Reservoir self-adaptive dispatching method and system for coping with salt tide based on distribution robust optimization

Abstract

The application discloses a reservoir self-adaptive scheduling method and system based on distribution robust optimization for coping with salt tide, and specifically is as follows: a reservoir, a water plant and a water demand area affected by a salt tide are abstracted as directed graph nodes, a connection relationship between the nodes is abstracted as a directed edge, and a flow on the directed edge is taken as a decision variable, so that a multi-water source-multi-water plant-multi-water area hierarchical water supply topology structure is constructed; a multi-objective multi-water source collaborative scheduling model is established, and a constraint condition is set for the decision variable; a Vassstein fuzzy set is constructed, a dynamic robust radius parameter is introduced, and a distribution robust aggregation evaluation is performed on an original objective function in the multi-objective multi-water source collaborative scheduling model; and an improved third-generation non-dominated sorting genetic algorithm is used to solve the multi-objective multi-water source collaborative scheduling model to obtain optimal decision variables. The application greatly improves the anti-risk resilience of the system.

Description

Technical Field

[0001] This invention belongs to the field of water resource scheduling technology, and in particular relates to an adaptive scheduling method and system for reservoirs to cope with saltwater intrusion based on distributed bar optimization. Background Technology

[0002] Estuarine cities with multiple water sources are frequently affected by the combined effects of tides, upstream runoff anomalies, and extreme weather events, resulting in significant uncertainty and unpredictability in chloride concentration and available water volume at their water source intakes. Traditional deterministic scheduling methods, failing to consider uncertain parameters, are ill-suited to address these risks. Some existing studies employ stochastic programming (SP) to handle the uncertainty of hydrological parameters; however, conventional SP relies on the assumption of stationarity, ensuring that the future distribution remains consistent with historical experience. However, in the context of climate change, when faced with severe data distribution shifts due to extreme drought years or anomalous saltwater intrusions, rigid SP models are prone to overfitting, failing to effectively control tail risk exposure under extreme conditions, potentially leading to severe water shortages and water quality safety incidents.

[0003] On the other hand, regional water supply systems affected by salinity intrusion typically include multiple water sources, water treatment plants, and water demand areas, exhibiting a complex hierarchical physical topology. Optimal scheduling of the system not only needs to meet stringent engineering boundary constraints such as maximum reservoir supply capacity, water treatment plant capacity, node water balance, and pipeline distribution capacity, but also requires seeking a Pareto equilibrium among three mutually constraining objectives: minimizing water shortage, minimizing salinity intrusion-affected water supply, and minimizing differences in water supply costs. Conventional scheduling algorithms, when dealing with such high-dimensional multi-objective conflicts, often lack heuristic search and adaptive repair mechanisms tailored to the hydraulic engineering characteristics of water supply. This easily leads to the generation of numerous infeasible solutions that violate water balance or physical capacity limits during the optimization process, rendering the algorithm-generated scheduling schemes impossible to implement on the actual physical pipeline network.

[0004] Furthermore, current research on joint scheduling of water supply systems largely remains at the stage of outputting massive numerical solution sets, failing to effectively establish a dynamic mapping relationship between the evolution of external risk states and the specific actions executed by the system. In actual water dispatching scenarios, when faced with sudden saltwater intrusion warnings, managers find it difficult to quickly extract control instructions with engineering guidance value directly from complex mathematical optimization results. Currently, the industry lacks an engineering decision-making tool that can transform multi-objective trade-off solution sets into intuitive triggering conditions and scheduling action rules, which severely restricts the application of complex operations research optimization algorithms to actual front-line water dispatching operations. Summary of the Invention

[0005] Purpose of the invention: To overcome the overfitting defects of existing technologies in dealing with the distribution shift of saline intrusion data, and the lack of engineering feasibility and adaptive decision support of conventional algorithms in solving high-dimensional multi-objective water supply scheduling, the primary objective of this invention is to provide a method and system for adaptive scheduling of reservoirs in response to saline intrusion based on distributed bar optimization.

[0006] Technical solution: This invention proposes an adaptive scheduling method for reservoirs to cope with saltwater intrusion based on distributed bar optimization, specifically as follows:

[0007] The reservoirs, water plants, and water demand areas in the target area affected by saltwater intrusion are abstracted as nodes in a directed graph, and the connection relationships between nodes are abstracted as directed edges. The flow on the directed edges is used as a decision variable, thereby constructing a multi-source-multi-water plant-multi-water demand area stratified water supply topology. The decision variables include the water supply from the reservoir to the water plant and the water transfer from the water plant to the water demand area.

[0008] Based on the stratified water supply topology of multiple water sources, multiple water plants, and multiple water demand areas, a multi-objective multi-water source collaborative scheduling model is established, and constraints are set for decision variables.

[0009] Construct Wasserstein fuzzy sets and introduce dynamically robust radius parameters. The original objective function in the multi-objective, multi-source collaborative scheduling model is evaluated by sub-Brussels bar aggregation, and the original objective function is transformed into a sub-Brussels bar objective function.

[0010] An improved third-generation non-dominated sorting genetic algorithm is used to solve the multi-objective multi-water source collaborative scheduling model: an engineering-oriented strategy is used to generate the initial population, an adaptive penalty mechanism of dynamic doubling is introduced when calculating fitness, and finally the iteration is terminated based on the set dual indicators to obtain the optimal decision variables.

[0011] Furthermore, the multi-objective, multi-water-source collaborative scheduling model includes an objective function for expected water shortage, an objective function for expected operating costs, and an objective function for the risk of saltwater intrusion tail:

[0012] ;

[0013] ;

[0014] ;

[0015] Where s represents the scenario, Let represent the objective function for the expected water shortage in scenario s. This represents the objective function for the expected running cost in scenario s. This represents the objective function for the tail risk of saltwater intrusion in scenario s; Represents the set of scheduling decision periods: This is the water shortage penalty term for time period t. Let be the operating cost for time period t. The risk loss due to saltwater intrusion in time period t is calculated based on the deviation between the chloride concentration monitored at the reservoir intake and the safety threshold. , , Let be the chloride concentration value at the reservoir intake during time period t. The expression is:

[0016] ;

[0017] ;

[0018] in, Let z be the total water demand forecast for water-demand area z during time period t. Let z be the actual total water demand of water-demanding area z during time period t. This represents the water shortage gap in water-demanding area z during time period t. This is a collection of areas requiring water.

[0019] Furthermore, the objective functions for expected water shortage and expected operating cost are analyzed using the following formulas:

[0020] ;

[0021] in, This represents the i-th objective function, when i=1. Let i be the objective function for the desired water shortage. When i=2, Let the objective function be the desired operating cost. Let be the mean of the corresponding objective function. For the i-th objective function, the sub-bar aggregation form is... Robust scaling function characterizing sample dispersion and amplitude:

[0022] ;

[0023] Where N represents the total number of scenarios;

[0024] For the objective function of saltwater intrusion tail risk, a conditional value risk based on the Bruker bar is introduced to quantify the safety margin:

[0025] ;

[0026] in, For a given confidence level, The final aggregated target value for the risk of saltwater intrusion tails in the Brussels Basin. To extract from various scenarios The tail sample values; and These respectively characterize the dispersion and magnitude of the risk sample distribution. and All are preset amplification factors that control the robustness depth.

[0027] Furthermore, considering the effective utilization of water resources, supply and demand balance, and physical boundaries, the following constraints are set:

[0028] Reservoir water supply capacity constraints:

[0029] ;

[0030] in, For time period t, the distance from the reservoir to the water plant is... Water supply Indicates a collection of water plants. This represents the maximum water supply capacity of reservoir r during time period t;

[0031] Water plant water supply capacity constraints:

[0032] ;

[0033] in, This represents the water volume delivered from water plant p to the water demand area z during time period t. Collected for water-demand areas, Let p be the maximum water supply capacity of water plant during time period t.

[0034] Pipeline transmission and distribution capacity constraints:

[0035] ;

[0036] ;

[0037] in, and These represent the maximum flow capacity of the corresponding water supply pipeline;

[0038] Water plant water balance constraints:

[0039] ;

[0040] Nonnegativity constraint:

[0041] ;

[0042] .

[0043] Furthermore, it adaptively adjusts according to different modes. ;

[0044] If the current hydrological forecast indicates no saltwater intrusion warning in the near future, and the predicted chloride concentration Q at the intakes of each reservoir remains stable and , If the current effective water storage capacity of the main reservoir is higher than the preset safety threshold, then the reservoir is currently in economic mode. The value range is [0, 0.01]. The objective function of expected operating cost has a higher priority than the objective function of expected water shortage and the objective function of saltwater intrusion tail risk.

[0045] If the reservoir is currently in a normal hydrological tidal cycle and the chloride concentration at the water intake is U±u, where u is a preset value, then the reservoir is currently in equilibrium mode. The value range is [0.01, 0.05]; the three objective functions have the same priority.

[0046] If current hydrological forecasts indicate saltwater intrusion in the near future, and this continues for a considerable period... If so, it is currently in conservative mode. The value range is [0.05, 0.1]. The expected operating cost objective function and the saltwater tail risk objective function have higher priority than the expected water shortage objective function.

[0047] Furthermore, the improved third-generation non-dominated sorting genetic algorithm is as follows:

[0048] Step 1: Jointly encode the water supply set from the reservoir to the water plant and the water transfer set from the water plant to the water demand area into a one-dimensional real number chromosome. The theoretical length of the chromosome is... , The total number of reservoirs The total number of water plants. Indicates the total number of scheduling decision periods. The total number of water-required areas is represented by the value range of each gene locus in the chromosome, which is strictly limited to the closed interval of the corresponding maximum water supply or maximum water delivery.

[0049] Step 2: Generate the initial population using an engineering-oriented strategy;

[0050] Step 3: Calculate multi-objective fitness and introduce an adaptive penalty mechanism with dynamic doubling;

[0051] Step 4: Crossover and mutation work together;

[0052] Step 5: Environmental selection and population renewal;

[0053] Step 6: Determine when to terminate the iteration;

[0054] Step 7: After the iteration terminates, a non-dominated solution set is obtained. Select the comprehensive optimal inflection point solution and the single sub-bar objective function optimal boundary solution from the non-dominated solution set. Use the inflection point solution and the boundary solution as representative schemes, and forcibly correct infeasible solutions that violate the physical boundaries of the reservoir or pipeline network in the representative schemes.

[0055] Furthermore, the engineering-oriented strategy is as follows:

[0056] Strategy 1, Water Supply Capacity Balancing Strategy: Distribute raw water according to the proportion of each water plant's maximum water supply capacity.

[0057] ;

[0058] in, Let p be the predicted total raw water demand of water plant in time period t. For strategy 1, the water flow from reservoir r to water plant in time period t. Water supply This represents the maximum water supply capacity of reservoir r in time period t. For reservoir collection;

[0059] Strategy 2, Historical Scheduling Fitting Strategy: Fit the measured scheduling data of the same period in the previous several years in the area affected by saltwater intrusion and add random perturbation to generate an initial solution;

[0060] ;

[0061] in, For strategy 2, the distance from reservoir r to water plant in time period t. Water supply For the historical period t, the water level at the reservoir reached the water plant. Water supply This is random disturbance noise;

[0062] Strategy 3, Reservoir Balancing Strategy: Minimize the variance of the utilization rate of each reservoir. To achieve dynamic balance, raw water is allocated according to the proportion of the current remaining water supply capacity of each reservoir. The logic is as follows:

[0063] ;

[0064] in, For strategy 3, the distance from reservoir r to water plant in time period t. Water supply

[0065] Strategy 4, Demand-Oriented Strategy: Allocate water plant output in reverse order based on the predicted water demand percentage for each water-demand area.

[0066] ;

[0067] in, Let this be the water transfer volume from water plant p to water demand area z in time period t of strategy 4. Let p be the maximum water supply capacity of water plant during time period t. Let z be the total water demand forecast for water-demand area z during time period t. For collection of water demand areas;

[0068] Strategy 5, Regional Priority Strategy: Allocate water according to the administrative level and population density of the water demand area;

[0069] ;

[0070] in, Let this be the water transfer volume from water plant p to water demand area z in time period t of strategy 5. Weighting coefficients are set based on administrative rank;

[0071] Strategy 6, a hybrid strategy, against , as well as Weighted fusion is performed to generate an initial population corresponding to the water supply from the reservoir to the water plant. as well as Weighted fusion is performed to generate an initial population corresponding to the water supply from the water plant to the water demand area. If the scheduling deviation rate of a certain strategy in historical verification is higher than 10%, its corresponding weight is adaptively reduced by 5%-10% and proportionally increased to other strategies.

[0072] Furthermore, the value of the multi-objective objective function is used as the original fitness value, and the multi-objective fitness is calculated using the following formula:

[0073] ;

[0074] in, This represents the fitness value corresponding to the objective function of the i-th sub-bar. At that time, Let be the fitness value corresponding to the objective function of the desired water deficit. At that time, Let the fitness value be the objective function corresponding to the desired operating cost. At that time, The fitness value corresponds to the objective function for the risk of saltwater intrusion tail; superscript Represents the original fitness value. This is the corrected fitness value. Water shortage in water-demanding areas exceeds their total demand. Exceeding the limit The reservoir is supplying water beyond its maximum capacity. For the excess limits, I and E are constants; The supply and demand balance penalty coefficient, This is the capacity constraint penalty coefficient; when the water shortage in a water-demanding area exceeds its total demand. hour, The value doubles; when the reservoir's water supply exceeds its maximum capacity. hour, Double.

[0075] Furthermore, based on the average reverse generation distance MIGD and the effective hypervolume variation coefficient... The two indicators determine the timing of iteration termination:

[0076] The three-dimensional Euclidean distance between the non-dominated solution set of the current population and the theoretical optimal solution set of the current iteration is calculated based on multi-objective fitness. The three-dimensional Euclidean distances are summed to obtain MIGD.

[0077] Using the minimum zero point as the origin and several times the worst historical value of each objective function as the boundary reference point, the overlapping regions in the cuboid space enclosed by the non-dominated solutions in the current non-dominated solution set and the boundary reference points are eliminated. Then, the cuboid space is divided into m independent effective hypervolume elements, and the average volume of the effective hypervolume elements is calculated. and volume standard deviation Calculate the coefficient of variation :

[0078] ;

[0079] If several consecutive generations occur during the iteration process Smaller than the preset Threshold, and Greater than the preset If the threshold is reached, the iteration terminates.

[0080] A reservoir adaptive scheduling system for saltwater intrusion response based on distributed bar optimization includes:

[0081] The topology module is used to abstract reservoirs, water plants and water demand areas in the target area affected by saltwater intrusion as directed graph nodes, the connection relationship between nodes as directed edges, and the flow on the directed edges as decision variables, thereby constructing a multi-source-multi-water plant-multi-water demand area stratified water supply topology.

[0082] The collaborative scheduling model module is used to establish a multi-objective, multi-source collaborative scheduling model based on the multi-source, multi-water-plant, multi-demand area stratified water supply topology, and to set constraints for decision variables;

[0083] The debulking bar aggregation evaluation module is used to construct Wasserstein fuzzy sets, while introducing a dynamically robust radius parameter. The original objective function in the multi-objective, multi-source collaborative scheduling model is evaluated by sub-Brussels bar aggregation, and the original objective function is transformed into a sub-Brussels bar objective function.

[0084] The optimization calculation module uses an improved third-generation non-dominated sorting genetic algorithm to solve the multi-objective multi-water source collaborative scheduling model: an engineering-oriented strategy is used to generate the initial population, an adaptive penalty mechanism with dynamic doubling is introduced when calculating fitness, and finally the iteration is terminated based on the set dual indicators to obtain the optimal decision variables.

[0085] Beneficial effects: 1. This invention reconstructs the objective function through the Wasserstein fuzzy set sub-bar framework, which completely overcomes the overfitting defect of traditional deterministic or stochastic programming models when encountering data distribution shifts caused by extreme saline intrusion, and greatly improves the system's resilience against risks.

[0086] 2. This invention constructs a complete three-layer network mathematical expression for water source, water plant, and water demand area, and introduces an engineering-oriented strategy and dynamic penalty and repair mechanism to deeply improve the NSGA-III (third generation non-dominated sorting genetic) algorithm, ensuring that each output scheduling scheme strictly satisfies the physical boundary and has engineering feasibility.

[0087] 3. This invention ensures efficient convergence of the algorithm solution through dual-index dynamic monitoring and ultimately generates an intuitive decision-making operation guide, providing water management departments with a complete decision support closed loop from accurate early warning to intelligent execution in response to complex hydrological uncertainties. Attached Figure Description

[0088] Figure 1 This is the overall flowchart of the present invention;

[0089] Figure 2 A schematic diagram of Wasserstein fuzzy sets;

[0090] Figure 3 For robust radius Vertical plot of the effects on the three objective functions; where (a) is the robust radius. Vertical plot of the effect on the objective function of expected water deficit, (b) is the robust radius. Vertical plot of the impact on the expected operating cost objective function, (c) is the robust radius. Vertical plot of the impact of the objective function on the risk of saltwater intrusion tail;

[0091] Figure 4 Box plot of total water deficit for SP and WDRO in out-of-sample testing. Detailed Implementation

[0092] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0093] This embodiment uses a multi-reservoir combined water supply system in the Yangtze River Estuary as an example. The Yangtze River Estuary has long faced the serious threat of saltwater intrusion. In recent years, influenced by extreme weather events, saltwater intrusion has exhibited significant non-stationarity and suddenness. For example, in the autumn and winter of 2022, the Yangtze River Estuary experienced a historically rare combination of extreme drought and saltwater intrusion, leading to a severe shift in the distribution of hydrological data. Current conventional stochastic programming (SP) models, due to their rigid adherence to the static assumption that historical experience distributions equal future actual distributions, are prone to overfitting when faced with such extreme distribution shifts, resulting in uncontrolled water shortage risks at the tail end. Therefore, it is urgent to introduce a divergent bloom optimization (WDRO) framework to achieve synergistic optimization of water supply reliability, economic efficiency, and tail end water quality safety.

[0094] In this embodiment, as Figure 1 As shown in the overall flowchart, the adaptive scheduling method for multiple reservoirs in the estuary to cope with saltwater intrusion based on distributed bar optimization, using real water supply and hydrological data from the Yangtze River estuary as the application scenario, specifically includes the following steps:

[0095] Step S1: Construct a multi-source, multi-water treatment plant, and multi-demand stratified water supply topology covering the Yangtze River estuary region affected by saltwater intrusion, and define decision variables, specifically including:

[0096] Step S1.1: In this embodiment, a three-layer network structure is constructed. The water sources include Chenxing Reservoir, Qingcaosha Reservoir, and Jinze Reservoir, totaling three core reservoirs. Based on actual physical parameters: Chenxing Reservoir has an effective storage capacity of 1 million to 7.34 million m³, with a maximum water supply capacity of 1.6 million m³ / d; Qingcaosha Reservoir has an effective storage capacity of 50 million to 336 million m³, with a maximum water supply capacity of 4.7 million m³ / d; and Jinze Reservoir has an effective storage capacity of 1 million to 8.17 million m³, with a maximum water supply capacity of 2.1 million m³ / d.

[0097] Step S1.2: Collect data on actual daily water demand, reservoir and water plant capacity, water level-reservoir capacity relationship, and chloride concentration at water intake points for each district. The constructed multi-scenario time-series dataset is strictly divided into training scenarios (70%) and out-of-sample testing (30%). The decision variable is determined as the water supply from the water source to the water plant cluster.

[0098] Water supply from water source to water plant:

[0099] X={ | };

[0100] In the formula, R is the set of reservoirs, R = {Chenxing Reservoir, Qingcaosha Reservoir, Jinze Reservoir}, P is the set of water plants, P = {Northern Water Plant Group, Central Water Plant Group, Pudong Water Plant Group, Southern Water Plant Group, Western Water Plant Group, Songjin Water Plant Group, Suburban Water Plant Group}, and T is the set of scheduling periods.

[0101] Water flow from the water plant to the demand area:

[0102] Y={ ;

[0103] In the formula, Let Z be the set of water-demanding areas, where Z = {central urban area, northern area, western area, southern area}; This indicates the amount of water that water plant p delivers to the water-demanding area z during time period t.

[0104] Step S2: Based on the topology and dataset constructed in Step S1, establish a three-dimensional multi-objective cooperative scheduling model that includes expected water shortage, expected operating cost, and risk of saline intrusion tail at the basalt estuary, and set physical engineering constraints; specifically including:

[0105] Step S2.1: Construct the scenario-level objective function for the multi-objective, multi-water-source optimization scheduling model:

[0106] Objective function 1: Expected water shortage objective function (The objective function for the expected water shortage in this embodiment is the objective function for the expected water shortage under the worst-case scenario):

[0107] ;

[0108] In the formula, Let be the water shortage target assessment value under scenario s (i.e., sample), where , It is a collection of time-series datasets generated based on historical data, containing different available water volumes and saltwater intrusion intensities; This indicates the unified scheduling decision period (day). This is the water shortage penalty term for time period t, i.e. ,in , Let z be the predicted total water demand of the water-demanding region in time period t. The actual total water demand of water-required area z during time period t.

[0109] Objective Function 2: Expected Operating Cost Objective Function:

[0110] ;

[0111] In the formula, This represents the cost target assessment value under scenario s; Let t be the operating cost (CNY) for period t.

[0112] Objective Function 3: Objective function for the tail risk of saltwater intrusion in the Brussels basin:

[0113] ;

[0114] In the formula, The risk assessment value for saltwater intrusion under scenario s; The risk loss due to saltwater intrusion in time period t is calculated based on the deviation between the chloride concentration (mg / L) monitored at the reservoir intake and the safety threshold of 250 mg / L. , The value represents the predicted chloride concentration at the reservoir intake during time period t, which is affected by saltwater intrusion.

[0115] Step S2.2: Considering the effective utilization of water resources, supply and demand balance, and physical boundaries, the following constraints are set:

[0116] ① Reservoir water supply capacity constraints:

[0117] ;

[0118] In the formula, Let r be the maximum water supply capacity of reservoir r in time period t (m³).

[0119] ② Constraints on the water supply capacity of the water plant:

[0120] ;

[0121] In the formula, Let p be the maximum water supply capacity (m³) of water plant group p in time period t.

[0122] ③ Pipeline transmission and distribution capacity constraints:

[0123] ;

[0124] ;

[0125] In the formula, and These represent the maximum flow capacity (m³) of the corresponding water pipeline.

[0126] ④ Water plant water balance constraints:

[0127] ;

[0128] ⑤ Non-negativity constraint:

[0129] ;

[0130] .

[0131] Step S3: Construct a Wasserstein fuzzy set centered on the historical experience distribution, and introduce a dynamic robust radius parameter. The multi-objective collaborative scheduling model is evaluated by sub-bar aggregation.

[0132] Combination Figure 2 As shown in the schematic diagram of Wasserstein fuzzy sets, when When the fuzzy set degenerates into a single empirical distribution center, the model is equivalent to traditional stochastic programming (SP); while when The system then expands a Wasserstein spatial sphere centered on the empirical distribution and searches for the probability distribution leading to the worst water supply condition on the boundary of this sphere. In this embodiment, robust aggregation is performed on the water shortage and cost targets:

[0133] ;

[0134] in, This represents the i-th objective function, when i=1. Let i be the objective function for the desired water shortage. When i=2, Let the objective function be the desired operating cost. Let be the mean of the corresponding objective function. For the i-th objective function, the sub-bar aggregation form is... A robust scaling function characterizing the sample dispersion and amplitude:

[0135] ;

[0136] in, .

[0137] To address the risk of saline tide tails under high salinity conditions, a risk formula based on the conditional value of the Bruker bar is introduced to quantify the safety margin:

[0138] ;

[0139] in, For a given confidence level, The final aggregated target value for the risk of saltwater intrusion tails in the Brussels Basin. To extract from various scenarios The tail sample values; and These respectively characterize the dispersion and magnitude of the risk sample distribution. and All are preset amplification factors that control the robustness depth.

[0140] when At that time, the model is equivalent to traditional stochastic programming (SP); as As the magnitude of the magnitude increases, the model's conservatism in resisting unknown risks also increases.

[0141] Step S4: Solve using the improved NSGA-III algorithm. To overcome the problem of easily getting trapped in local optima and the lack of engineering guidance in conventional high-dimensional multi-objective optimization, the improvement of the classic NSGA-III algorithm in this invention is mainly reflected in three core aspects: ① diversified population initialization combined with engineering-oriented strategies; ② adaptive penalty function with dynamic doubling mechanism to handle constraint violations; ③ based on the average backward generation distance (MIGD) and average hypervolume (MHV) coefficient of variation. Dynamic termination criteria for dual indicators.

[0142] In the parameter settings and operation of this embodiment, the population size is set to 100, and the maximum number of generations is relaxed to 500 to match the dynamic termination criterion. When generating the initial population using the historical scheduling fitting strategy at level ①, the decision variables of some individuals are forcibly incorporated into the experience-based joint scheduling ratio benchmark for water sources. For example, under normal non-extreme salinity conditions, the priority allocation of water supply loads for Qingcaosha, Jinze, and Chenxing reservoirs is 7:2:1. Using this as heuristic prior knowledge greatly improves the feasibility and search efficiency of the initial solution in the real physical pipe network. Subsequently, the search process is continuously optimized through adaptive crossover, mutation operators, and feasibility repair operators based on engineering heuristic rules. When MIGD and [other conditions] are satisfied for 50 consecutive generations, the search is completed. When the convergence threshold is reached, the system triggers the termination mechanism, and finally outputs a fully converged Pareto optimal solution set in the three-dimensional target space.

[0143] Physical repair is triggered for infeasible solutions resulting from algorithm mutations. In this embodiment, the water supply allocated to Chenxing Reservoir by a candidate solution exceeds its capacity limit of 1.6 million m³ / d. The repair operator will forcibly reduce the excess flow and, based on geographical topology, shift the supply to Qingcaosha Reservoir, which has a stronger resistance to salinity, thus rigidly restoring the engineering legality of the solution.

[0144] Step S4.1: Encoding Decision Variables: This involves jointly encoding the set of water supply from the water source to the water treatment plant group and the set of water transfer from the water treatment plant group to the water demand area into a one-dimensional real-number chromosome. The theoretical length of the chromosome is... The value range of each gene locus is strictly limited to the corresponding closed interval of maximum water supply or maximum water delivery.

[0145] Step S4.2: Diversified initial population generation: To avoid the initial solution getting trapped in local optima and to ensure engineering feasibility, seven engineering-oriented strategies are weighted and fused to generate an initial population of 100-150.

[0146] against Three generation strategies:

[0147] Water supply capacity balancing strategy (strategy 1): Allocate raw water according to the proportion of the maximum water supply capacity of each water source to generate candidate solutions. The logic is as follows:

[0148] ;

[0149] In the formula, Let p be the predicted total raw water demand of water plant in time period t. For strategy 1, the water flow from reservoir r to water plant in time period t. The allocation amount.

[0150] Historical scheduling fitting strategy (strategy 2): Fit historical measured scheduling quantities for the same period and add random perturbations to generate candidate solutions. The logic is as follows:

[0151] ;

[0152] In the formula, For strategy 2, the distance from reservoir r to water plant in time period t. The allocation amount, For the historical period t, the water level at the reservoir reached the water plant. The allocation amount, This is random perturbation noise.

[0153] Reservoir balancing strategy (strategy 3): The objective is to minimize the reservoir utilization rate at each water source. To determine the variance of the water source and achieve dynamic equilibrium, raw water is allocated according to the proportion of the current remaining water supply capacity of each reservoir, generating candidate solutions. The logic is as follows:

[0154] ;

[0155] In the formula, the numerator represents the remaining dispatch capacity of the reservoir in the current period. For strategy 3, the distance from reservoir r to water plant in time period t. The allocation amount.

[0156] against Two generation strategies:

[0157] Demand-oriented strategy (Strategy 4): Distribute water plant output in reverse order based on the predicted water demand percentage of different water-demand areas. Generate candidate solutions. The logic is as follows:

[0158] ;

[0159] in, Let this be the amount of water allocated from water plant p to water demand area z in time period t of strategy 4. Let p be the maximum water supply capacity of water plant during time period t. Let z be the total water demand forecast for water-demand area z during time period t. For collection of water demand areas;

[0160] Regional Priority Strategy (Strategy 5): Allocate water demand areas according to their administrative level and population density to generate candidate solutions. The logic is as follows:

[0161] ;

[0162] in These are weighting coefficients set based on administrative level.

[0163] Hybrid Strategy (Adaptive Weighted Fusion, Strategy 6): Employs a dual-channel weighted fusion formula to generate initial individuals independently for variables X and Y:

[0164] ;

[0165] ;

[0166] In the formula and These are the dynamic fusion weights for generating strategies corresponding to variables X and Y (satisfying Σβ=1). If the scheduling deviation rate of a certain strategy in historical verification is higher than 10%, the system adaptively reduces its corresponding weight by 5%-10% and increases it proportionally to other strategies.

[0167] Step S4.3: Fitness Evaluation and Constraint Handling: The aggregation results of the three sub-bars are used as candidate solutions x. The original multi-objective fitness in the evolutionary algorithm is calculated, and a dynamic penalty function is introduced to correct the fitness. The correction formula is as follows: ;

[0168] In the formula, Represents chromosomes. This represents the fitness value corresponding to the objective function of the i-th sub-bar. At that time, Let be the fitness value corresponding to the objective function of the desired water deficit. At that time, Let the fitness value be the objective function corresponding to the desired operating cost. At that time, The fitness value corresponds to the objective function for the risk of saltwater intrusion tail; superscript Represents the original fitness value. This is the corrected fitness value. Water shortage in water-demanding areas exceeds their total demand. Exceeding the limit The reservoir is supplying water beyond its maximum capacity. For the excess limits, I and E are constants; The supply and demand balance penalty coefficient, This is the capacity constraint penalty coefficient; when the water shortage in a water-demanding area exceeds its total demand. hour, The value doubles; when the reservoir's water supply exceeds its maximum capacity. hour, Double.

[0169] Step S4.4: Crossover and mutation co-operation: The population iteration is driven by simulated binary crossover (SBX) and polynomial mutation strategy.

[0170] Step S4.5: Environment selection and population update: Merge the parent and offspring populations and perform fast non-dominated sorting, and perform target normalization screening at reference points that are pre-distributed evenly in the three-dimensional target space.

[0171] Step S4.6: Dynamically monitor the convergence and solution set distribution of the algorithm to determine the timing for terminating the iteration.

[0172] Based on mean reverse generation distance MIGD and effective hypervolume variation coefficient The two indicators determine the timing of iteration termination:

[0173] The three-dimensional Euclidean distance between the non-dominated solution set of the current population and the theoretical optimal solution set of the current iteration is calculated based on multi-objective fitness. The three-dimensional Euclidean distances are summed to obtain MIGD.

[0174] Let the non-dominated solution set of the current evolutionary population be:

[0175] ;

[0176] The theoretical optimal solution set or reference set is:

[0177] ;

[0178] For the solution set The i-th candidate solution Calculate its reference set The three-dimensional Euclidean distance of the nearest solution:

[0179] ;

[0180] calculate :

[0181] .

[0182] Using the minimum zero point as the origin and several times the worst historical value of each objective function as the boundary reference point, the overlapping regions in the cuboid space enclosed by the non-dominated solutions in the current non-dominated solution set and the boundary reference points are eliminated. Then, the cuboid space is divided into m independent effective hypervolume elements, and the average volume of the effective hypervolume elements is calculated. and volume standard deviation Calculate the coefficient of variation :

[0183] Let its volume set be:

[0184] ;

[0185] Calculate the average volume of the effective cells:

[0186] ;

[0187] And volume standard deviation:

[0188] ;

[0189] Finally, the hypervolume variation coefficient was obtained:

[0190] ;

[0191] Termination criterion: When the evolutionary iteration satisfies fifty consecutive generations... and When the system determines that the current non-dominated solution set has fully converged and uniformly covered the three-dimensional target space, the system terminates the evolution and outputs the Pareto optimal solution set.

[0192] Step S5: From the non-dominated solution set, select the comprehensive optimal inflection point solution and the boundary solution with a single objective function based on the minimum Euclidean distance method with respect to the ideal point. Use these as representative scheduling schemes to perform constraint consistency repair and forcibly correct infeasible solutions that violate the physical boundaries of the reservoir or pipeline network.

[0193] Step S6: For multiple robust radius parameters Batch experiments were conducted to output comprehensive evaluation data such as Pareto front comparison charts under different robust radii, system water supply guarantee rate, and saltwater intrusion response costs. A decision matrix and scheduling operation guidelines were also constructed to guide actual scheduling.

[0194] For multiple robust radius parameters Perform batch experiments, output comprehensive evaluation data under different robustness levels, and construct a decision matrix and scheduling operation guidelines to guide actual scheduling.

[0195] In this embodiment, the system is in Batch validation was performed under four gradients. Table 1 summarizes the core performance evaluation results of the algorithm under different blobs conservatism.

[0196] Table 1. Core Performance Evaluation Results of the Yangtze River Estuary Regional Water Supply System

[0197]

[0198] Combination Figure 3 (a), (b), and (c) (robust radius) The combined analysis of the impact on the three objective functions and the data in Table 1 shows that: At that time, because the model cannot identify extreme uncertainty, its expected cost estimate under the worst distribution appears to be the lowest; however, as... Gradually increase from 0.01 to 0.1, Figure 3 (b) shows that the expected cost exhibits a clear non-linear upward trend, with costs climbing; at the same time, the model's prediction of the worst-case expected water shortage and saltwater intrusion tail risk also expands accordingly. This vertical graph evolution pattern profoundly reveals the core physical significance of WDRO: the system actively identifies a deeper threat of data distribution shift, and therefore automatically adds scheduling weight reservations for high-cost, high-security water sources (Qingcaosha Reservoir) in the mathematical model.

[0199] To verify the actual defense capability of the above WDRO model under real extreme working conditions, combined with Figure 4 The box plots showing the total water deficit of SP and WDRO in out-of-sample tests are used for illustration. Figure 4 The statistical distribution characteristics and evolution patterns of the model in response to unknown hydrological data are intuitively presented: In the box plot of the traditional SP scheme, although the median of the water deficit is relatively low, its overall fluctuation range is extremely large, with a long upper and lower edge span; this high dispersion, mapped to the physical world, means that when encountering unexpected extreme salinity periods, traditional stochastic algorithms are easily overwhelmed by tail risk, triggering an uncontrollable systemic water shortage crisis. Conversely, when the WDRO model of this invention is introduced and a suitable robust radius (such as...) is set... and When the system water shortage reaches a certain threshold, the box plot exhibits risk compression and peak shaving characteristics. Although the median water shortage is slightly higher due to the safety margin, its interquartile range (box height), representing uncertainty, is significantly narrowed, and the upper edge representing extreme conditions is significantly lowered and stabilized. Furthermore, Figure 4 It also reveals the limiting characteristics of the system's defense boundary: when the robustness radius is set too extreme (such as...). In the extremely conservative mode, not only does the median of the container continue to rise, but its upper edge even rebounds; this indicates that excessive pursuit of decentralization will lead to over-defense of the system, which will exacerbate the overall water shortage pressure due to resource misallocation. This data pattern based on real data provides irrefutable engineering and physical support for the reasonable selection of parameter ranges in the adaptive operation guidelines for actual scheduling.

[0200] Finally, based on the aforementioned multidimensional trade-off data, the system generated the adaptive operational decision guide shown in Table 2:

[0201] Table 2. Adaptive Operation Decision-Making Guidelines for Multi-Reservoir Systems Under the Influence of Saltwater Intrusion

[0202]

[0203] The table shows that the risk of saltwater intrusion is low and the water supply security margin is sufficient. Specifically, hydrological forecasts indicate that there will be no saltwater intrusion warnings in the next 7-15 days, and the predicted chloride concentrations at the water intakes of the Yangtze River Estuary are stable and far below the safety line of 250 mg / L. At the same time, the current effective water storage of the major reservoirs is at a high level, and the system has extremely strong self-sufficiency and resistance to breaches.

[0204] The system is within a normal fluctuation range and supply and demand are controllable. Specifically, it is in a normal hydrological and tidal cycle with short-term mild salinity fluctuations. The chloride concentration at the water intake may hover around 250 mg / L, but the duration of exceeding the standard is extremely short. The dynamic joint regulation of the existing reservoirs is sufficient to absorb the fluctuations and will not cause a substantial water supply gap.

[0205] The specific reasons for triggering early warnings of saltwater intrusion or increasing risk of continuous water shortage are as follows: meteorological and hydrological forecasts indicate that a severe or extreme saltwater intrusion is about to occur, and the chloride concentration at the water intake will exceed 250 mg / L for several consecutive days; in addition, the effective storage capacity of reservoirs with weak conventional saltwater resistance will decrease sharply, the number of self-sustaining days will approach the shutdown threshold, and the system will soon face the threat of large-scale water outages.

[0206] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A reservoir adaptive scheduling method for dealing with saltwater intrusion based on distributed bar optimization, characterized in that, Specifically: The reservoirs, water plants, and water demand areas in the target area affected by saltwater intrusion are abstracted as nodes in a directed graph, and the connection relationships between nodes are abstracted as directed edges. The flow on the directed edges is used as a decision variable, thereby constructing a multi-source-multi-water plant-multi-water demand area stratified water supply topology. The decision variables include the water supply from the reservoir to the water plant and the water transfer from the water plant to the water demand area. Based on the stratified water supply topology of multiple water sources, multiple water plants, and multiple water demand areas, a multi-objective multi-water source collaborative scheduling model is established, and constraints are set for decision variables. Construct Wasserstein fuzzy sets and introduce dynamically robust radius parameters. The original objective function in the multi-objective, multi-source collaborative scheduling model is evaluated by sub-Brussels bar aggregation, and the original objective function is transformed into a sub-Brussels bar objective function. An improved third-generation non-dominated sorting genetic algorithm is used to solve the multi-objective multi-water source collaborative scheduling model: an engineering-oriented strategy is used to generate the initial population, an adaptive penalty mechanism of dynamic doubling is introduced when calculating fitness, and finally the iteration is terminated based on the set dual indicators to obtain the optimal decision variables. The multi-objective, multi-water-source collaborative scheduling model includes the objective function of expected water shortage, the objective function of expected operating cost, and the objective function of saltwater intrusion tail risk: ； ； ； Where s represents the scenario, Let represent the objective function for the expected water shortage in scenario s. This represents the objective function for the expected running cost in scenario s. This represents the objective function for the tail risk of saltwater intrusion in scenario s; Represents the set of scheduling decision periods: This is the water shortage penalty term for time period t. Let be the operating cost for time period t. The risk loss due to saltwater intrusion in time period t is calculated based on the deviation between the chloride concentration monitored at the reservoir intake and the safety threshold. , , Let be the chloride concentration value at the reservoir intake during time period t. The expression is: ； ； in, Let z be the total water demand forecast for water-demand area z during time period t. Let z be the actual total water demand of water-demanding area z during time period t. This represents the water shortage gap in water-demanding area z during time period t. For collection of water demand areas; The improved third-generation non-dominated sorting genetic algorithm is as follows: Step 1: Jointly encode the water supply set from the reservoir to the water plant and the water transfer set from the water plant to the water demand area into a one-dimensional real number chromosome. The theoretical length of the chromosome is... , The total number of reservoirs, The total number of water plants. Indicates the total number of scheduling decision periods. The total number of water-required areas is represented by the value range of each gene locus in the chromosome, which is strictly limited to the closed interval of the corresponding maximum water supply or maximum water delivery. Step 2: Generate the initial population using an engineering-oriented strategy; Step 3: Calculate multi-objective fitness and introduce an adaptive penalty mechanism with dynamic doubling; Step 4: Crossover and mutation work together; Step 5: Environmental selection and population renewal; Step 6: Determine when to terminate the iteration; Step 7: After the iteration terminates, a non-dominated solution set is obtained. Select the comprehensive optimal inflection point solution and the single sub-bar objective function optimal boundary solution from the non-dominated solution set. Use the inflection point solution and the boundary solution as representative schemes, and forcibly correct infeasible solutions that violate the physical boundaries of the reservoir or pipeline network in the representative schemes.

2. The adaptive scheduling method for reservoirs to cope with saltwater intrusion based on bibliometric bar optimization according to claim 1, characterized in that, The objective functions for expected water shortage and expected operating cost are analyzed using the following formulas: ； in, This represents the i-th objective function, when i=1. Let i be the objective function for the desired water shortage. When i=2, Let the objective function be the desired operating cost. Let be the mean of the corresponding objective function. For the i-th objective function, the sub-bar aggregation form is... Robust scaling function characterizing sample dispersion and amplitude: ； Where N represents the total number of scenarios; For the objective function of saltwater intrusion tail risk, a conditional value risk based on the Bruker bar is introduced to quantify the safety margin: ； in, For a given confidence level, The final aggregated target value for the risk of saltwater intrusion tails in the Brussels Basin. To extract from various scenarios The tail sample values; and These respectively characterize the dispersion and magnitude of the risk sample distribution. and All are preset amplification factors that control the robustness depth.

3. The adaptive scheduling method for reservoirs to cope with saltwater intrusion based on bibliometric bar optimization according to claim 1, characterized in that, Considering the effective utilization of water resources, supply and demand balance, and physical boundaries, the following constraints are set: Reservoir water supply capacity constraints: ； in, For time period t, the distance from the reservoir to the water plant is... Water supply Indicates a collection of water plants. This represents the maximum water supply capacity of reservoir r during time period t; Water plant water supply capacity constraints: ； in, This represents the water volume delivered from water plant p to the water demand area z during time period t. Collected for water-demand areas, Let p be the maximum water supply capacity of water plant during time period t. Pipeline transmission and distribution capacity constraints: ； ； in, and These represent the maximum flow capacity of the corresponding water supply pipeline; Water plant water balance constraints: ； Nonnegativity constraint: ； 。 4. The adaptive scheduling method for reservoirs to cope with saltwater intrusion based on bibliometric bar optimization according to claim 1, characterized in that, Adaptive adjustment based on different modes ; If the current hydrological forecast indicates no saltwater intrusion warning in the near future, and the predicted chloride concentration Q at the intakes of each reservoir remains stable and , If the current effective water storage capacity of the main reservoir is higher than the preset safety threshold, then the reservoir is currently in economic mode. The value range is [0, 0.01]. The objective function of expected operating cost has a higher priority than the objective function of expected water shortage and the objective function of saltwater intrusion tail risk. If the reservoir is currently in a normal hydrological tidal cycle and the chloride concentration at the water intake is U±u, where u is a preset value, then the reservoir is currently in equilibrium mode. The value range is [0.01, 0.05]; The three objective functions have the same priority; If current hydrological forecasts indicate saltwater intrusion in the near future, and this continues for a considerable period... If so, it is currently in conservative mode. The value range is [0.05, 0.1]. The expected operating cost objective function and the saltwater tail risk objective function have higher priority than the expected water shortage objective function.

5. The adaptive scheduling method for reservoirs to cope with saltwater intrusion based on bibliometric bar optimization according to claim 1, characterized in that, The engineering-oriented strategy is: Strategy 1, Water Supply Capacity Balancing Strategy: Distribute raw water according to the proportion of each water plant's maximum water supply capacity. ； in, Let p be the predicted total raw water demand of water plant in time period t. For strategy 1, the water flow from reservoir r to water plant in time period t. Water supply This represents the maximum water supply capacity of reservoir r in time period t. For reservoir collection; Strategy 2, Historical Scheduling Fitting Strategy: Fit the measured scheduling data of the same period in the previous several years in the area affected by saltwater intrusion and add random perturbation to generate an initial solution; ； in, For strategy 2, the distance from reservoir r to water plant in time period t. Water supply For the historical period t, the water level at the reservoir reached the water plant. Water supply This is random disturbance noise; Strategy 3, Reservoir Balancing Strategy: Minimize the variance of the utilization rate of each reservoir. To achieve dynamic balance, raw water is allocated according to the proportion of the current remaining water supply capacity of each reservoir. The logic is as follows: ； in, For strategy 3, the distance from reservoir r to water plant in time period t. Water supply For the first From reservoir to water plant Water supply; Strategy 4, Demand-Oriented Strategy: Allocate water plant output in reverse order based on the predicted water demand percentage for each water-demand area. ； in, Let this be the water transfer volume from water plant p to water demand area z in time period t of strategy 4. Let p be the maximum water supply capacity of water plant during time period t. Let z be the total water demand forecast for water-demand area z during time period t. Collected for areas requiring water; Strategy 5, Regional Priority Strategy: Allocate water according to the administrative level and population density of the water demand area; ； in, Let this be the water transfer volume from water plant p to water demand area z in time period t of strategy 5. Weighting coefficients are set based on administrative rank; Strategy 6, a hybrid strategy, against , as well as Weighted fusion is performed to generate an initial population corresponding to the water supply from the reservoir to the water plant. as well as Weighted fusion is performed to generate an initial population corresponding to the water supply from the water plant to the water demand area. If the scheduling deviation rate of a certain strategy in historical verification is higher than 10%, its corresponding weight is adaptively reduced by 5%-10% and proportionally increased to other strategies.

6. The adaptive scheduling method for reservoirs to cope with saltwater intrusion based on bibliometric bar optimization according to claim 1, characterized in that, Using the value of the objective function of the multi-objective bar as the initial fitness value, the multi-objective fitness is calculated using the following formula: ； in, This represents the fitness value corresponding to the objective function of the i-th sub-bar. At that time, Let be the fitness value corresponding to the objective function of the desired water deficit. At that time, Let the fitness value be the objective function corresponding to the desired operating cost. At that time, The fitness value corresponds to the objective function for the risk of saltwater intrusion tail; superscript Represents the original fitness value. This is the corrected fitness value. Water shortage in water-demanding areas exceeds their total demand. Exceeding the limit The reservoir is supplying water beyond its maximum capacity. For the excess limits, I and E are constants; The supply and demand balance penalty coefficient, This is the capacity constraint penalty coefficient; when the water shortage in a water-demanding area exceeds its total demand. hour, The value doubles; when the reservoir's water supply exceeds its maximum capacity. hour, Double.

7. The adaptive scheduling method for reservoirs to cope with saltwater intrusion based on bibliometric bar optimization according to claim 1, characterized in that, Based on mean reverse generation distance MIGD and effective hypervolume variation coefficient The dual indicators determine the timing of iteration termination: The three-dimensional Euclidean distance between the non-dominated solution set of the current population and the theoretical optimal solution set of the current iteration is calculated based on multi-objective fitness. The three-dimensional Euclidean distances are summed to obtain MIGD. Using the minimum zero point as the origin and several times the worst historical value of each objective function as the boundary reference point, the overlapping regions in the cuboid space enclosed by the non-dominated solutions in the current non-dominated solution set and the boundary reference points are eliminated. Then, the cuboid space is divided into m independent effective hypervolume elements, and the average volume of the effective hypervolume elements is calculated. and volume standard deviation Calculate the coefficient of variation : ； If several consecutive generations occur during the iteration process Smaller than the preset Threshold, and Greater than the preset If the threshold is reached, the iteration terminates.

8. A system applied to the adaptive scheduling method for reservoirs to cope with saltwater intrusion based on bibliometric bar optimization as described in claim 1, characterized in that, include: The topology module is used to abstract reservoirs, water plants and water demand areas in the target area affected by saltwater intrusion as directed graph nodes, the connection relationship between nodes as directed edges, and the flow on the directed edges as decision variables, thereby constructing a multi-source-multi-water plant-multi-water demand area stratified water supply topology. The collaborative scheduling model module is used to establish a multi-objective, multi-source collaborative scheduling model based on the multi-source, multi-water-plant, multi-demand area stratified water supply topology, and to set constraints for decision variables; The debulking bar aggregation evaluation module is used to construct Wasserstein fuzzy sets, while introducing a dynamically robust radius parameter. The original objective function in the multi-objective, multi-source collaborative scheduling model is evaluated by sub-Brussels bar aggregation, and the original objective function is transformed into a sub-Brussels bar objective function. The optimization calculation module uses an improved third-generation non-dominated sorting genetic algorithm to solve the multi-objective multi-water source collaborative scheduling model: an engineering-oriented strategy is used to generate the initial population, an adaptive penalty mechanism with dynamic doubling is introduced when calculating fitness, and finally the iteration is terminated based on the set dual indicators to obtain the optimal decision variables.

Images