Water resource scheduling optimization method based on empirical model

Abstract

The invention discloses a water resource scheduling optimization method based on an empirical model, and the method comprises the following steps: 1, constructing a dynamic empirical model library, and training a water volume prediction module, a water demand analysis module and an ecological response module based on historical scheduling data; 2, real-time scene feature extraction: extracting feature vectors from meteorological data, hydrological monitoring equipment and ecological remote sensing data; 3, model dynamic migration optimization: selecting an optimal empirical model combination from the dynamic empirical model library through a migration selection function; adjusting parameters of the selected optimal empirical model combination by adopting a parameter migration function; and 4, constructing a double-layer game decision-making tree based on the water volume predicted value, the water demand distribution matrix and the ecological sensitivity coefficient. According to the invention, by integrating the double-layer game decision tree, the multi-objective optimization algorithm and the dynamic parameter adjustment mechanism, an intelligent scheduling system suitable for a complex water network system is constructed, and water resource scheduling optimization is carried out better.

Description

Technical Field

[0001] The present invention relates to the field of water resource scheduling, and in particular to a water resource scheduling optimization method based on an empirical model. Background Art

[0002] Water resource scheduling is the process of regulating and allocating water resources across time and space through hydraulic engineering facilities (such as reservoirs, pumping stations, and water pipelines). Its core goal is to balance diverse water demands for domestic, industrial, and ecological uses, thereby achieving sustainable water resource utilization. As a major country with a long history of water conservancy, my country has historically implemented the rudiments of inter-basin water transfer through projects such as the Dujiangyan Irrigation System and the Beijing-Hangzhou Grand Canal. With economic and social development, water resource scheduling has gradually shifted from demand-driven supply to a sustainable development model, evolving from a single-unit water allocation to a comprehensive regulation that coordinates multiple economic, environmental, and ecological objectives.

[0003] Traditional water resources scheduling has problems such as dynamic and single-objective orientation, insufficient multi-objective coordination, and delayed response to emergencies. Water resources scheduling optimization is needed. Therefore, a water resources scheduling optimization method based on empirical model is proposed. Summary of the Invention

[0004] The present invention solves the problems of the prior art through the following technical solutions, and the present invention includes the following steps: Step 1: Build a dynamic empirical model library and train the water volume prediction module, water demand analysis module, and ecological response module based on historical scheduling data; Step 2: Real-time scene feature extraction, extracting feature vectors from meteorological data, hydrological monitoring equipment, and ecological remote sensing data ; Step 3: Dynamically migrate and optimize the model by migrating the selected function Select the optimal empirical model combination from the dynamic empirical model library; Using parameter transfer function Adjust the parameters of the selected optimal empirical model combination; Step 4: Based on the water volume forecast value, water demand distribution matrix and ecological sensitivity coefficient, a two-layer game decision tree is constructed: The upper layer defines the municipal water supply benefit function, agricultural water supply benefit function and ecological water replenishment benefit function, and uses Nash equilibrium to solve the optimal water supply allocation ratio; The lower layer establishes the flow balance constraints of the pipeline network nodes and optimizes the path with the total water delivery time and the total energy consumption of the pumping station as the objective function; The Pareto optimal solution set is solved by using a non-dominated sorting genetic algorithm, and the scheduling strategy including water supply allocation ratio, water transmission path and node scheduling quantity is output based on actual demand. Step 5: Closed-loop feedback optimization, correct the empirical model parameters based on the actual scheduling effect and update the dynamic empirical model library.

[0005] Furthermore, in step 1: The water volume prediction module is used to process historical rainfall and river flow data based on the LSTM network and output the water volume forecast value for the future period; The water demand analysis module is used to combine the crop growth cycle and urban water use patterns to output the spatial and temporal distribution matrix of regional water demand; The ecological response module is used to output the ecological sensitivity coefficient vector of wetlands or river sections based on historical ecological water replenishment effect data.

[0006] Furthermore, the feature vector in step 2 The construction methods include: Meteorological characteristics: Collect the precipitation forecast and temperature change rate for the next 72 hours; Hydrological characteristics: current reservoir storage rate and river flow change trends; Ecological characteristics: vegetation cover index and water eutrophication index; The above features are normalized and concatenated into .

[0007] Furthermore, the parameter transfer function of step three is defined as: ; in: is the empirical model parameter vector; is the emergency event feature vector; is the feature encoder; and is the trainable parameter matrix and T is the matrix transpose.

[0008] Furthermore, the feature encoder It is a three-layer fully connected neural network, and its network structure includes: Input layer: The number of neurons is equal to Dimensions; Hidden layer: 128-dimensional fully connected layer using ReLU activation function; Output layer: 64-dimensional feature vector.

[0009] Furthermore, the emergency feature vector The construction methods include: Calculate the drought index: ; Where Pt is the current rainfall, Pavg is the historical average rainfall, Pstd is the standard deviation of historical rainfall, Rt is the current river flow, Ravg is the historical average flow, Rstd is the standard deviation of historical flow, α and β are the preset weight coefficients in the drought index calculation formula, which are used to weigh the contribution of rainfall deviation and river flow deviation to the drought index respectively; Encode the device fault status into a binary vector f device ∈{0,1} 2 ; Combine the drought index with the equipment failure status code to .

[0010] Furthermore, the construction of the two-layer game decision tree in step 4 includes: Upper water supply allocation game: Define the municipal water supply benefit function Um: ; Agricultural water supply benefit function Ua: ; Ecological water replenishment benefit function Ue: ; Where ωk is the ecological sensitivity coefficient; is the municipal water priority coefficient, is the crop water requirement weight; i is the index of the municipal water unit, which is used to traverse all municipal water areas; j is the index of the agricultural water unit, which is used to traverse all agricultural irrigation areas; k is the index of the ecological water replenishment unit, which is used to traverse all areas that need ecological water replenishment.

[0011] Nm is the total number of municipal water use units, and the value of i ranges from 1 to Nm, indicating that there are Nm municipal water use areas in total; Na is the total number of agricultural water use units, and the value of j ranges from 1 to Na, indicating that there are a total of Na agricultural irrigation areas; Ne is the total number of ecological water replenishment units, and the value of k ranges from 1 to Ne, indicating that there are Ne ecological water replenishment areas in total; Based on the municipal water supply benefit function Um, the agricultural water supply benefit function Ua and the ecological water replenishment benefit function Ue, the Nash equilibrium is solved to obtain the optimal water supply allocation ratio [pm, pa, pe], and the solution is found to meet the constraints pm+pa+pe=1 and pi≥0, where: pm is the municipal water allocation ratio, pa is the agricultural water allocation ratio, and pe is the ecological water replenishment allocation ratio; Optimization of lower layer water delivery path: Establish flow balance constraints at network nodes: ; Where Ein(v) is the set of inlet edges of node v, Eout(v) is the set of outlet edges of node v, qe is the water delivery, and dv is the water demand of the node; Construct the objective function: ; Where Ttotal is the total water delivery time, Epump is the total energy consumption of the pumping station, and α and β are weight coefficients; The ecological sensitivity coefficient ωk in the ecological water replenishment benefit function is output by the ecological response module; The municipal water supply benefit function The actual water supply in the regional water demand spatiotemporal distribution matrix output by the water demand analysis module, To ensure the minimum water supply; In the agricultural water supply benefit function, Wj is the actual irrigation water volume, The optimal irrigation water volume; In the ecological water replenishment benefit function, Rk is the ecological water replenishment volume, and are the lower limit and critical value of ecological water demand respectively.

[0012] Furthermore, in step four, the upper-layer water supply allocation ratio [pm, pa, pe] and the lower-layer water transmission path plan qe are used as decision variables, and {Um, Ua, Ue, Ttotal, Epump} are used as multi-objectives. A non-dominated sorting genetic algorithm (NSGA-II) is used to solve the Pareto optimal solution set, and the optimal scheduling strategy including the water supply allocation ratio, water transmission path and node scheduling quantity is selected from the solution set according to actual scheduling needs.

[0013] Furthermore, it also includes collaborative optimization steps, specifically: When the emergency feature vector Drought Index hour: Adjust the ecological sensitivity coefficient. The adjustment process is as follows: ; Modify the weight of the lower layer objective function: , ; in and is the preset response coefficient.

[0014] Compared with the existing technology, the present invention has the following advantages: the water resource scheduling optimization method based on the empirical model combines the dynamic empirical model library with historical data to train the water volume prediction, water demand analysis and ecological response modules, thereby improving the accuracy of water volume prediction, water demand distribution and ecological sensitivity coefficient assessment, extracting meteorological, hydrological, ecological and other scene features in real time, and combining the dynamic migration optimization of the model to enable the scheduling model to quickly adapt to real-time scene changes. Through the two-layer game decision tree, the upper layer solves the water supply allocation ratio based on Nash equilibrium, taking into account the municipal, agricultural and ecological benefits; the lower layer optimizes the water transmission path, balances water transmission efficiency and energy consumption, and improves the rationality of resource allocation. The closed-loop feedback mechanism corrects the model parameters and updates the model library based on the actual scheduling effect, promoting the continuous iterative improvement of the scheduling method. The collaborative optimization mechanism for emergencies such as drought can dynamically adjust the ecological sensitivity coefficient and water transmission target weight, improving the scheduling adaptability under extreme scenarios. The design of the ecological response module and the ecological water replenishment benefit function ensures that ecological protection is taken into account during the scheduling process, promotes the sustainable use of water resources, and makes the system more worthy of promotion and use. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is the overall flow chart of the present invention. DETAILED DESCRIPTION

[0016] The following is a detailed description of an embodiment of the present invention. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process. However, the protection scope of the present invention is not limited to the following embodiment.

[0017] like Figure 1 As shown, this embodiment provides a technical solution: a water resources scheduling optimization method based on an empirical model, comprising the following steps: Step 1: Build a dynamic empirical model library and train the water volume prediction module, water demand analysis module, and ecological response module based on historical scheduling data; Step 2: Real-time scene feature extraction, extracting feature vectors from meteorological data, hydrological monitoring equipment, and ecological remote sensing data ; Step 3: Dynamically migrate and optimize the model by migrating the selected function Select the optimal empirical model combination from the dynamic empirical model library; Using parameter transfer function Adjust the parameters of the selected optimal empirical model combination; Step 4: Based on the water volume forecast value, water demand distribution matrix and ecological sensitivity coefficient, a two-layer game decision tree is constructed: The upper layer defines the municipal water supply benefit function, agricultural water supply benefit function and ecological water replenishment benefit function, and uses Nash equilibrium to solve the optimal water supply allocation ratio; The lower layer establishes the flow balance constraints of the pipeline network nodes and optimizes the path with the total water delivery time and the total energy consumption of the pumping station as the objective function; The Pareto optimal solution set is solved by using a non-dominated sorting genetic algorithm, and the scheduling strategy including water supply allocation ratio, water transmission path and node scheduling quantity is output based on actual demand. Step 5: Closed-loop feedback optimization, correct the empirical model parameters based on the actual scheduling effect and update the dynamic empirical model library.

[0018] In the step 1: The water volume prediction module is used to process historical rainfall and river flow data based on the LSTM network and output the water volume forecast value for the future period; The water demand analysis module is used to combine the crop growth cycle and urban water use patterns to output the spatial and temporal distribution matrix of regional water demand; The ecological response module is used to output the ecological sensitivity coefficient vector of wetlands or river sections based on historical ecological water replenishment effect data; The water volume forecasting module processes historical data based on the LSTM network to improve the accuracy of water volume forecasts for future periods and provide a reliable water volume basis for scheduling; The water demand analysis module combines crop growth cycles and urban water use patterns to make regional water demand analysis more in line with actual needs. The output spatiotemporal distribution matrix more accurately reflects the water demand situation in different regions and at different times. The ecological response module is based on historical ecological water replenishment effect data. The output ecological sensitivity coefficient vector can specifically reflect the sensitivity of wetlands or river sections to water replenishment, provide a scientific basis for ecological water replenishment decision-making, and ensure the accuracy of ecological protection.

[0019] The feature vector in step 2 The construction methods include: Meteorological characteristics: Collect the precipitation forecast and temperature change rate for the next 72 hours; Hydrological characteristics: current reservoir storage rate and river flow change trends; Ecological characteristics: vegetation cover index and water eutrophication index; The above features are normalized and concatenated into ; The specific composition of the real-time scene feature vector fs is clarified, with a total of 6 features and standardized methods in three categories: meteorology, hydrology, and ecology, making the feature extraction process operational and quantifiable, ensuring that the subsequent model dynamic migration optimization can select the optimal model combination based on comprehensive and accurate scene information, and improve the adaptability of the scheduling model to the real-time environment.

[0020] If the feature vector fs extracted from a certain area shows that the meteorological characteristics show that the predicted precipitation value in the next 72 hours is extremely low, the hydrological characteristics show that the current reservoir water storage rate is only 30%, and the ecological characteristics show that the vegetation cover index has decreased significantly, then the model can give priority to the empirical model combination that focuses on water-saving scheduling based on these clear characteristics to avoid model selection bias due to feature ambiguity.

[0021] The parameter migration function of step three is defined as: ; in: is the empirical model parameter vector; is the emergency event feature vector; is the feature encoder; and is the trainable parameter matrix, T is the matrix transpose; The mathematical definition of the parameter transfer function is clarified. Through the feature encoder and the trainable parameter matrix, a quantitative mapping relationship is established between the emergency feature vector fe and the empirical model parameter vector θ, so that the model parameters can be dynamically adjusted according to real-time emergencies (such as droughts, equipment failures, etc.), enhancing the adaptability of the empirical model to emergency scenarios and ensuring that the scheduling basis of the model output is more in line with the actual emergency situation.

[0022] If the emergency event eigenvector fe shows that the drought index DIt increases significantly (such as a persistent drought), the parameter transfer function The drought characteristics will be processed through the feature encoder, and then combined with the trainable parameters W and b, the parameters θ of the water volume prediction module will be automatically adjusted, so that the prediction model will focus more on the water volume attenuation trend under drought conditions, avoiding the prediction deviation caused by fixed model parameters, and providing more accurate basic data for subsequent water supply allocation.

[0023] The feature encoder It is a three-layer fully connected neural network, and its network structure includes: Input layer: The number of neurons is equal to Dimensions; Hidden layer: 128-dimensional fully connected layer using ReLU activation function; Output layer: 64-dimensional feature vector; Clarified feature encoder The specific network structure is a three-layer fully connected neural network with an input layer, a 128-dimensional ReLU hidden layer, and a 64-dimensional output layer. This makes the encoding process of the emergency feature vector fe quantifiable and reproducible. This deep network structure effectively extracts nonlinear features and key information from fe , avoiding feature loss and ensuring that the parameter transfer function accurately adjusts model parameters based on high-quality encoded features, improving the accuracy of the empirical model's adaptation to emergency scenarios.

[0024] If the emergency event feature vector fe contains the drought index DI t = 1.5 and the device fault vector f device = [1, 0], indicating a pump station failure. After the feature encoder's input layer receives these raw features, the hidden layer uses a ReLU activation function to process the correlation between drought and equipment failure. For example, a pump station failure may exacerbate the impact of drought. The output layer compresses these features into 64-dimensional deep features. The parameter transfer function adjusts model parameters based on these encoded features, allowing the water volume prediction module to focus more on flow attenuation caused by drought, while also avoiding water transfer routes from failed pump stations, improving scheduling accuracy.

[0025] The emergency event feature vector The construction methods include: Calculate the drought index: ; Where Pt is the current rainfall, Pavg is the historical average rainfall, Pstd is the standard deviation of historical rainfall, Rt is the current river flow, Ravg is the historical average flow, Rstd is the standard deviation of historical flow, α and β are the preset weight coefficients in the drought index calculation formula, which are used to weigh the contribution of rainfall deviation and river flow deviation to the drought index respectively; Encode the device fault status into a binary vector f device ∈{0,1} 2 ; Combine the drought index with the equipment failure status code to ; The quantitative construction rules of the emergency feature vector fe are clarified. By converting abstract emergencies such as drought and equipment failure into standardized numerical vectors, accurate input is provided for the dynamic migration optimization of the model, ensuring that the empirical model can quickly identify the core characteristics of the emergency scenario and adjust the parameters in a targeted manner, thereby improving the response accuracy of the dispatching system to emergencies.

[0026] The construction of the two-layer game decision tree in step 4 includes: Upper water supply allocation game: Define the municipal water supply benefit function Um: ; Agricultural water supply benefit function Ua: ; Ecological water replenishment benefit function Ue: ; Where ωk is the ecological sensitivity coefficient; is the municipal water priority coefficient, is the crop water requirement weight; i is the index of the municipal water unit, which is used to traverse all municipal water areas; j is the index of the agricultural water unit, which is used to traverse all agricultural irrigation areas; k is the index of the ecological water replenishment unit, which is used to traverse all areas that need ecological water replenishment.

[0027] Nm is the total number of municipal water use units, and the value of i ranges from 1 to Nm, indicating that there are Nm municipal water use areas in total; Na is the total number of agricultural water use units, and the value of j ranges from 1 to Na, indicating that there are a total of Na agricultural irrigation areas; Ne is the total number of ecological water replenishment units, and the value of k ranges from 1 to Ne, indicating that there are Ne ecological water replenishment areas in total; Based on the municipal water supply benefit function Um, the agricultural water supply benefit function Ua and the ecological water replenishment benefit function Ue, the Nash equilibrium is solved to obtain the optimal water supply allocation ratio [pm, pa, pe], and the solution is found to meet the constraints pm+pa+pe=1 and pi≥0, where: pm is the municipal water allocation ratio, pa is the agricultural water allocation ratio, and pe is the ecological water replenishment allocation ratio; pi refers to any one of the municipal water allocation ratio pm, agricultural water allocation ratio pa, and ecological water replenishment allocation ratio pe. The constraint condition pi ≥ 0 indicates that the allocation ratio of each type of water use cannot be negative, which conforms to the actual allocation logic; Optimization of lower layer water delivery path: Establish flow balance constraints at network nodes: ; Where Ein(v) is the set of inlet edges of node v, Eout(v) is the set of outlet edges of node v, qe is the water delivery, and dv is the water demand of the node; Construct the objective function: ; Where Ttotal is the total water delivery time, Epump is the total energy consumption of the pumping station, and α and β are weight coefficients; The ecological sensitivity coefficient ωk in the ecological water replenishment benefit function is output by the ecological response module; The municipal water supply benefit function The actual water supply in the regional water demand spatiotemporal distribution matrix output by the water demand analysis module, To ensure the minimum water supply; In the agricultural water supply benefit function, Wj is the actual irrigation water volume, The optimal irrigation water volume; In the ecological water replenishment benefit function, Rk is the ecological water replenishment volume, and are the lower limit and critical value of ecological water demand respectively; The specific construction logic of the two-layer game decision tree was clarified. Through the coordinated design of the upper-layer water supply allocation game and the lower-layer path optimization, the connection between multi-objective benefit balance and efficient water delivery execution was achieved: The upper layer defines three types of benefit functions: municipal, agricultural, and ecological. It clarifies the relationship between the parameters of each function and the previous modules to ensure that the scheduling basis is consistent with the basic data. Solve the water supply allocation ratio through Nash equilibrium, taking into account the core needs of different water users and avoiding the imbalance of resource allocation caused by prioritizing a single goal; The lower layer defines the pipe network flow balance constraints and water delivery objective function. This is the total time plus energy consumption, ensuring that the water delivery plan takes into account both efficiency and cost while meeting the water demand of the nodes. The two-tier structure forms a closed loop from allocation strategy to execution plan, making scheduling both consistent with the interests of multiple entities and practical; Assume that the total water supply in a certain area is 1 million m 3 , scheduling needs to be performed through a two-layer game decision tree: Upper water supply allocation game: Municipal water supply benefit function Um: The water demand analysis module outputs the regional water demand spatiotemporal distribution matrix, where two municipal areas (N m =2) Actual water supply 250,000 m 3 150,000 m 3 , minimum guaranteed water volume 200,000 m 3 100,000 m 3 , municipal water priority coefficient They are 0.6 and 0.4 respectively.

[0028] Calculation: Um=0.6×(25-20)+0.4×(15-10)=0.6×5+0.4×5=3+2=5.

[0029] Agricultural water supply benefit function Ua: The water demand analysis module outputs 1 agricultural area (N a =1) the actual irrigation water volume Wj = 300,000 m 3 , optimal irrigation water volume =400,000 m 3 , crop water requirement weight =0.8.

[0030] Calculation: Ua=0.8×tanh(30 / 40)=0.8×tanh(0.75)≈0.8×0.604≈0.483.

[0031] Ecological water replenishment benefit function Ue: The ecological response module outputs 1 wetland (Ne =1) with an ecological sensitivity coefficient of ωk = 0.5 and an actual water replenishment volume of Rk = 300,000 m 3 , ecological water demand lower limit =200,000 m 3 , critical value =500,000 m 3 .

[0032] calculate: Ue=0.5×tanh[(30-20) / (50-20)]=0.5×tanh(10 / 30)=0.5×tanh(1 / 3)≈0.5×0.321≈0.161.

[0033] By solving the Nash equilibrium, satisfying pm+pa+pe=1 and pi≥0, the optimal allocation ratio is [pm=0.4, pa=0.3, pe=0.3] (i.e., the municipal 400,000 m 3 Agriculture: 300,000 m 3 、Ecological 300,000 m 3 ), which matches the calculation result of the above benefit function to achieve multi-objective balance.

[0034] Lower layer path optimization: Flow balance constraints: The network consists of three nodes (v1, v2, and v3). V1 is the water source node, V2 is the municipal / agricultural water node, and V3 is the ecological water replenishment node. The following must be true: Total inlet flow - Total outlet flow = Node water demand (dv).

[0035] If the water demand of v2 is dv2=40+30=700,000 m 3 , v3 water demand dv3 = 300,000 m 3 , then the edge flow q into v2 12 =700,000 m 3 , the edge flow q entering v3 13 =300,000 m 3 , satisfying q 12 -0=70,q 13 -0=30 (no water edge), which meets the constraints.

[0036] Objective function optimization: Taking total water delivery time (Ttotal) and pumping station energy consumption (Epump) as targets, set α = 0.3 and β = 0.7, and the objective function is min(0.3 × Ttotal + 0.7 × Epump). If the total time from the water source to v2 and then to v3 is shorter and consumes less energy than the total time from the water source to v2 and then to v3, the former is selected as the optimal path, ensuring efficient and low-energy water delivery.

[0037] Through this two-tiered optimization, the final scheduling solution not only balances the water demands of municipal, agricultural, and ecological needs, but also ensures the efficiency of the water delivery process, avoiding resource waste or supply-demand imbalance caused by deviation from a single objective. The specific process of Nash finding equilibrium is as follows: Game model construction; Participants: Municipal (m), agriculture (a), and ecology (e) are three major water users.

[0038] Decision variables: water supply allocation ratio [pm, pa, pe], satisfying the constraints: pm+pa+pe=1, pi≥0(i=m,a,e); Obtaining the benefit function; Municipal water supply benefit function Um: ; Agricultural water supply benefit function Ua: ; Ecological water replenishment benefit function Ue: ; in, =pm·Qtotal (total amount of water resources available for dispatch), similarly Wj=pa·Qtotal, Rk=pe·Qtotal; Nash equilibrium requires: For any party i∈{m,a,e}, when the allocation ratio of other parties is fixed, pi maximizes Ui, that is: Ui(pi ∗ ,p ∗ −i )≥Ui(pi,p ∗ −i ),∀pi≥0,pi ∗ +p ∗ −i =1, where p ∗ −i represents the equilibrium proportion of all parties except i.

[0039] The solution steps are: Step 1: Reduce the dimensionality of the three-dimensional problem; By constraining pm+pa+pe=1, let pe=1−pm−pa, and the problem is transformed into a two-dimensional optimization: max pm,pa≥0,pm+pa≤1 {Um(pm),Ua(pa),Ue(1−pm−pa)}; Step 2: Derive the unilateral optimal response function; Municipal optimal response (fixed pa): Um increases linearly with pm, so the municipal optimal response is: ; Agricultural optimal response (fixed PM): Derivative of Ua: ; Since cosh2(∙)>0, Ua increases monotonically with pa, and the optimal agricultural response is: ; Ecological optimal response (fixed PM, PA): Derivative of Ue: ; Similarly, Ue increases monotonically with pe, and the ecological optimal response is: pe∗=1−pm−pa (only when peQtotal≥ Valid when ); Step 3: Find the equilibrium point; Nash equilibrium must satisfy the intersection of the three parties' optimal responses, namely: pm ∗ =1−pa ∗ −pe ∗ ; pa∗=1−pm ∗ −pe ∗ ; pe ∗ =1−pm ∗ −pa ∗ ; Combined with the non-negativity constraint, possible equilibrium scenarios include: Perfect competition scenario (none of the three parties is saturated): ; Where λ, γ, ω are the marginal coefficients of the benefit function (normalized by derivative); Unilateral saturation scenario (e.g. municipal water demand is saturated): ; Agriculture and ecology are allocated according to marginal benefits in the remaining water volume.

[0040] Multi-party saturation scenario (e.g., both municipal administration and agriculture are saturated): The solution is achieved through analytical derivation and numerical verification of Nash equilibrium, specifically: Analytical derivation, exploiting the monotonicity of the benefit function, we derive explicit expressions for the optimal responses of each party.

[0041] Constrained projection, projecting the solution to the feasible region where pi ≥ 0 and pm + pa + pe = 1.

[0042] Nash equilibrium verification, check Um(pm ∗ ,pa ∗ )≥Um(pm,pa ∗ ) holds for any pm; Similarly verify Ua and Ue; If there are multiple equilibria, the Pareto optimal solution is selected through non-dominated sorting, i.e., NSGA-II algorithm.

[0043] For example, balanced distribution during drought period: Assume that the total water volume Qtotal = 1 million m 3 , the parameters are as follows: Municipal government: minimum guarantee =400,000 m 3 , priority λ=0.6; Agriculture: Optimal Irrigation =300,000 m 3 , weight γ=0.3; Ecology: Critical Value =200,000 m 3 , sensitivity coefficient ω=0.1.

[0044] Solution process: Municipal water demand is not saturated (40<100), the optimal response pm ∗ ≥0.4; Agricultural irrigation is not saturated (30<100), the optimal response pa ∗ ≥0.3; Ecological water demand lower limit Rkmin = 100,000 m 3 , the remaining water volume is 100-400-300=300,000 m 3 >100,000m 3 ,efficient.

[0045] Balanced ratio: pm ∗ =0.4,pa ∗ =0.3,pe ∗ =0.3, satisfying Um=5,Ua=0.48,Ue=0.16.

[0046] Verification: Unilaterally adjusting pm=0.5 will cause Ua and Ue to decrease, and vice versa, so it is a Nash equilibrium.

[0047] In step 4, the upper layer water supply allocation ratio [pm, pa, pe] and the lower layer water transmission path plan are used as decision variables, and {Um, Ua, Ue, Ttotal, Epump} are used as multi-objectives. The non-dominated sorting genetic algorithm (NSGA-II) is used to solve the Pareto optimal solution set, and the optimal scheduling strategy including the water supply allocation ratio, water transmission path and node scheduling quantity is selected from the solution set according to the actual scheduling demand; The specific implementation method for multi-objective optimization in step 4 was clarified: using the core output of the two-layer game decision tree (upper-layer water supply allocation ratio, lower-layer water transmission path) as the decision variable, incorporating municipal / agricultural / ecological benefits, water transmission time, and pumping station energy consumption into the multi-objective system, and solving the Pareto optimal solution set using the non-dominated sorting genetic algorithm (NSGA-II), ultimately selecting the optimal strategy based on actual needs. The benefits of this design are: Taking into account conflicting objectives, such as improving ecological benefits may increase energy consumption, and shortening water delivery time may reduce distribution fairness, to avoid single-objective optimization that leads to loss of balance; Provides a variety of optimal solution options to adapt to different actual scenario requirements, such as prioritizing efficiency during droughts and reducing energy consumption during equipment failures, thereby enhancing scheduling flexibility. Clarify the relationship between the algorithm and decision variables and goals, make the solution process reproducible and the results quantifiable, and ensure the scientific nature of the scheduling strategy.

[0048] For example, the total water supply in a certain area is 1 million m 3 , based on the output of the two-layer game decision tree, through the optimization of the scheduling strategy: Decision variables and multiple objectives: Decision variables: upper-layer water supply allocation ratio [pm, pa, pe] (such as [0.4, 0.3, 0.3], [0.35, 0.35, 0.3], etc.), lower-layer water transmission path plan (such as water source to node 2 and then to node 3 water source, directly supplying nodes 2 and 3 respectively, etc.).

[0049] Multi-objective: {Um, Ua, Ue, Ttotal, Epump}, i.e. {municipal benefit, agricultural benefit, ecological benefit, total water delivery time, total energy consumption of the pumping station}; NSGA-II solves the Pareto optimal solution set: After the algorithm is iterated, multiple non-dominated solutions are obtained, for example: Solution 1: [pm=0.45, pa=0.25, pe=0.3], the path is a direct supply dual-node path, corresponding to the target values ​​{5.8, 0.42, 0.16, 2.5 hours, 800 kWh}, focusing on improving municipal benefits, with medium duration and energy consumption; Solution 2: [pm=0.3, pa=0.4, pe=0.3], the path is series water supply, corresponding to the target values ​​{4.2, 0.55, 0.16, 3.2 hours, 650kWh}, focusing on improving agricultural efficiency, lower energy consumption but longer duration; Solution 3: [pm=0.35, pa=0.3, pe=0.35], the path is mixed water supply, corresponding to the target values ​​{5, 0.48, 0.2, 2.8 hours, 720 kWh}, balancing the three types of benefits, with moderate duration and energy consumption; If the current scenario is a drought period and a critical period for agricultural irrigation, the actual demand is to prioritize crop irrigation while taking into account energy consumption control. Then, solution 2 is selected from the solution set, which has the highest agricultural benefits and the lowest energy consumption. The corresponding water supply allocation ratio [0.3, 0.4, 0.3], the water supply and node scheduling amount of the water transmission path are output in series, and node 2 is allocated 700,000 m 3 Node 3 allocates 300,000 m 3 , to achieve precise matching with actual needs.

[0050] It also includes collaborative optimization steps, specifically: When the emergency feature vector Drought Index hour: Adjust the ecological sensitivity coefficient. The adjustment process is as follows: ; Modify the weight of the lower layer objective function: , ; in and is the preset response coefficient; The collaborative optimization mechanism for emergencies, such as drought, is clarified. By dynamically adjusting the ecological sensitivity coefficient and the weight of the water delivery objective function, the scheduling strategy can quickly respond to extreme scenarios and avoid resource allocation imbalances caused by fixed parameters: Enhanced targeting of sudden droughts: When the drought index exceeds the threshold, the ecological sensitivity coefficient will be adjusted to balance the priorities of ecological protection and other water needs in special circumstances; Optimize the adaptability of water delivery targets: Modify the weights of the underlying objective functions to prioritize reducing energy consumption or ensuring water delivery stability in emergency scenarios such as droughts, thereby avoiding resource waste; A region experiences sudden drought, and the drought index DI t =3.0 (exceeding the preset threshold τ=1.5), triggering collaborative optimization: Adjust the ecological sensitivity coefficient, the original ecological sensitivity coefficient ωk=0.5, the preset response coefficient σ=0.2, then after adjustment ωk′=0.5×(1+0.2×3.0)=0.5×1.6=0.8. After adjustment, ωk′ in the ecological water replenishment benefit function Ue increases, which means that in the water supply allocation game, the weight of ecological water replenishment is relatively increased, but combined with the actual needs of drought scenarios, priority is given to ensuring basic water use in cities and agriculture. This adjustment can avoid excessive compression of ecological water use without excessively squeezing out core needs.

[0051] Modify the weight of the lower layer objective function: The original lower-level objective function weights are α=0.3 (total water delivery time weight), β=0.7 (pump station energy consumption weight), and the preset response coefficient ρ=2. After adjustment, α′=0.3 / 2=0.15 and β′=0.7×2=1.4.

[0052] After the adjustment, the lower-level objective function becomes min(0.15×Ttotal+1.4×Epump), which focuses more on reducing the energy consumption of the pumping station. Due to the shortage of water resources during the drought period, unnecessary energy consumption needs to be reduced. For example, when selecting the water transmission route, pipelines with lower energy consumption are given priority. Even if the water transmission time is slightly longer, energy costs can be saved during the drought period.

[0053] After coordinated optimization, the upper-level water supply allocation is still mainly based on ensuring basic water use in cities (pm=0.4) and agriculture (pa=0.3), and ecological water replenishment (pe=0.3) is avoided from excessive reduction through the adjustment of ωk′; the lower-level water transmission route selection is more inclined to low-energy consumption solutions to ensure the efficient use of water resources and energy during drought periods, and avoid the double waste of both water shortage and energy consumption due to fixed parameters.

[0054] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0055] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0056] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A water resources scheduling optimization method based on an empirical model, characterized in that: The following steps are involved: Step 1: Build a dynamic empirical model library and train the water volume prediction module, water demand analysis module, and ecological response module based on historical scheduling data; Step 2: Real-time scene feature extraction, extracting feature vectors from meteorological data, hydrological monitoring equipment, and ecological remote sensing data ; Step 3: Dynamically migrate and optimize the model by migrating the selected function Select the optimal empirical model combination from the dynamic empirical model library; Using parameter transfer function Adjust the parameters of the selected optimal empirical model combination; Step 4: Construct a two-layer game decision tree based on the water volume forecast value, water demand distribution matrix and ecological sensitivity coefficient; Step 5: Closed-loop feedback optimization, correct the empirical model parameters based on the actual scheduling effect and update the dynamic empirical model library.

2. The water resource scheduling optimization method based on an empirical model according to claim 1, characterized in that: In the step 1: The water volume prediction module is used to process historical rainfall and river flow data based on the LSTM network and output the water volume forecast value for the future period; The water demand analysis module is used to combine the crop growth cycle and urban water use patterns to output the spatial and temporal distribution matrix of regional water demand; The ecological response module is used to output the ecological sensitivity coefficient vector of wetlands or river sections based on historical ecological water replenishment effect data.

3. The water resource scheduling optimization method based on an empirical model according to claim 1, characterized in that: The feature vector in step 2 The construction methods include: Meteorological characteristics: Collect the predicted precipitation value and temperature change rate for the preset time period in the future; Hydrological characteristics: current reservoir storage rate and river flow change trends; Ecological characteristics: vegetation cover index and water eutrophication index; The above features are normalized and concatenated into .

4. The water resource scheduling optimization method based on an empirical model according to claim 1, characterized in that: The parameter migration function of step three is defined as: ； in: is the empirical model parameter vector; is the emergency event feature vector; is the feature encoder; and is the trainable parameter matrix and T is the matrix transpose.

5. The water resource scheduling optimization method based on the empirical model according to claim 4 is characterized in that: The feature encoder It is a three-layer fully connected neural network, and its network structure includes: Input layer: The number of neurons is equal to Dimensions; Hidden layer: 128-dimensional fully connected layer using ReLU activation function; Output layer: 64-dimensional feature vector.

6. The water resource scheduling optimization method based on an empirical model according to claim 4, characterized in that: The emergency event feature vector The construction methods include: Calculate the drought index: ; Where Pt is the current rainfall, Pavg is the historical average rainfall, Pstd is the standard deviation of historical rainfall, Rt is the current river flow, Ravg is the historical average flow, Rstd is the standard deviation of historical flow, α and β are the preset weight coefficients in the drought index calculation formula, which are used to weigh the contribution of rainfall deviation and river flow deviation to the drought index respectively; Encode the device fault status into a binary vector f device ∈{0,1} 2 ; Combine the drought index with the equipment failure status code to .

7. The water resource scheduling optimization method based on an empirical model according to claim 1, characterized in that: The construction of the two-layer game decision tree in step 4 includes: Upper water supply allocation game: Define the municipal water supply benefit function Um: ； Agricultural water supply benefit function Ua: ； Ecological water replenishment benefit function Ue: ； Where ωk is the ecological sensitivity coefficient; is the municipal water priority coefficient, is the crop water requirement weight; i is the index of the municipal water unit; j is the index of the agricultural water unit; k is the index of the ecological water replenishment unit; Nm is the total number of municipal water units; Na is the total number of agricultural water units; Ne is the total number of ecological water replenishment units; Based on the municipal water supply benefit function Um, the agricultural water supply benefit function Ua and the ecological water replenishment benefit function Ue, the Nash equilibrium is solved to obtain the optimal water supply allocation ratio [pm, pa, pe], and the solution is found to meet the constraints pm+pa+pe=1 and pi≥0, where: pm is the municipal water allocation ratio, pa is the agricultural water allocation ratio, and pe is the ecological water replenishment allocation ratio; Optimization of lower layer water delivery path: Establish flow balance constraints at network nodes: ； Where Ein(v) is the set of inlet edges of node v, Eout(v) is the set of outlet edges of node v, qe is the water delivery, and dv is the water demand of the node; Construct the objective function: ; Where Ttotal is the total water delivery time, Epump is the total energy consumption of the pumping station, and α and β are weight coefficients; The ecological sensitivity coefficient ωk in the ecological water replenishment benefit function is output by the ecological response module; The municipal water supply benefit function The actual water supply in the regional water demand spatiotemporal distribution matrix output by the water demand analysis module, To ensure the minimum water supply; In the agricultural water supply benefit function, Wj is the actual irrigation water volume, The optimal irrigation water volume; In the ecological water replenishment benefit function, Rk is the ecological water replenishment volume, and are the lower limit and critical value of ecological water demand respectively.

8. The water resource scheduling optimization method based on an empirical model according to claim 7, characterized in that: In step 4, the upper layer water supply allocation ratio [pm, pa, pe] and the lower layer water transmission path plan are used as decision variables, and {Um, Ua, Ue, Ttotal, Epump} are used as multi-objectives. A non-dominated sorting genetic algorithm is used to solve the Pareto optimal solution set, and the optimal scheduling strategy including the water supply allocation ratio, water transmission path and node scheduling quantity is selected from the solution set according to the actual scheduling needs, where the lower layer water transmission path plan is composed of the water transmission quantity qe of each edge.

9. The water resource scheduling optimization method based on an empirical model according to claim 7, characterized in that: It also includes collaborative optimization steps, specifically: When the emergency feature vector Drought Index hour: Adjust the ecological sensitivity coefficient. The adjustment process is as follows: ; Modify the weight of the lower layer objective function: , ; in and is the preset response coefficient.

Images