biogenic substances is enhanced while the power generation amount is guaranteed, and therefore comprehensive quantity, energy and quality regulation and control considering power generation, ecological flow and biogenic substances are achieved; and a technical support is provided for optimizing comprehensive management of basin water resource utilization and ecological environment protection.

Description

Technical Field

[0001] The invention belongs to the technical field of water resource management and environmental protection, and relates to an ecological scheduling method for alleviating the retention of biogenic substances in deep and large reservoirs. Background Art

[0002] The construction and operation of large and deep reservoirs significantly impact the water quality and ecosystems of the reservoir area and downstream rivers. Reservoir impoundment traps biogenic substances such as nitrogen and phosphorus in the water, leading to the accumulation of biogenic substances such as total nitrogen (TN) and total phosphorus (TP) in the reservoir area, potentially leading to long-term eutrophication. Furthermore, downstream rivers are susceptible to oligotrophication due to the discharge of low-temperature water and reduced nutrient availability. Therefore, developing operational strategies that can mitigate the retention of biogenic substances in large and deep reservoirs is crucial.

[0003] To address these issues, Chinese invention patent CN111080157B proposes a method and system for scheduling phosphorus discharge from cascade hydropower stations. This system integrates the dual objectives of total power generation and phosphorus discharge from cascade hydropower stations into a single scheduling objective, generates a sequence of scheduling objectives at different weights, and then optimizes each scheduling objective to achieve a water level scheduling process. This approach can alleviate the ecological and environmental issues associated with hydropower station construction. Chinese invention patent CN115659641B proposes a machine learning-based method for scheduling water bloom control. This method primarily constructs a water bloom control scheduling model that incorporates both economic and ecological objective functions. The ecological objective function is established using an LSTM model of the water body's comprehensive nutrient index and phytoplankton density. A cuckoo optimization algorithm is used to solve the multi-objective problem, obtaining a non-inferior solution, from which a satisfactory ecological scheduling solution is selected. However, the water quality scheduling objectives in this patent primarily rely on downstream flux or reservoir nitrogen and phosphorus concentrations and their derived nutrient indices, and the scheduling target is a fixed water intake level. The migration and transformation of nitrogen and phosphorus within large and deep reservoirs exhibits a distinct three-stage process: retention, transformation, and transport. The dominant mechanisms vary significantly across time periods and spatial locations: during the non-flood season, the activation and release of dissolved nitrogen and phosphorus dominates in the dam front area, while during the flood season, the retention and interception of particulate nitrogen and phosphorus dominates in the central reservoir area. Existing technologies do not fully consider the migration and transformation mechanisms of biogenic matter in large and deep reservoirs, resulting in deficiencies in scientific decision-making and refined scheduling, leaving room for improvement.

[0004] In summary, how to combine the migration and transformation mechanism of biogenic elements in deep and large reservoirs to provide an ecological scheduling method to alleviate the retention of biogenic substances in deep and large reservoirs, balance the contradiction between power generation and the environment, reduce eutrophication in the reservoir area while ensuring the water quality of downstream rivers has become an urgent problem to be solved. Summary of the Invention

[0005] In order to solve the problem that the existing technology does not fully consider the temporal and spatial evolution of biogenic matter inside the reservoir when considering ecological scheduling of water quality, and it is difficult to effectively alleviate the retention of nitrogen and phosphorus while ensuring fine power generation, the present invention provides an ecological scheduling method for alleviating the retention of biogenic matter in deep and large reservoirs. The present invention reduces the retention of biogenic matter by improving hydrodynamic conditions through discharge, and utilizes the water temperature stratification conditions in the reservoir area to achieve activation and transportation. While ensuring the power generation, the effective biogenic matter is strengthened to the downstream. Thus, the "comprehensive regulation of quantity, energy and quality" considering power generation, ecological flow and biogenic matter is realized, and technical support is provided for the comprehensive management of optimizing water resource utilization and ecological environment protection in the basin.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] An ecological scheduling method for alleviating the retention of biogenic matter in deep and large reservoirs comprises the following steps:

[0008] The first step is to collect characteristic parameters of deep and large reservoirs in the study basin and various types of hydrological and water quality monitoring data, as follows:

[0009] The characteristic parameters of deep and large reservoirs mainly include: digital elevation model DEM, reservoir dam height, total storage capacity, normal water level, and water level-reservoir capacity curve.

[0010] The various types of hydrological and water quality monitoring data mainly include: meteorological data, hydrological data, pollution load data, inflow data, discharge data, water level data, surface water temperature data, vertical water temperature data, surface water quality data, vertical water quality profile data, point source and non-point source pollution data.

[0011] The second step is to build a three-dimensional hydrodynamic-water quality model for the deep and large reservoir. Based on the characteristic parameters of the deep and large reservoir collected in the first step and various types of hydrological and water quality monitoring data, a three-dimensional hydrodynamic module and a water quality module for the deep and large reservoir are constructed. The two are coupled and calibrated to form a reliable three-dimensional hydrodynamic-water quality model for the deep and large reservoir. The details are as follows:

[0012] Step 2.1: Based on the characteristic parameters of the deep and large reservoirs and various types of hydrological and water quality monitoring data collected in the first step, the horizontal and vertical grids are divided, the terrain and water depth data are input, and the boundary conditions are set to construct the three-dimensional hydrodynamic module of the deep and large reservoirs. Specifically:

[0013] Grid type selection: Use curvilinear coordinates for planar grid division to effectively conform to irregular, curved reservoir boundaries. Grid resolution is increased in areas with significant river curvature, while maintaining a moderate simplification in areas with smooth boundaries. Select the σ coordinate system for vertical grid division to better conform to complex underwater topography and accurately reflect changes in water depth.

[0014] Topography and water depth data input: Based on the spatial data of the digital elevation model (DEM), the elevation data of the entire reservoir area is extracted, and accurate elevation values ​​are assigned to each grid cell within the reservoir area to reflect the undulating changes in the terrain, providing a basis for the simulation of water flow, material transport and water quality changes.

[0015] Boundary Condition Setting: Setting boundary conditions primarily involves input data, initial background value settings, and warm-up period selection. For inflow boundaries, physical elements such as flow and temperature, meteorological elements such as precipitation and evaporation, and upstream water quality load input are required. The model's initial conditions are based on key elements such as flow, water temperature, meteorology, and water quality corresponding to the first day of the simulation period. A hot start method is used for initialization, with the output at the end of the first year of the simulation period serving as the hot start input for the following year.

[0016] At this point, a three-dimensional hydrodynamic module for deep and large reservoirs has been constructed. To further optimize computing performance, efficient computing functions such as parallel computing, nested grids, and computational domain decomposition are adopted to enable it to perform large-scale hydrodynamic simulations and achieve local refined calculations in key areas, thereby improving its adaptability and simulation accuracy in complex hydrodynamic environments.

[0017] Step 2.2: Select water quality related processes and set key parameters to build a water quality module. Specifically:

[0018] Selection of water quality-related processes: Comprehensively consider the migration and transformation processes of key water quality components in water bodies, such as dissolved oxygen (DO), ammonia nitrogen (NH4), nitrate nitrogen (NO3), orthophosphate (PO4), particulate nitrogen (DetN) and particulate phosphorus (DetP).

[0019] Key parameter settings include: gravitational acceleration, water density, and air density; setting corresponding wind resistance coefficients and wind speed thresholds for different wind speed ranges; and related parameters such as friction coefficient, horizontal eddy viscosity coefficient, and eddy diffusion coefficient. Regarding heat exchange at the water-air interface, the Secchi depth, Dalton number, and Stanton number are key parameters controlling the evaporation and heat exchange processes at the water surface. Regarding water quality parameter settings, parameters such as the decomposition rate constant and sedimentation velocity of nitrogen and phosphorus are set to simulate the transformation of particulate nitrogen and phosphorus into dissolved nitrogen and phosphorus, which plays an important role in simulating the migration, sedimentation, and re-release of nitrogen and phosphorus.

[0020] At this point, a water quality module has been constructed, which uses the finite volume method for numerical solution. Its numerical solver can efficiently process convection, diffusion and reaction equations, accurately calculate water flow transport, pollutant diffusion and biogeochemical processes, ensure the conservation of mass of various substances in the numerical calculation process, and improve the accuracy and stability of the simulation.

[0021] In step 2.3, the water quality module is coupled with the hydrodynamic module to form a three-dimensional hydrodynamic-water quality model for deep and large reservoirs, so that it has comprehensive water quality processes and efficient computing capabilities.

[0022] Coupling mechanism: A loose coupling mechanism is adopted between the three-dimensional hydrodynamic module and the water quality module of deep and large reservoirs to achieve coordinated operation. The basic principle is to use the hydrodynamic process as the external driving force for water quality evolution, and drive the transport and transformation process of substances in the water body through dynamic flow field information, thereby realizing the dynamic simulation of material migration and changes in complex water bodies.

[0023] Coupling steps: First, the hydrodynamic module simulates and outputs key hydrodynamic parameters such as flow velocity, water depth, temperature, and turbulent diffusion coefficient. These parameters vary with time and space, reflecting the dynamic characteristics of real hydrological processes. Subsequently, these parameters are passed as forced inputs to the water quality module within each spatial grid and each time step, driving its dynamic simulation of the convection, diffusion, and reaction processes of substances.

[0024] In step 2.4, based on the characteristic parameters of the deep and large reservoirs and the multi-category hydrological and water quality monitoring data collected in the first step, the three-dimensional hydrodynamic-water quality model of the deep and large reservoirs is calibrated and verified with an integrated approach of "multi-process, multi-factor, multi-time and multi-indicator".

[0025] In terms of "multi-process": focus on the dynamic linkage and physical consistency of key processes such as hydrodynamic evolution, water temperature structure, and water quality transformation.

[0026] In terms of "multiple factors": multiple variables such as water level, flow, water temperature, total nitrogen (TN), total phosphorus (TP), dissolved oxygen (DO), and biochemical oxygen demand (BOD) are included to improve the comprehensiveness and accuracy of parameter constraints.

[0027] In terms of "multi-time and space": covering high-frequency water level and water temperature data at the daily scale throughout the year, as well as typical seasonal scales and key periods such as flood season and non-flood season, and conducting multi-time period, multi-section, and multi-depth collaborative verification at key locations such as the main stream, tributaries, reservoirs, and in front of the dam, as well as vertically.

[0028] In terms of "multiple indicators": avoid relying on only one or a few indicators to judge model performance, and use a combination of multiple indicators such as Nash-Sutcliffe efficiency coefficient (NSE), determination coefficient (R²), root mean square error (RMSE), absolute relative deviation (ADR%) to comprehensively evaluate the performance and physical consistency of the three-dimensional hydrodynamic-water quality model of deep and large reservoirs under different processes, elements and spatiotemporal scales.

[0029] In step 2.5, based on the calibrated and verified 3D hydrodynamic-water quality model for deep and large reservoirs, organize its input and simulation output results to form a training dataset for constructing subsequent data-driven proxy models.

[0030] The third step is to build a data-driven proxy model based on the three-dimensional hydrodynamic-water quality model of the deep and large reservoir constructed in the second step. The details are as follows:

[0031] Step 3.1, organize and divide the training data set:

[0032] Based on step 2.5, obtain a training dataset for the data-driven proxy model. To ensure the comprehensiveness and representativeness of the proxy model training, divide the training dataset into a training set, a validation set, and a test set, with a ratio of 60%, 20%, and 20%, respectively.

[0033] Step 3.2: Build the proxy model. After the training dataset in step 3.1 is prepared, the proxy model is built. The following are the detailed steps:

[0034] Step 3.2.1, proxy model architecture design:

[0035] Input layer design: The input layer of the proxy model receives multiple long-sequence features, including long-series hydrological and water quality data input to the three-dimensional hydrodynamic-water quality model of deep and large reservoirs, and multi-layer water quality concentration data output by the three-dimensional hydrodynamic-water quality model of large reservoirs; if the input feature dimension is n, then the input layer contains n neurons.

[0036] Hidden Layer Design: To capture the complex nonlinear relationship between water quality and environmental drivers, the model includes multiple hidden layers and uses Rectified Linear Unit (ReLU) as the activation function. ReLU effectively alleviates the vanishing gradient problem and improves convergence speed.

[0037] Output Layer Design: The model's output layer, composed of multiple neurons, is used to predict water quality indicators. Single-point prediction is used, outputting the temporal evolution of water quality at a representative location or layer. Because water quality prediction is a regression problem, the output layer uses a linear activation function, which is suitable for numerical prediction.

[0038] Step 3.2.2, training process:

[0039] Training Dataset: Using the training data from step 3.1, input the meteorological data, hydrological data, and pollution load data collected in the first step of the feature set. The target variable is the water quality change process (such as surface water quality and vertical water quality). The loss function uses the mean square error (MSE) to calculate the difference between the predicted value of the proxy model and the output value of the deep and large reservoir 3D hydrodynamic-water quality model:

[0040] (1)

[0041] in, For the The predicted value of the sample, is the true observation value, is the total number of samples, Indicates the sample index.

[0042] Proxy model performance optimization: The Adam optimizer is used. This optimization algorithm combines adaptive learning rate and momentum methods to improve the efficiency and convergence speed of proxy model training. The Adam algorithm updates the parameters of each layer using the following update rules:

[0043] (2)

[0044] in, is the learning rate, and are estimates of the first and second moments, A constant to prevent division by zero; For the Model parameters obtained by time period iteration; For the The model parameters obtained by time period iteration.

[0045] Regularization: To avoid overfitting, regularization techniques, also known as weight decay, are used to control model complexity by adding regularization terms to the loss function:

[0046] (3)

[0047] in, is the regularization coefficient, is the weight parameter in the model; For the total loss.

[0048] Training Strategy: Use an early stopping strategy to prevent overfitting. When performance on the validation set stops improving, training is stopped early. This strategy can effectively save training time while avoiding overfitting the training data.

[0049] Step 3.3, model validation and evaluation:

[0050] To ensure the accuracy and generalization ability of the model, the proxy model was verified and evaluated in various ways. The cross-validation method was mainly used to ensure the stability and reliability of the proxy model. Specifically:

[0051] Cross-validation: Use K-fold cross-validation to evaluate the model. The data is divided into K subsets, one of which is used as the validation set. The remaining subsets are used for training, and the average performance of the surrogate model is evaluated. This method can effectively reduce the accidental errors caused by data partitioning.

[0052] Evaluation indicators: Root mean square error (RMSE) and Nash efficiency coefficient (NSE) are mainly used. The closer the value is to 1, the better the model performance is.

[0053] Step 4: Build a reservoir optimization operation model and couple the data-driven proxy model obtained in Step 3 as a water quality response prediction tool with the reservoir optimization operation model. This approach can meet multiple water quality objectives while taking into account power generation needs, and derive appropriate operation rules for decision makers. Specifically:

[0054] Step 4.1, set the decision variables:

[0055] Determining decision variables: To ensure that established ecologically relevant water quality and power generation targets can be achieved under different water inflow scenarios, the reservoir operation process must be considered as a decision variable, ensuring applicability under uncertainty. The operation rules are composed of a combination of target reservoir capacity, output coefficient, and discharge location set for each operation period, which are used to determine the corresponding relationship between unit output and reservoir capacity.

[0056] Step 4.2, set the optimization goal:

[0057] Determine power generation targets: Maximizing the average multi-year power generation of deep and large reservoirs is used as the power generation efficiency optimization target to provide an economic evaluation basis for subsequent scheduling plans.

[0058] (4)

[0059] in, is the total number of scheduling periods, Indicates the calculation period; To contribute to the power station, is the multi-year average power generation; is the time step.

[0060] Determine downstream water quality targets: Considering that the main focus of deep and large reservoirs during the non-flood season is the activation and release of dissolved nitrogen and phosphorus, while during the flood season it is the retention and interception of particulate nitrogen and phosphorus, different water quality targets are set for different time periods to alleviate eutrophication within the reservoir area and oligotrophication downstream. During the non-flood season, the control target is mainly the concentration of dissolved nitrogen and phosphorus in the reservoir water, which is controlled to be as low as possible to improve the self-purification capacity of the water body and reduce the risk of eutrophication in the reservoir area. During the flood season, the control target is mainly the downstream flux of biogenic nitrogen and phosphorus. By optimizing the discharge strategy, as much effective biogenic material as possible is transported downstream, thereby alleviating the oligotrophic phenomenon of downstream water bodies.

[0061] (5)

[0062] (6)

[0063] (7)

[0064] (8)

[0065] in, For the reservoir The discharge flow at the moment, This is the layer where the reservoir discharge is located. , time period TP concentration in This is the layer where the reservoir discharge is located. , time period TN concentration in is the average TN concentration of multiple layers in the reservoir area; CS TP is the average TP concentration of multiple layers in the reservoir area; N represents the total number of monitoring times, and n represents the index of monitoring times; is the TP flux through the reservoir discharge section; is the TN flux through the reservoir discharge section.

[0066] Step 4.3, set constraints:

[0067] Constraints: Reservoir optimization scheduling generally follows water balance constraints, reservoir capacity constraints, reservoir discharge constraints, power generation flow constraints, reservoir output constraints, and scheduling rule parameter range constraints.

[0068] 1) Reservoir water balance constraints:

[0069] (9)

[0070] 2) Reservoir capacity constraints:

[0071] (10)

[0072] 3) Hydropower station output constraints:

[0073] (11)

[0074] 4) River ecological flow constraints:

[0075] (12)

[0076] Where: For the reservoir Water storage at the time, S t-1 is the water storage capacity of the reservoir at time t-1; is the dead storage capacity of the reservoir; is the normal water storage capacity of the reservoir; For reservoirs Inflow flow at any moment, is the installed capacity of the reservoir, MW; It is the ecological flow of the river downstream of the reservoir.

[0077] Step 4.4, set the solution algorithm:

[0078] Selection of optimization solution algorithm: Considering that the reservoir optimization operation model is multi-objective, the heuristic algorithm NSGA-III algorithm that takes into account both computational efficiency and solution quality is selected to solve it.

[0079] By solving a multi-objective reservoir optimization scheduling model, a set of scheduling rules that meet power generation needs and water quality requirements were obtained, providing a scientific basis and technical support for the comprehensive management of water resources and water environment to alleviate the retention of biogenic substances in the reservoir area and improve downstream water quality.

[0080] An ecological scheduling method for alleviating the retention of biogenic matter in deep and large reservoirs is implemented based on the deep and large reservoir ecological scheduling system. The deep and large reservoir ecological scheduling system includes a data acquisition module, a deep and large reservoir hydrodynamic-water quality model construction module, a data-driven agent model construction module, and a reservoir scheduling optimization module. Specifically:

[0081] Data acquisition module: used to obtain characteristic parameters of deep and large reservoirs in the study basin and various types of hydrological and water quality monitoring data, providing basic input for hydrodynamic and water quality simulation of deep and large reservoirs.

[0082] Module for constructing a 3D hydrodynamic-water quality model for large and deep reservoirs: This module constructs a 3D hydrodynamic module and a water quality module for large and deep reservoirs. By solving the continuity equation, momentum equation, and water quality component transport equation, it simulates the spatiotemporal distribution characteristics of water quality indicators under historical operating conditions. The module then couples and calibrates the two to form a reliable 3D hydrodynamic-water quality model for large and deep reservoirs.

[0083] Data-driven surrogate model construction module: Based on the simulation results of the 3D hydrodynamic-water quality model of deep and large reservoirs and related input characteristics and output results, a surrogate model suitable for water quality prediction is constructed;

[0084] Reservoir scheduling optimization module: Coupled with the data-driven agent model to build a reservoir optimization scheduling model, and deduce a set of scheduling rules while meeting power generation demand and water quality requirements.

[0085] The beneficial effects of the present invention are:

[0086] (1) This invention combines the biogenic matter migration and transformation mechanism of deep and large reservoirs and sets differentiated water quality targets by time period. It constructs a refined ecological scheduling strategy that integrates discharge volume, discharge time, and discharge location, and proposes an ecological scheduling technology to alleviate biogenic matter retention. This technology overcomes the shortcomings of existing technologies in terms of scientific decision-making and refined scheduling, providing a scientific basis and technical support for the operation of deep and large reservoirs.

[0087] (2) This paper uses a data-driven proxy model to provide a high-fidelity approximation of the hydrodynamic-water quality model for deep and large reservoirs, enabling rapid prediction of multi-layer water quality in the reservoir area. This approach overcomes the inefficiency of traditional physical hydrodynamic-water quality models in practical applications due to computational complexity and long runtimes, thereby improving the real-time and operability of reservoir ecological scheduling.

[0088] (3) The present invention provides a set of efficient and intelligent decision-making support tools for ecological scheduling to alleviate the retention of biogenic substances in deep and large reservoirs. It not only improves the scientificity and refinement of management, but also promotes the protection of water ecological environment and the sustainable utilization of water energy resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 A flow chart of the ecological scheduling method for alleviating the retention of biogenic matter in deep and large reservoirs provided by the present invention;

[0090] Figure 2 This is a schematic diagram of the structure of the deep and large reservoir hydrodynamic-water quality coupling model provided by the present invention;

[0091] Figure 3 Performance evaluation diagram of the data-driven proxy model provided by the present invention. DETAILED DESCRIPTION

[0092] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0093] The purpose of the present invention is to provide an ecological scheduling method and system for alleviating the retention of biogenic matter in deep and large reservoirs, which can take into account the benefits of hydropower generation and the water quality control needs of the reservoir area and downstream, scientifically and efficiently deduce the ecological scheduling strategy of deep and large reservoirs, and provide decision-making support for the sustainable development and ecological protection of the region.

[0094] To make the above-mentioned objects, features and advantages of the present invention more clearly understood, the Xiaowan Reservoir in the Lancang River Basin is taken as an example. The present invention is further described in detail with reference to the accompanying drawings and specific embodiments to promote the application of the present invention to the Yarlung Zangbo River Basin and other deep and large reservoirs on the plateau.

[0095] See also Figure 1 Ecological scheduling methods to alleviate the retention of biomass in deep and large reservoirs include:

[0096] Step 1: Collect characteristic parameters of deep and large reservoirs and various types of hydrological and water quality monitoring data, as follows:

[0097] The characteristic parameters of deep and large reservoirs mainly include: elevation model (DEM), reservoir dam height, total storage capacity, normal water level, and water level-reservoir capacity curve. The Xiaowan Reservoir dam is 292 meters high and has a total storage capacity of 151.32×10 8 The normal water level of Xiaowan Reservoir is 1,240 meters, and the corresponding water surface area is about 189 square kilometers.

[0098] The various types of hydrological and water quality monitoring data mainly include: meteorological data, inflow data, discharge data, water level data, surface water temperature data, vertical water temperature data, surface water quality data, vertical water quality profile data, point source and non-point source pollution data.

[0099] The second step is to build a three-dimensional hydrodynamic-water quality model for deep and large reservoirs. Based on the characteristic parameters of deep and large reservoirs and various types of hydrological and water quality monitoring data from the first step, a three-dimensional hydrodynamic module and a water quality module for deep and large reservoirs are built, and the two are coupled and calibrated to form a reliable three-dimensional hydrodynamic-water quality model for deep and large reservoirs. Figure 2 , as follows:

[0100] Step 2.1: Based on the characteristic parameters of deep and large reservoirs and the collection of various types of hydrological and water quality monitoring data from the first step, the horizontal and vertical grids are divided, and the terrain and water depth data are input and boundary conditions are set to construct the hydrodynamic module. Specifically:

[0101] Grid type selection: Use curvilinear coordinates for planar grid division to effectively conform to irregular, curved reservoir boundaries. Grid resolution is increased in areas with significant river curvature, while maintaining a moderate simplification in areas with smooth boundaries. Select the σ coordinate system for vertical grid division to better conform to complex underwater topography and accurately reflect changes in water depth.

[0102] Topography and water depth data input: Based on the spatial data of the digital elevation model (DEM), the elevation data of the entire reservoir area was extracted, and accurate elevation values ​​were assigned to each grid cell within the reservoir area to reflect the undulating changes in the terrain, providing a basis for the simulation of water flow, material transport and water quality changes.

[0103] Boundary Condition Setting: Setting boundary conditions primarily involves input data, initial background value settings, and warm-up period selection. For inflow boundaries, physical elements such as flow and temperature, meteorological elements such as precipitation and evaporation, and upstream water quality load input are required. The model's initial conditions are based on key elements such as flow, water temperature, meteorology, and water quality corresponding to the first day of the simulation period. A hot start method is used for initialization, with the output at the end of the first year of the simulation period serving as the hot start input for the following year.

[0104] At this point, a hydrodynamic module has been constructed. To further optimize computing performance, efficient computing functions such as parallel computing, nested grids, and computational domain decomposition are adopted to enable it to perform large-scale hydrodynamic simulations and achieve local refined calculations in key areas, thereby improving its adaptability and simulation accuracy in complex hydrodynamic environments.

[0105] Step 2.2: Rationally configure water quality related processes and set key parameters to build a water quality module. Specifically:

[0106] Selection of water quality-related processes: The migration and transformation processes of key water quality components in water bodies, such as dissolved oxygen (DO), ammonia nitrogen (NH4), nitrate nitrogen (NO3), orthophosphate (PO4), particulate nitrogen (DetN) and particulate phosphorus (DetP), were comprehensively considered.

[0107] Key parameter settings: Gravity acceleration 9.81m / s 2 、Water density 1000kg / m 3 and air density 1kg / m 3 ; Set the corresponding drag coefficient and wind speed threshold for different wind speed ranges, where the drag coefficient A is 0.00063, the drag coefficient B is 0.00723, and the drag coefficient C is 0.00723; the friction coefficient is 0.022, and the horizontal eddy viscosity coefficient is 2m 2 / s and eddy diffusion coefficient 2m 2 / s and other related parameters. In terms of heat exchange at the water-air interface, the Secchi depth of 2m, the Dalton number of 0.0013 and the Stanton number of 0.0013 are key parameters for controlling the evaporation and heat exchange process of the water surface. In terms of water quality parameter setting, the decomposition rate constant and sedimentation velocity of nitrogen and phosphorus are set, among which the sedimentation velocity is 0.1m / d and the N decomposition rate constant is 0.0003d -1 , P decomposition rate constant d -1 .

[0108] At this point, a water quality module has been constructed, which uses the finite volume method for numerical solution. Its numerical solver can efficiently process convection, diffusion and reaction equations, accurately calculate water flow transport, pollutant diffusion and biogeochemical processes, ensure the conservation of mass of each substance in the numerical calculation process, and improve the accuracy and stability of the simulation.

[0109] In step 2.3, the water quality module is coupled with the hydrodynamic module to form a three-dimensional hydrodynamic-water quality model for deep and large reservoirs, so that it has comprehensive water quality processes and efficient computing capabilities.

[0110] Coupling mechanism: A loose coupling mechanism is adopted between the hydrodynamic module and the water quality module to achieve collaborative operation. The basic principle is to use the hydrodynamic process as the external driving force for water quality evolution, and drive the transport and transformation process of substances in the water body through dynamic flow field information, thereby realizing the dynamic simulation of material migration and changes in complex water bodies.

[0111] Coupling steps: First, the hydrodynamic module simulates and outputs key hydrodynamic parameters such as flow velocity, water depth, temperature, and turbulent diffusion coefficient. These parameters vary with time and space, reflecting the dynamic characteristics of real hydrological processes. Subsequently, these parameters are passed as forced inputs to the water quality module within each spatial grid and each time step, driving its dynamic simulation of the convection, diffusion, and reaction processes of substances.

[0112] In step 2.4, based on the first step of collecting various types of hydrological and water quality monitoring data of deep and large reservoirs, the three-dimensional hydrodynamic-water quality model of deep and large reservoirs is calibrated and verified with an integrated approach of "multi-process, multi-factor, multi-time and multi-indicator".

[0113] In terms of "multi-process": focus on the dynamic linkage and physical consistency of key processes such as hydrodynamic evolution, water temperature structure, and water quality transformation.

[0114] In terms of "multiple factors": multiple variables such as water level, flow, water temperature, total nitrogen (TN), total phosphorus (TP), dissolved oxygen (DO), and biochemical oxygen demand (BOD) are included to improve the comprehensiveness and accuracy of parameter constraints.

[0115] In terms of "multi-time and space": covering high-frequency water level and water temperature data at the daily scale throughout the year, as well as typical seasonal scales and key periods such as flood season and non-flood season, and conducting multi-time period, multi-section, and multi-depth collaborative verification at key locations such as the main stream, tributaries, reservoirs, and in front of the dam, as well as at a vertical depth of 150m.

[0116] Regarding "multi-indicator" evaluation, we avoid relying solely on a single or limited number of indicators to judge model performance. Instead, we use a combination of indicators, including the Nash-Sutcliffe efficiency coefficient (NSE), coefficient of determination (R²), root mean square error (RMSE), and absolute relative deviation (ADR%), to comprehensively evaluate the model's performance and physical consistency across different processes, elements, and spatiotemporal scales. The NSE for water level, flow rate, and water temperature are all greater than 0.9, the R² is greater than 0.95, the RMSE is less than 1.2, and the ADR% is less than 6.5%. For water quality, the standard for model performance evaluation is slightly lowered. Previous studies have indicated that an NSE > 0.3 has been the accepted accuracy standard for most water quality modeling studies. With the exception of individual points where the NSE, R², RMSE, and ADR% for water quality elements in this paper generally meet these requirements.

[0117] In step 2.5, based on the calibrated and verified 3D hydrodynamic-water quality model for deep and large reservoirs, organize its input and simulation output results to form a training dataset for constructing subsequent data-driven proxy models.

[0118] In the third step, based on the three-dimensional hydrodynamic-water quality model of deep and large reservoirs constructed in the second step, and combining its long series of multi-input and multi-output sequences, a data-driven proxy model is constructed as follows:

[0119] Step 3.1, organize and divide the training data set:

[0120] Training Dataset Partitioning: Based on step 2.5, we obtained the training dataset for the proxy model. To ensure the extensiveness and representativeness of the proxy model training, we divided the dataset into a training set, a validation set, and a test set, with a ratio of 60%, 20%, and 20%, respectively.

[0121] Step 3.2: Build the proxy model. After the training dataset in step 3.1 is prepared, the proxy model construction phase begins. The following are the detailed steps for model construction:

[0122] Proxy model architecture design:

[0123] Input layer design: The input layer of the data-driven model receives multiple long-sequence features, including long-series hydrological and water quality data input to the three-dimensional hydrodynamic-water quality model of deep and large reservoirs, and multi-layer water quality concentration data output by the three-dimensional hydrodynamic-water quality model of large reservoirs; if the input feature dimension is n, the input layer contains n neurons.

[0124] Hidden layer design: To capture the complex nonlinear relationship between water quality and environmental drivers, the model includes multiple hidden layers and uses ReLU as the activation function. ReLU can effectively alleviate the vanishing gradient problem and improve convergence speed.

[0125] Output Layer Design: The model's output layer, composed of multiple neurons, is used to predict water quality indicators. Single-point prediction is used, outputting the water quality change over time at a representative location / layer. Because water quality prediction is a regression problem, the output layer uses a linear activation function, which is suitable for numerical prediction.

[0126] Training process:

[0127] Training Dataset: Using the training data from step 3.1, the input features are meteorological data, hydrological data, and pollution load data, and the target variable is the water quality process (such as surface water quality and vertical water quality). The loss function uses the mean squared error (MSE) to calculate the difference between the predicted value of the surrogate model and the simulated output of the hydrodynamic-water quality model.

[0128] Model performance optimization: The Adam optimizer is used. This optimization algorithm combines adaptive learning rate and momentum methods to improve the efficiency and convergence speed of agent model training. The Adam algorithm updates the parameters of each layer using the following update rules:

[0129] Regularization: To avoid overfitting, regularization techniques, also known as weight decay, are used to control model complexity by adding regularization terms to the loss function.

[0130] Training Strategy: Use early stopping to prevent overfitting. Stop training early when performance on the validation set stops improving. This strategy can effectively save training time while avoiding overfitting the training data.

[0131] Step 3.3, model validation and evaluation:

[0132] To ensure the accuracy and generalization ability of the model, the proxy model was verified and evaluated in various ways. The cross-validation method was mainly used to ensure the stability and reliability of the proxy model. Specifically:

[0133] Cross-validation: Use K-fold cross-validation to evaluate the model. The data is divided into K subsets, one of which is used as the validation set. The remaining subsets are used for training, and the average performance of the model is evaluated. This method can effectively reduce the accidental errors caused by data partitioning.

[0134] Evaluation indicators: Root mean square error (RMSE) and Nash efficiency coefficient (NSE) are mainly used. The closer the value is to 1, the better the model performance is.

[0135] Performance reference for proxy models Figure 3 , where the horizontal axis represents the predicted values ​​from the proxy model, and the vertical axis represents the simulated values ​​from the deep and large reservoir hydrodynamic-water quality model. All data points are closely distributed on both sides of the 1:1 line, indicating that the proxy model has high prediction accuracy and reliability and can accurately reflect the changing patterns of actual physical processes.

[0136] Step 4: Build a reservoir optimization operation model and couple the data-driven proxy model obtained in Step 3 as a water quality response prediction tool with reservoir optimization operation. This approach can meet multiple water quality objectives while taking into account power generation needs, and help decision makers develop an appropriate operation process. Specifically:

[0137] Step 4.1, set the decision variables:

[0138] Determining decision variables: To ensure that established ecologically relevant water quality and power generation targets can be achieved under different water inflow scenarios, the reservoir operation process must be considered as a decision variable, ensuring applicability under uncertainty. The operation rules are composed of a combination of target reservoir capacity, output coefficient, and discharge location set for each operation period, which are used to determine the corresponding relationship between unit output and reservoir capacity.

[0139] Step 4.2, set the optimization goal:

[0140] Determine the power generation target: Maximize the average power generation of deep and large reservoirs over many years as the power generation benefit optimization target, and provide an economic evaluation basis for subsequent scheduling plans, as shown in formula (4).

[0141] Determine the downstream water quality target: Considering that the activation and release of dissolved nitrogen and phosphorus in the non-flood season is the main focus of deep and large reservoirs, while the retention and interception of particulate nitrogen and phosphorus is the main focus during the flood season, different water quality targets are set for different time periods to alleviate the problems of eutrophication in the reservoir area and oligotrophication downstream. In the non-flood season, the control target is mainly the concentration of dissolved nitrogen and phosphorus in the reservoir water body, which is controlled to be as low as possible to improve the self-purification capacity of the water body and reduce the risk of eutrophication in the reservoir area; in the flood season, the control target is mainly the downstream flux of biogenic nitrogen and phosphorus. By optimizing the discharge strategy, as much effective biogenic material as possible is transported downstream, thereby alleviating the oligotrophic phenomenon of the downstream water body. For example, as shown in formulas (5)-(8).

[0142] Step 4.3, set constraints:

[0143] Constraints: Reservoir optimization scheduling generally follows water balance constraints, reservoir storage capacity constraints, reservoir discharge constraints, power generation flow constraints, reservoir output constraints, and scheduling rule parameter range constraints, such as formulas (9)-(12).

[0144] Step 4.4, set the solution algorithm:

[0145] Optimization Algorithm Selection: Considering the multi-objective nature of the reservoir optimization operation model, the NSGA-III heuristic algorithm, designed to balance computational efficiency and solution quality, was selected to solve the optimization operation model. Key algorithm parameters were set as follows: population size of 200, maximum number of evolutionary generations of 500, crossover probability of 0.9, mutation probability of 1 / D (where D is the dimension of the decision variable), reference point generation using the normalized boundary crossover method of Das and Dennis, crossover distribution index of 30, mutation distribution index of 20, and the use of simulated binary crossover and polynomial mutation operators. This parameter combination effectively improves the algorithm's convergence performance and distribution uniformity in high-dimensional objective spaces while maintaining population diversity.

[0146] By solving the multi-objective optimization scheduling model, a set of scheduling rules that meet power generation needs and water quality requirements were obtained, providing a scientific basis and technical support for the comprehensive management of water resources and water environment to alleviate the retention of biogenic matter in the reservoir area and improve downstream water quality.

[0147] An ecological scheduling method for alleviating the retention of biogenic matter in deep and large reservoirs is implemented based on the deep and large reservoir ecological scheduling system. The deep and large reservoir ecological scheduling system includes a data acquisition module, a deep and large reservoir hydrodynamic-water quality model construction module, a data-driven agent model construction module, and a reservoir scheduling optimization module. Specifically:

[0148] Data acquisition module: used to obtain characteristic parameters of deep and large reservoirs in the study basin and various types of hydrological and water quality monitoring data, providing basic input for hydrodynamic and water quality simulation of deep and large reservoirs.

[0149] Module for constructing a 3D hydrodynamic-water quality model for large and deep reservoirs: This module constructs a 3D hydrodynamic module and a water quality module for large and deep reservoirs. By solving the continuity equation, momentum equation, and water quality component transport equation, it simulates the spatiotemporal distribution characteristics of water quality indicators under historical operating conditions. The module then couples and calibrates the two to form a reliable 3D hydrodynamic-water quality model for large and deep reservoirs.

[0150] Data-driven surrogate model construction module: Based on the simulation results of the 3D hydrodynamic-water quality model of deep and large reservoirs and related input characteristics and output results, a surrogate model suitable for water quality prediction is constructed;

[0151] Reservoir scheduling optimization module: Coupled with the data-driven agent model to build a reservoir optimization scheduling model, and deduce a set of scheduling rules while meeting power generation demand and water quality requirements.

[0152] The above-described embodiments merely express the implementation methods of the present invention, but should not be understood as limiting the scope of the present invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, which all fall within the scope of protection of the present invention.

Claims

1. An ecological scheduling method for alleviating the retention of biogenic matter in deep and large reservoirs, characterized in that: The ecological scheduling method includes the following steps: The first step is to collect characteristic parameters of deep and large reservoirs in the study basin and various types of hydrological and water quality monitoring data; The second step is to construct a three-dimensional hydrodynamic-water quality model for deep and large reservoirs; Based on the characteristic parameters of deep and large reservoirs and various types of hydrological and water quality monitoring data collected in the first step, a three-dimensional hydrodynamic module and a water quality module for deep and large reservoirs are constructed, and the two are coupled and calibrated to form a three-dimensional hydrodynamic-water quality model for deep and large reservoirs; Step 2.1: Based on the characteristic parameters of the deep and large reservoirs and the collection of various types of hydrological and water quality monitoring data from the first step, the horizontal and vertical grids are divided, the terrain and water depth data are input, and the boundary conditions are set to construct the three-dimensional hydrodynamic module of the deep and large reservoirs; Step 2.2, selection of water quality related processes and setting of key parameters to construct the water quality module; Step 2.3: Couple the water quality module with the deep and large reservoir 3D hydrodynamic module to form the deep and large reservoir 3D hydrodynamic-water quality model, as follows: The three-dimensional hydrodynamic module of the deep and large reservoir obtained in step 2.1 and the water quality module obtained in step 2.2 are loosely coupled to achieve collaborative operation, thus realizing the dynamic simulation of material migration and changes in complex water bodies. First, the deep and large reservoir 3D hydrodynamic module simulates and outputs key hydrodynamic parameters such as flow velocity, water depth, temperature, and turbulent diffusion coefficient. These key hydrodynamic parameters are then passed as forced inputs to the water quality module within each spatial grid and each time step, driving its dynamic simulation of the convection, diffusion, and reaction processes of the substances. This completes the construction of the deep and large reservoir 3D hydrodynamic-water quality model. Step 2.4: Based on the characteristic parameters of the deep and large reservoirs and the multi-category hydrological and water quality monitoring data collected in the first step, the three-dimensional hydrodynamic-water quality model of the deep and large reservoirs is calibrated and verified with a multi-process, multi-factor, multi-spatiotemporal, and multi-index integrated approach. Step 2.5: Based on the calibrated and validated 3D hydrodynamic-water quality model for deep and large reservoirs, organize its input and simulation output results to form a training dataset for constructing subsequent data-driven proxy models. The third step is to build a data-driven proxy model based on the three-dimensional hydrodynamic-water quality model of the deep and large reservoir constructed in the second step; Step 4: Construct a reservoir optimization scheduling model, and couple the data-driven proxy model obtained in the third step with the reservoir optimization scheduling model as a water quality response prediction tool, and solve it to obtain scheduling rules that meet power generation demand and water quality requirements.

2. The ecological scheduling method for alleviating the retention of biomass in deep and large reservoirs according to claim 1 is characterized in that: In the first step, the characteristic parameters of the deep and large reservoir include digital elevation model (DEM), reservoir dam height, total storage capacity, normal water level, and water level-storage capacity curve; The multiple types of hydrological and water quality monitoring data include: meteorological data, hydrological data, pollution load data, inflow data, outflow data, water level data, surface water temperature data, vertical water temperature data, surface water quality data, vertical water quality profile data, point source and non-point source pollution data.

3. The ecological scheduling method for alleviating the retention of biomass in deep and large reservoirs according to claim 2 is characterized in that: In the second step: The step 2.1 is specifically as follows: Grid type selection: Use curvilinear coordinate grid for plane grid division; select σ coordinate system for vertical grid division; Topography and water depth data input: Based on the spatial data of the digital elevation model (DEM), the elevation data of the entire reservoir area is extracted, and accurate elevation values ​​are assigned to each grid cell within the reservoir area; Boundary condition setting: mainly includes input data, initial background value setting and warm-up period selection; At this point, the construction of the three-dimensional hydrodynamic module of the deep and large reservoir has been completed; The step 2.2 is specifically as follows: Selection of water quality-related processes: Comprehensive consideration of the migration and transformation processes of key water quality components in the water body, such as dissolved oxygen, ammonia nitrogen, nitrate nitrogen, orthophosphate, particulate nitrogen, and particulate phosphorus; Key parameter settings: gravitational acceleration, water density, and air density; setting corresponding drag coefficients and wind speed thresholds for different wind speed ranges; parameters related to friction coefficient, horizontal eddy viscosity coefficient, and eddy diffusion coefficient; in terms of heat exchange at the water-air interface, Secchi depth, Dalton number, and Stanton number are key parameters controlling water surface evaporation and heat exchange processes; in terms of water quality parameter settings, setting nitrogen and phosphorus decomposition rate constants and sedimentation velocity parameters to simulate the conversion of particulate nitrogen and phosphorus to dissolved nitrogen and phosphorus; At this point, the construction of the water quality module is completed.

4. The ecological scheduling method for alleviating the retention of biomass in deep and large reservoirs according to claim 3 is characterized in that: The step 2.4 is specific as follows: In terms of multiple processes: focus on the dynamic linkage and physical consistency of key processes of hydrodynamic evolution, water temperature structure, and water quality transformation; In terms of multiple factors: water level, flow, water temperature, total nitrogen, total phosphorus, dissolved oxygen, and biochemical oxygen demand are included; In terms of multi-time and space: high-frequency water level and water temperature data covering the entire year on a daily basis; In terms of multiple indicators: Nash efficiency coefficient NSE, determination coefficient R², root mean square error RMSE, and absolute relative deviation ADR% are jointly used to comprehensively evaluate the expressiveness and physical consistency of the three-dimensional hydrodynamic-water quality model of deep and large reservoirs under different processes, elements and temporal and spatial scales.

5. The ecological scheduling method for alleviating the retention of biomass in deep and large reservoirs according to claim 4 is characterized in that: The third step is specifically as follows: Step 3.1, organize and divide the training data set: Based on step 2.5, obtain the training dataset for the data-driven surrogate model and divide it into a training set, a validation set, and a test set; Step 3.2, construction of proxy model: Step 3.2.1, proxy model architecture design: Input layer design: The input layer of the proxy model receives multiple long-sequence features, including hydrological and water quality data input to the deep and large reservoir 3D hydrodynamic-water quality model, as well as multi-layer water quality concentration data output by the deep and large reservoir 3D hydrodynamic-water quality model; If the input feature dimension is n, then the input layer contains n neurons; Hidden layer design: contains multiple hidden layers and uses ReLU as the activation function; Output layer design: The output layer consists of multiple neurons and is used to predict water quality indicators; Step 3.2.2, training process: Training data set: Using the training set from step 3.1, input the meteorological data, hydrological data, and pollution load data collected in the first step of the feature set, and the target variable is the water quality change process; The loss function uses the mean square error (MSE) to calculate the difference between the predicted value of the proxy model and the output value of the deep and large reservoir three-dimensional hydrodynamic-water quality model: (1) in, For the The predicted value of the sample, is the true observation value, is the total number of samples, Indicates the sample index; Proxy model performance optimization: Adopt Adam optimizer and update the parameters of each layer according to the following update rules: (2) in, is the learning rate, and are estimates of the first and second moments, A constant to prevent division by zero; For the Model parameters obtained by time period iteration; For the Model parameters obtained during the period; Regularization: Regularization technology is used to control the complexity by adding regularization terms in the loss function: (3) in, is the regularization coefficient, is the weight parameter in the model; is the total loss; Training strategy: Use early stopping strategy to prevent overfitting; In step 3.3, the stability and reliability of the proxy model are ensured through cross-validation.

6. The ecological scheduling method for alleviating the retention of biomass in deep and large reservoirs according to claim 5 is characterized in that: The step 3.3 is specifically as follows: the proxy model is evaluated using the K-fold cross-validation method, the data is divided into K subsets, one of the subsets is used as the validation set each time, and the remaining subsets are used for training, and finally the average performance of the proxy model is evaluated; the evaluation indicators include the root mean square error (RMSE) and the Nash efficiency coefficient (NSE), and the closer the value is to 1, the better the performance.

7. The ecological scheduling method for alleviating the retention of biomass in deep and large reservoirs according to claim 6 is characterized in that: The fourth step is specifically as follows: Step 4.1, set the decision variables: The reservoir operation process is used as a decision variable. The operation rules are composed of the target storage capacity, output coefficient and discharge position set in each operation period, which are used to determine the corresponding relationship between unit output and storage capacity. Step 4.2, set the optimization goal: Determine power generation targets: Maximizing the average annual power generation of deep and large reservoirs is the goal of optimizing power generation benefits; (4) in, is the total number of scheduling periods, Indicates the calculation period; Provide power for the power station; is the multi-year average power generation; is the time step; Determine the downstream water quality targets: Set different water quality targets for different time periods; during the non-flood season, the control targets are mainly based on the concentration of dissolved nitrogen and phosphorus in the reservoir water; during the flood season, the control targets are mainly based on the downstream flux of biogenic nitrogen and phosphorus; (5) (6) (7) (8) in, For the reservoir The discharge flow at the moment; This is the layer where the reservoir discharge is located. , time period TP concentration within; This is the layer where the reservoir discharge is located. , time period TN concentration within; is the average TN concentration of multiple layers in the reservoir area; CS TP is the average TP concentration of multiple layers in the reservoir area; N represents the total number of monitoring times, and n represents the index of monitoring times; is the TP flux through the reservoir discharge section; is the TN flux through the reservoir discharge section; Step 4.3: Set constraints, including water balance constraints, reservoir capacity constraints, reservoir discharge constraints, power generation flow constraints, reservoir output constraints, and dispatch rule parameter range constraints. Step 4.4, set the solution algorithm: Selection of optimization solution algorithm: Considering that the reservoir optimization operation model is multi-objective, the heuristic algorithm NSGA-III algorithm is selected to solve it; By solving a multi-objective reservoir optimization scheduling model, a set of scheduling rules that meet power generation demand and water quality requirements are obtained.

8. The ecological scheduling method for alleviating the retention of biomass in deep and large reservoirs according to claim 7, characterized in that: The ecological scheduling method is implemented based on the deep and large reservoir ecological scheduling system, which includes a data acquisition module, a deep and large reservoir hydrodynamic-water quality model construction module, a data-driven agent model construction module, and a reservoir scheduling optimization module.

9. The ecological scheduling method for alleviating the retention of biogenic matter in deep and large reservoirs according to claim 8, characterized in that: In the deep and large reservoir ecological dispatching system: Data acquisition module: used to obtain characteristic parameters of deep and large reservoirs in the study basin and various types of hydrological and water quality monitoring data; Module for constructing a 3D hydrodynamic-water quality model for large and deep reservoirs: This module constructs a 3D hydrodynamic module and a water quality module for large and deep reservoirs. By solving the continuity equation, momentum equation, and water quality component transport equation, it simulates the spatiotemporal distribution characteristics of water quality indicators under historical operating conditions. The module then couples and calibrates the two to form a reliable 3D hydrodynamic-water quality model for large and deep reservoirs. Data-driven surrogate model construction module: Based on the simulation results of the 3D hydrodynamic-water quality model of deep and large reservoirs and related input characteristics and output results, a surrogate model suitable for water quality prediction is constructed; Reservoir scheduling optimization module: Coupled with the data-driven agent model to build a reservoir optimization scheduling model, and deduce a set of scheduling rules while meeting power generation demand and water quality requirements.

Images