A thermal power multi-water source mobilization optimization method and system based on NSGA-III

Abstract

The application discloses a thermal power multi-water source mobilization optimization method and system based on NSGA-III, and the method comprises the following steps: obtaining water source cost and combining GIS spatial analysis to calculate water transmission cost; a multi-objective optimization model is established with the minimum total water distribution cost, the minimum external water diversion dependence and the maximum regenerated water utilization as targets; initial allocation is carried out based on cost weight, and the initial population is generated through upper limit correction, balance repair and disturbance; an improved NSGA-III algorithm is used for iterative evolution, the optimal solution set is output through hierarchical management of constraint violation degree, differential variation and reference point encryption screening; and the final water distribution scheme is obtained through post-processing. The application improves the initial population feasibility, convergence speed and solution set quality, and realizes fine configuration of multi-water sources in the thermal power industry.

Description

Technical Field

[0001] This application belongs to the field of water resources management technology, specifically relating to a method and system for optimizing the dispatch of multiple water sources in thermal power plants. Background Technology

[0002] Optimal allocation of water resources is a core means to achieve sustainable regional water resource utilization. It aims to meet the water needs of various users by scientifically scheduling multiple water sources, while simultaneously considering both economic and ecological benefits. With the continuous advancement of inter-basin water transfer projects in my country, water-receiving areas face a complex situation of coordinated scheduling of multiple water sources, including local water, externally transferred water, and unconventional water. Water security for high water-consuming industrial users (such as thermal power plants) within the region is directly related to regional energy security and economic development. Therefore, establishing efficient multi-source optimal allocation models and designing reliable solution algorithms has become an important research direction in the field of water resource management.

[0003] Existing technologies typically employ multi-objective models to solve water resource management problems. However, handling constraints remains a core and challenging issue in the optimization process of multi-objective models. When the types of constraints and the number of decision variables are excessive, and there are strong correlations between these variables, the feasible space becomes a small proportion of the overall search space. Algorithms then need to increase the population and iteration count to expand the search, resulting in longer computation times, poor convergence, and even failure to find a feasible solution.

[0004] Existing technologies typically employ the standard NSGA-III algorithm to solve multi-objective optimization models. While effective in regional water resource allocation, the NSGA-III algorithm can optimize objectives such as water shortage, cost, and pollution, and considers unconventional water sources. However, in the context of coordinated scheduling of multiple water sources (local water, externally diverted water, and unconventional water), it is difficult to guarantee finding a high-quality and evenly distributed optimization solution within a limited computation time. Furthermore, the standard NSGA-III algorithm uses random generation when generating the population, which easily leads to a large number of infeasible solutions, resulting in low efficiency in feasible region search and algorithm convergence. In addition, existing technologies often simplify spatial factors such as distance and pipeline networks into fixed cost coefficients, multiplying the water supply by a fixed coefficient to obtain the water transmission cost, resulting in high water transmission costs. In the multi-source allocation process for thermal power plants, there are differences in the unit water supply cost, maximum available water volume, and water demand of different thermal power plants.

[0005] In summary, while existing multi-objective optimization methods can optimize water distribution schemes, there is still room for further optimization in the synergistic utilization of the aforementioned cost information and water supply constraints during the initial scheme generation stage. Furthermore, how to balance the water balance constraints of thermal power plants and the maximum available water volume constraints of each water source while simultaneously satisfying these constraints is also an issue that needs further consideration in the optimal allocation of multiple water sources for thermal power plants. Summary of the Invention

[0006] This application provides a method and system for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ, aiming to solve the problem that existing technologies cannot guarantee finding a high-quality and uniformly distributed optimization scheme within a limited computation time.

[0007] Firstly, a method for optimizing the multi-water source regulation of thermal power plants based on NSGA-Ⅲ, the method comprising:

[0008] Obtain the water supply cost of each water source unit, and calculate the shortest reachable distance between each water source and each water user based on GIS spatial analysis, so as to calculate the water distribution cost from each water source to each water user.

[0009] The system acquires the real-time maximum available water volume of each water source, the real-time water demand information of each thermal power plant, and the water distribution cost from each water source to each water user. It then determines whether the changes in the real-time maximum available water volume, real-time water demand information, or water distribution cost of each water source relative to the previous period exceed a preset threshold. If they do, the system proceeds to the next step.

[0010] A multi-objective optimization model is established with N thermal power plants in the water receiving area as water users and M types of water sources as water supply sources. The multi-objective optimization model takes the minimum total water distribution cost, the minimum total dependence on external water transfer, and the maximum total utilization of reclaimed water as optimization objectives, and the water balance constraints of each thermal power plant and the upper limit of the maximum available water volume of each water source as constraints.

[0011] Based on the unit water supply cost of each water source, an allocation weight is calculated. An initial water allocation is then performed according to the water demand of each thermal power plant and the allocation weight, resulting in an initial allocation scheme. It is then determined whether the total allocation amount of each water source in the initial allocation scheme exceeds its maximum available water capacity. If it does, a scaling factor is calculated, and the allocation amount of that water source to each thermal power plant is proportionally scaled and corrected, resulting in a first corrected scheme. Next, it is determined whether the total water supply of each thermal power plant in the first corrected scheme is equal to its water demand. If they are not equal, a water difference is calculated, and this difference is allocated to water sources with remaining capacity according to the allocation weight, resulting in a second corrected scheme. A controllable random perturbation is applied to the second corrected scheme to generate an initial population.

[0012] The initial population is input into the improved NSGA-Ⅲ algorithm for iterative evolution to solve the multi-objective optimization model. Each iteration performs the following operations: crossover and mutation of the parent population to generate the offspring population; the mutation is applied differently based on the individual constraint violation degree, with individuals having a constraint violation degree greater than zero exhibiting a greater mutation magnitude than those with a constraint violation degree equal to zero; the parent and offspring populations are merged into a mixed population, and the objective vector and constraint violation degree of each individual in the mixed population are calculated; the mixed population is stratified based on the constraint violation degree, forming a priority layer with a constraint violation degree equal to zero and a secondary layer with constraint violation degrees greater than zero, ordered numerically; the normalized objective vectors of the stratified population individuals are projected onto a reference direction, which densifies the dimension of the reclaimed water utilization objective; and individuals of equal size to the population are selected from the stratified population based on the individual distribution density under each reference direction to form a new generation population.

[0013] When the preset number of iterations is reached or the rate of change of the target value of the population is lower than the preset threshold for multiple consecutive generations, the iteration is terminated and the optimal solution set is output.

[0014] The optimal solution set is then filtered for feasible solutions and optimized to obtain the final water distribution scheme.

[0015] Optionally, in the above scheme, the minimum total cost of water distribution is:

[0016]

[0017]

[0018] in, The water supply from water source j to power plant i. The unit cost of water supply from the water source. This is the unit distance water distribution cost coefficient. The reachable distance of the water system between water source j and thermal power plant i, calculated for GIS purposes;

[0019] The minimum total amount of externally diverted water used is:

[0020]

[0021] in, The amount of water supplied to power plant i by external water transfer;

[0022] The maximum total amount of reclaimed water utilized is:

[0023]

[0024] in, The amount of reclaimed water supplied to power plant i.

[0025] Optionally, in the above scheme, the water balance constraint is:

[0026]

[0027] in, The water demand of thermal power plant i;

[0028] The maximum available water supply capacity of the water source is subject to the following constraints:

[0029]

[0030] in, The maximum available water volume of water source j;

[0031] The non-negativity constraint on the available water supply from the water source:

[0032]

[0033] in, The amount of water supplied from water source j to power plant i.

[0034] Optionally, in the above scheme, the weighting calculation based on the water supply cost of each water source unit includes:

[0035] Cost weights are calculated based on the reciprocal of the unit water supply cost for each water source:

[0036] ;

[0037] in, The unit cost of water supply from the water source;

[0038] The allocation weights are obtained by normalizing the cost weights of each water source:

[0039] .

[0040] Optionally, in the above scheme, calculating the scaling factor and proportionally scaling the distribution of the water source to each thermal power plant includes:

[0041] The scaling factor is calculated based on the ratio of the maximum available water volume to the total allocation of each water source in the initial allocation scheme:

[0042]

[0043] in, The maximum available water volume of water source j;

[0044] The allocation of water sources exceeding the maximum available water volume in the initial allocation scheme to each thermal power plant is scaled and corrected according to a scaling factor.

[0045] Optionally, in the above scheme, the water volume difference is calculated, and the difference is allocated among water sources with remaining capacity according to the allocation weight, resulting in the second modified scheme, which is as follows:

[0046] Calculate the water consumption difference for thermal power plants:

[0047] ;in,

[0048] The difference In water sources with remaining capacity, water is allocated according to the allocation weight; if it cannot be fully replenished, the constraint violation degree of the population is recorded as greater than zero.

[0049] Optionally, the above scheme further includes applying a controllable random perturbation to the second modified scheme:

[0050] Traverse each water source and calculate the amount by which the total allocation of any water source to all thermal power plants in the first generation population exceeds its maximum available water capacity. If the excess is greater than zero, then compress the allocation of that water source to all thermal power plants.

[0051] Determine whether the total water supply of each thermal power plant after compression is equal to its water demand. If they are not equal, calculate the water difference and allocate the difference to water sources with remaining capacity according to the allocation weight. After repair, recalculate the constraint violation degree of the individual. If the constraint violation degree is still greater than zero, it is retained as an infeasible solution.

[0052] In the above scheme, optionally, the constraint violation degree is calculated using the following formula:

[0053] .

[0054] Optionally, in the above scheme, the feasible solution screening and optimization correction of the optimal solution set specifically includes:

[0055] Individuals with a constraint violation rate of zero are retained as final candidates; if the constraint violation rate of all individuals is greater than zero, the individuals with the lowest constraint violation rate are selected for constraint repair, and their constraint violation rate is reduced before they are selected as backup candidates.

[0056] The final or backup candidates are simplified, the optimal compromise is identified, and they are sorted or clustered according to the decision-maker's preference weight to obtain representative solutions.

[0057] The representative schemes are fine-tuned locally using linear programming or quadratic programming to optimize the main objective or improve the utilization rate of reclaimed water while satisfying the constraints.

[0058] For the locally optimal fine-tuned scheme, multiple scenarios are tested under the fluctuation of the maximum water supply from the water source and the water demand of the thermal power plant. The fluctuation range of water distribution cost, total external water transfer and reclaimed water utilization is statistically analyzed. Schemes with fluctuation ranges exceeding the preset threshold are eliminated to determine the final water distribution scheme.

[0059] Secondly, a thermal power multi-water source dispatching and optimization system based on NSGA-Ⅲ, the system comprising:

[0060] Water distribution cost calculation module: used to obtain the water supply cost of each water source unit, and calculate the shortest reachable distance between each water source and each water user based on GIS spatial analysis, so as to calculate the water distribution cost from each water source to each water user.

[0061] Water distribution scheme update trigger module: used to obtain the real-time maximum available water volume of each water source, the real-time water demand information of each thermal power plant, and the water distribution cost from each water source to each water user in real time, and to determine whether the changes of the real-time maximum available water volume, real-time water demand information, or water distribution cost of each water source relative to the previous period exceed the preset threshold. If they exceed the threshold, the multi-objective optimization model construction module is executed.

[0062] Multi-objective optimization model construction module: used to establish a multi-objective optimization model with N thermal power plants in the water receiving area as water users and M types of water sources as water supply sources; the multi-objective optimization model takes the minimum total water distribution cost, the minimum total dependence on external water transfer, and the maximum total utilization of reclaimed water as optimization objectives, and takes the water balance constraints of each thermal power plant and the upper limit of the maximum available water volume of each water source as constraints.

[0063] The initial population generation module is used to calculate the allocation weight based on the unit water supply cost of each water source, and to perform initial water allocation according to the water demand of each thermal power plant and the allocation weight to obtain an initial allocation scheme; it determines whether the total allocation of each water source in the initial allocation scheme exceeds its maximum available water capacity. If it does, it calculates a scaling factor and proportionally scales the allocation of that water source to each thermal power plant to obtain a first correction scheme; it determines whether the total water supply of each thermal power plant in the first correction scheme is equal to its water demand. If they are not equal, it calculates the water difference and allocates the difference to water sources with remaining capacity according to the allocation weight to obtain a second correction scheme; it applies a controllable random perturbation to the second correction scheme to generate the initial population.

[0064] The solution module is used to input the initial population into the improved NSGA-Ⅲ algorithm for iterative evolution and solve the multi-objective optimization model. Each iteration performs the following operations: crossover and mutation of the parent population to generate the offspring population; the mutation is applied differently based on the individual constraint violation degree, with individuals having a constraint violation degree greater than zero exhibiting a greater mutation magnitude than those with a constraint violation degree equal to zero; merging the parent and offspring populations into a mixed population, calculating the objective vector and constraint violation degree of each individual in the mixed population; stratifying the mixed population based on the constraint violation degree, forming a priority layer with a constraint violation degree equal to zero and a secondary layer with constraint violation degrees greater than zero, sorted by numerical value; projecting the normalized objective vectors of the stratified population individuals onto a reference direction, where the reference direction densifies the dimension of the reclaimed water utilization objective; and selecting individuals of equal size from the stratified population based on the individual distribution density under each reference direction to form a new generation population.

[0065] Optimal solution set output module: used to terminate the iteration and output the optimal solution set when the preset number of iterations or the rate of change of the target value of the population for multiple consecutive generations is lower than a preset threshold.

[0066] The correction module is used to screen and optimize the optimal solution set to obtain the final water allocation scheme. Compared with the prior art, this application has at least the following beneficial effects:

[0067] This application employs a cost-oriented heuristic population initialization strategy. It calculates allocation weights based on the water supply cost of each water source unit and performs initial allocation according to water demand. Allocations exceeding the water source's upper limit are proportionally scaled and corrected. Allocations that do not meet water balance requirements are compensated for and subject to controllable perturbations. This effectively improves the feasibility of the initial population, ensuring that the vast majority of initial solutions satisfy or approximately satisfy the constraints. It significantly reduces the blind search in infeasible regions during the early stages of the algorithm, lowers computation time, and improves convergence performance. By embedding constraint violation degrees into the hierarchical management and differential mutation mechanism of the NSGA-III algorithm, it achieves high... Targeted repair of individuals with high constraint violation rates and preservation of excellent evolutionary information of individuals with low constraint violation rates enhance the algorithm's optimization ability in complex constraint scenarios. By calculating dynamic water transfer costs through GIS spatial analysis and incorporating them into the objective function, the simplification of traditional fixed cost coefficients is overcome, improving the accuracy of spatial information utilization. By densifying the reference points of the target dimension of reclaimed water utilization, the population is guided to generate optimization schemes with high unconventional water utilization rates, improving the quality of Pareto front distribution. After feasible solution screening, representative scheme extraction, and local optimization correction in the post-processing stage, a water distribution scheme that meets engineering applicability is finally obtained. Attached Figure Description

[0068] Figure 1A flowchart of a method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ is provided as an embodiment of this application;

[0069] Figure 2 Map showing the distribution of thermal power plants in Huai'an City;

[0070] Figure 3 This is a map showing the elevation distribution of Huai'an City. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0072] This invention aims to address the engineering challenge of refined and real-time water resource allocation for high water-consuming industries, particularly in the context of coordinated scheduling of multiple water sources, including local water, diverted water, and unconventional water, within the water-receiving area of ​​an inter-basin water transfer project (taking Huai'an City as an example). The core objective is to establish a multi-source, multi-user water resource optimization model to maximize the utilization of unconventional water, minimize reliance on high-cost diverted water, and improve the fairness and sustainability of overall regional water resource allocation, all while ensuring the safe operation of thermal power plants.

[0073] This application improves algorithm initialization and genetic operators, designs a crossover strategy that maintains a good spatial allocation pattern, and deeply embeds real-time spatial constraints calculated by GIS into the algorithm's fitness evaluation and constraint processing stages. It constructs a dynamic optimization model and establishes a model update mechanism linked with the GIS platform and real-time monitoring data. In addition to traditional economic and efficiency objectives, the optimization goals should emphasize regionally specific objectives such as "minimizing dependence on external water transfers" and "maximizing the utilization rate of unconventional water." It provides a complete methodology and tools from spatial analysis and real-time prediction to multi-objective dynamic optimization, specifically addressing the precise water allocation problem in the thermal power industry under multi-source coordination and multi-user competition within water-receiving areas, achieving a leap from static planning to dynamic real-time optimization.

[0074] In one embodiment, such as Figure 1 As shown, a method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ is provided, the method comprising:

[0075] Step S1: Obtain the water supply cost of each water source unit, and calculate the shortest reachable distance between each water source and each water user based on GIS spatial analysis, so as to calculate the water distribution cost from each water source to each water user.

[0076] In step S1, the water transmission cost is derived based on GIS network analysis. For example, the unit water transmission cost from water source i to power plant j. This function can be pre-formed into a cost matrix or relational expression through GIS network analysis and hydraulic calculations, and then embedded into the objective function.

[0077] Step S2: Obtain the real-time maximum available water volume of each water source, the real-time water demand information of each thermal power plant, and the water distribution cost from each water source to each water user in real time. Determine whether the changes in the real-time maximum available water volume, real-time water demand information, or water distribution cost of each water source relative to the previous period exceed a preset threshold. If they exceed the threshold, proceed to the next step.

[0078] In step S2, it is determined whether the changes in the real-time maximum available water volume, real-time water demand information, or water distribution cost of each water source relative to the previous period exceed their respective preset thresholds. If they exceed, the subsequent water distribution scheme update step is triggered.

[0079] Step S3: Establish a multi-objective optimization model with N thermal power plants in the water-receiving area as water users and M types of water sources as water suppliers; the multi-objective optimization model aims to minimize the total cost of water distribution, minimize the total dependence on external water transfer, and maximize the total utilization of reclaimed water, with constraints on the water balance of each thermal power plant and the upper limit of the maximum available water volume of each water source. Figure 2 The image shows a map showing the distribution of thermal power plants in Huai'an City. Figure 3 This is a map showing the elevation distribution of Huai'an City.

[0080] In step S3, the minimum total cost of water distribution is:

[0081]

[0082]

[0083] in, The water supply from water source j to power plant i. The unit cost of water supply from the water source. This is the unit distance water distribution cost coefficient. The reachable distance of the water system between water source j and thermal power plant i, calculated for GIS purposes;

[0084] The minimum total amount of water transferred from outside the region is: ;in, The amount of water supplied to power plant i by external water transfer;

[0085] The maximum total amount of reclaimed water utilized is: ;in, The amount of reclaimed water supplied to power plant i.

[0086] The water balance constraint is: ;in, The water demand of thermal power plant i;

[0087] Maximum water supply capacity constraint: ;in, The maximum available water volume of water source j

[0088] Non-negativity constraint on the available water supply: ;in, The amount of water supplied from water source j to power plant i.

[0089] Step S4: Calculate the allocation weight based on the unit water supply cost of each water source, and perform initial water allocation according to the water demand of each thermal power plant and the allocation weight to obtain an initial allocation scheme; determine whether the total allocation of each water source in the initial allocation scheme exceeds its maximum available water volume. If it does, calculate the scaling factor and proportionally scale the allocation of that water source to each thermal power plant to obtain a first correction scheme; determine whether the total water supply of each thermal power plant in the first correction scheme is equal to its water demand. If they are not equal, calculate the water volume difference and allocate the difference to water sources with remaining capacity according to the allocation weight to obtain a second correction scheme; apply a controllable random perturbation to the second correction scheme to generate the first generation population.

[0090] In step S4, the population size Pop is set, and the initial population P0 is generated according to a fixed four-step process:

[0091] (1) Calculate the cost weight based on the reciprocal of the water supply cost of each water source unit: ;in, The unit cost of water supply from the water source;

[0092] Allocated per power plant:

[0093] Initial allocation based on weight: ;

[0094] (2) Water source upper limit correction: For any water source j, if the total water intake of all power plants exceeds Sjmax, calculate the scaling factor. The water volume of all power plants corresponding to the water source is multiplied synchronously. ;

[0095] (3) The correction may result in a sum for each i. A second correction is needed: calculate the current total water volume of the power plant. ,like Adjustments are made based on priority or candidate water source weights: Calculation of water volume difference. The difference will be calculated as follows: Water should be allocated to sources with remaining capacity; if the shortfall cannot be made up, the individual should be marked as infeasible and recorded. However, minimize residual differences; ensure that the vast majority of solutions in the initial population are feasible. This reduces the search space.

[0096] (4) Add a controllable perturbation to the basic weight allocation: Apply a uniform perturbation to the allocation amount: , Subsequently, the upper limit was adjusted and the balance was repaired again to ensure diversity while favoring low cost.

[0097] (5) Scaling repair for individuals with large cv: Set a repair switch and repair threshold Tcv, only for Individual execution repair:

[0098] For each type of water source j, calculate the total water intake exceeding the upper limit. ,like Then let

[0099] like The power plant supply and demand gap will be redistributed, with priority given to allocating low-cost water sources.

[0100] After repair, the individual constraint violation degree is recalculated. If the constraint cannot be fully satisfied, it is retained as an infeasible solution.

[0101] Step S5: Input the initial population into the improved NSGA-Ⅲ algorithm for iterative evolution to solve the multi-objective optimization model; perform the following operations in each iteration: perform crossover and mutation on the parent population to generate the offspring population, and apply the mutation according to the individual constraint violation degree. Individuals with a constraint violation degree greater than zero have a greater mutation magnitude than individuals with a constraint violation degree equal to zero; merge the parent population and the offspring population into a mixed population, and calculate the target vector and constraint violation degree of each individual in the mixed population; stratify the mixed population based on the constraint violation degree to form a priority layer with a constraint violation degree equal to zero and a secondary layer with a constraint violation degree greater than zero in numerical order; project the normalized target vector of the individuals in the stratified population onto the reference direction. The reference direction densifies the dimension of the reclaimed water utilization target. Based on the individual distribution density under each reference direction, select individuals from the stratified population with a number equal to the population size to form a new generation population.

[0102] In step S5, the population size is set. Total number of iterations Evolutionary operator: SBX simulates binary crossover, crossover probability. Polynomial variation, baseline variation rate ;

[0103] Reference points: Uniform reference points are generated using the Das and Dennis methods. The total number of reference points is... Population size; densify reference points for reclaimed water target dimensions as needed to guide preferences;

[0104] Target normalization: Each generation dynamically uses the maximum and minimum target values ​​of the current population for linear normalization to eliminate dimensional differences. .

[0105] After the iteration enters the main loop, the following steps are included:

[0106] (1) The parent population Pt generates the offspring population through SBX crossover and polynomial mutation. Adaptive mutation improvement is adopted: the mutation rate decreases linearly with the number of iterations, and the mutation amplitude increases for infeasible individuals; scaling repair can be optionally performed for individuals with high CV.

[0107] (2) Merge parent and offspring generations to obtain a mixed population. ; Traverse all individuals and uniformly calculate the target vector. With constraint violation degree cv;

[0108] (3) Strictly apply the Deb constraint dominance rule to rank individuals: a. Feasible solutions Better than any infeasible solution b. For two feasible solutions, the frontier is determined according to the standard Pareto non-dominance relation; c. For two infeasible solutions, the individual with the smaller constraint violation degree (cv) has higher priority.

[0109] (4) Project the normalized target vector onto the reference direction and assign each individual to a reference point; at the same reference point, prioritize retaining feasible solutions and replace infeasible solutions in ascending order of cv; select Pop individuals to form a new generation population based on niche crowding. ;

[0110] (5) When the preset iteration number Gen is reached, or when HV and IGD converge without significant change for several consecutive generations, the iteration is terminated and the final population is output.

[0111] Traditional NSGA-III's non-dominated sorting judges the quality of solutions solely based on the objective function value, and niche selection is based solely on the distance to the reference point for population screening. When applied to thermal power plant water distribution, this may result in a power plant becoming overly dependent on a particular water source, which does not conform to the actual engineering water diversion logic. When performing non-dominated sorting, if the objective function values ​​of two solutions are similar, the solution with a more even water distribution should be selected first.

[0112]

[0113] in For solutions representing the average water supply percentage, during non-dominated sorting, solutions whose objective function values ​​are on the same frontier are preferred. A smaller solution.

[0114] In this way, when the number of power plants, the water demand of each power plant, the maximum water supply of each water source, and the water price are input, the system can output which water sources supply each power plant with how much water, the total cost, the amount of water transferred from other sources, and the amount of reclaimed water used.

[0115] Step S6: When the preset number of iterations is reached or the rate of change of the target value of the population is lower than the preset threshold for multiple consecutive generations, the iteration is terminated and the optimal solution set is output.

[0116] Step S7: Select feasible solutions and optimize the optimal solution set to obtain the final water distribution scheme.

[0117] In step S7, the feasible solution screening and optimization correction of the optimal solution set includes:

[0118] (1) Initial screening of solution sets: Only retain feasible solution sets. As the final candidate. If all are not feasible, select the subset with the smallest cv and fix it to reduce constraint violation, as the backup set.

[0119] (2) Pareto solution set simplification: Knee point identification: identify the compromise optimal point using quadratic curvature or equivalent geometric methods.

[0120] Weighted summation: The Pareto solutions are reordered for the preferences (weight vectors) of several decision-makers, and the solution with the optimal weight is selected.

[0121] Clustering: K-means or hierarchical clustering groups the Pareto solutions, finds representative solutions for each cluster, and reduces the number of alternative solutions.

[0122] (3) Local correction:

[0123] For a selected number of candidate solutions, linear programming or quadratic programming is used for local optimization (with fixed objective weights or multi-objective scalarization) to minimize the main objective or improve the reclaimed water utilization rate while satisfying all constraints. For example, the following single-objective LP (local optimization of a candidate solution) can be used: This approach ensures feasibility while finding the most cost-effective exchange allocation, making it particularly suitable for optimizing the structures provided by NSGA-III into engineering-usable solutions.

[0124] In this process, local optimal fine-tuning does not involve performing a global NSGA-III search again, but rather using a selected representative scheme. Centered on a specific point, linear or quadratic programming is used within its neighborhood for small-scale water exchange, allowing the scheme to further reduce costs or improve reclaimed water utilization while maintaining its original trade-off characteristics. Local optimal fine-tuning includes:

[0125] (a) Selecting a representative solution from the Pareto solution set .

[0126] (b) Set the local adjustment range .

[0127] (c) Fix the allowable degradation range of the other two objectives (e.g., the amount of water transferred from outside the project should not be higher than the original scheme plus the threshold, and the amount of reclaimed water used should not be lower than the original scheme minus the threshold).

[0128] (d) Minimize water distribution cost as the local optimization objective;

[0129] The water balance of thermal power plants, the upper limit of water supply from water sources, and non-negative constraints are retained.

[0130] (e) Replace the original candidate solution with the local optimization result after solving.

[0131] (4) Robustness test: Set up multi-scenario tests for the optimized candidate solutions: maximum available water supply Water demand of power plants The total cost of water distribution, the total amount of water transferred from other regions, and the fluctuation range of reclaimed water utilization for each scheme were statistically analyzed, and water distribution schemes with high sensitivity and poor stability were eliminated.

[0132] In the aforementioned optimization method for multi-source water allocation in thermal power plants based on NSGA-III, this application employs a cost-oriented heuristic population initialization strategy. It calculates allocation weights based on the unit water supply cost of each water source and performs initial allocation according to water demand. Allocations exceeding the water source limit are proportionally scaled and corrected. Allocations that do not meet water balance requirements are compensated for and subject to controllable perturbations. This effectively improves the feasibility of the initial population, ensuring that the vast majority of initial solutions satisfy or approximately satisfy the constraints. It significantly reduces the blind search in infeasible regions in the early stages of the algorithm, lowers computation time, and improves convergence performance. Furthermore, by embedding constraint violation degrees into the hierarchical management of the NSGA-III algorithm... The differentiated mutation mechanism enables targeted repair of individuals with high constraint violation rates and preservation of excellent evolutionary information of individuals with low constraint violation rates, enhancing the algorithm's optimization ability under complex constraint scenarios. By calculating dynamic water transfer costs through GIS spatial analysis and incorporating them into the objective function, it breaks through the simplification of traditional fixed cost coefficients and improves the accuracy of spatial information utilization. By densifying the reference points of the target dimension of reclaimed water utilization, it guides the population to generate optimization schemes with high unconventional water utilization rates, improving the quality of Pareto front distribution. After feasible solution screening, representative scheme extraction, and local optimization correction in the post-processing stage, a water distribution scheme that meets engineering applicability is finally obtained.

[0133] In one embodiment, the constraint violation degree cv is defined for each individual, and the violation amount is calculated and summed for equality constraints (balance) and inequality constraints (upper bound) respectively:

[0134] ;

[0135] in, This is a feasible solution; For a feasible solution, the larger the value, the more severe the constraint violation.

[0136] In one embodiment, densifying the reference direction for the target dimension of reclaimed water utilization includes: in the three-dimensional target space, increasing the density of reference points in the target dimension corresponding to the reclaimed water utilization to k times that of other target dimensions. .

[0137] In one embodiment, a thermal power multi-water source dispatching and optimization system based on NSGA-Ⅲ is provided, comprising:

[0138] Water distribution cost calculation module: used to obtain the water supply cost of each water source unit, and calculate the shortest reachable distance between each water source and each water user based on GIS spatial analysis, so as to calculate the water distribution cost from each water source to each water user.

[0139] Water distribution scheme update trigger module: used to obtain the real-time maximum available water volume of each water source, the real-time water demand information of each thermal power plant, and the water distribution cost from each water source to each water user in real time, and to determine whether the changes of the real-time maximum available water volume, real-time water demand information, or water distribution cost of each water source relative to the previous period exceed the preset threshold. If they exceed the threshold, the multi-objective optimization model construction module is executed.

[0140] Multi-objective optimization model construction module: used to establish a multi-objective optimization model with N thermal power plants in the water receiving area as water users and M types of water sources as water supply sources; the multi-objective optimization model takes the minimum total water distribution cost, the minimum total dependence on external water transfer, and the maximum total utilization of reclaimed water as optimization objectives, and takes the water balance constraints of each thermal power plant and the upper limit of the maximum available water volume of each water source as constraints.

[0141] The initial population generation module is used to calculate the allocation weight based on the unit water supply cost of each water source, and to perform initial water allocation according to the water demand of each thermal power plant and the allocation weight to obtain an initial allocation scheme; it determines whether the total allocation of each water source in the initial allocation scheme exceeds its maximum available water capacity. If it does, it calculates a scaling factor and proportionally scales the allocation of that water source to each thermal power plant to obtain a first correction scheme; it determines whether the total water supply of each thermal power plant in the first correction scheme is equal to its water demand. If they are not equal, it calculates the water difference and allocates the difference to water sources with remaining capacity according to the allocation weight to obtain a second correction scheme; it applies a controllable random perturbation to the second correction scheme to generate the initial population.

[0142] The solution module is used to input the initial population into the improved NSGA-Ⅲ algorithm for iterative evolution and solve the multi-objective optimization model. Each iteration performs the following operations: crossover and mutation of the parent population to generate the offspring population; the mutation is applied differently based on the individual constraint violation degree, with individuals having a constraint violation degree greater than zero exhibiting a greater mutation magnitude than those with a constraint violation degree equal to zero; merging the parent and offspring populations into a mixed population, calculating the objective vector and constraint violation degree of each individual in the mixed population; stratifying the mixed population based on the constraint violation degree, forming a priority layer with a constraint violation degree equal to zero and a secondary layer with constraint violation degrees greater than zero, sorted by numerical value; projecting the normalized objective vectors of the stratified population individuals onto a reference direction, where the reference direction densifies the dimension of the reclaimed water utilization objective; and selecting individuals of equal size from the stratified population based on the individual distribution density under each reference direction to form a new generation population.

[0143] Optimal solution set output module: used to terminate the iteration and output the optimal solution set when the preset number of iterations or the rate of change of the target value of the population for multiple consecutive generations is lower than a preset threshold.

[0144] Correction module: used to screen and optimize the optimal solution set to obtain the final water distribution scheme.

[0145] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for optimizing the dispatch of multiple water sources in thermal power plants based on NSGA-Ⅲ, characterized in that, The method includes: Obtain the water supply cost of each water source unit, and calculate the shortest reachable distance between each water source and each water user based on GIS spatial analysis, so as to calculate the water distribution cost from each water source to each water user. The system acquires the real-time maximum available water volume of each water source, the real-time water demand information of each thermal power plant, and the water distribution cost from each water source to each water user. It then determines whether the changes in the real-time maximum available water volume, real-time water demand information, or water distribution cost of each water source relative to the previous period exceed a preset threshold. If they do, the system proceeds to the next step. A multi-objective optimization model is established with N thermal power plants in the water receiving area as water users and M types of water sources as water supply sources. The multi-objective optimization model takes the minimum total water distribution cost, the minimum total dependence on external water transfer, and the maximum total utilization of reclaimed water as optimization objectives, and the water balance constraints of each thermal power plant and the upper limit of the maximum available water volume of each water source as constraints. Based on the unit water supply cost of each water source, an allocation weight is calculated. An initial water allocation is then performed according to the water demand of each thermal power plant and the allocation weight, resulting in an initial allocation scheme. It is then determined whether the total allocation amount of each water source in the initial allocation scheme exceeds its maximum available water capacity. If it does, a scaling factor is calculated, and the allocation amount of that water source to each thermal power plant is proportionally scaled and corrected, resulting in a first corrected scheme. Next, it is determined whether the total water supply of each thermal power plant in the first corrected scheme is equal to its water demand. If they are not equal, a water difference is calculated, and this difference is allocated to water sources with remaining capacity according to the allocation weight, resulting in a second corrected scheme. A controllable random perturbation is applied to the second corrected scheme to generate an initial population. The initial population is input into the improved NSGA-Ⅲ algorithm for iterative evolution to solve the multi-objective optimization model. Each iteration performs the following operations: crossover and mutation of the parent population to generate the offspring population; the mutation is applied differently based on the individual constraint violation degree, with individuals having a constraint violation degree greater than zero exhibiting a greater mutation magnitude than those with a constraint violation degree equal to zero; the parent and offspring populations are merged into a mixed population, and the objective vector and constraint violation degree of each individual in the mixed population are calculated; the mixed population is stratified based on the constraint violation degree, forming a priority layer with a constraint violation degree equal to zero and a secondary layer with constraint violation degrees greater than zero, ordered numerically; the normalized objective vectors of the stratified population individuals are projected onto a reference direction, which densifies the dimension of the reclaimed water utilization objective; and individuals of equal size to the population are selected from the stratified population based on the individual distribution density under each reference direction to form a new generation population. When the preset number of iterations is reached or the rate of change of the target value of the population is lower than the preset threshold for multiple consecutive generations, the iteration is terminated and the optimal solution set is output. The optimal solution set is then filtered for feasible solutions and optimized to obtain the final water distribution scheme.

2. The method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ according to claim 1, characterized in that, The minimum total cost of water distribution is: in, The water supply from water source j to power plant i. The unit cost of water supply from the water source. This is the unit distance water distribution cost coefficient. The reachable distance of the water system between water source j and thermal power plant i, calculated for GIS purposes; The minimum total amount of water diverted from outside the region is: in, The amount of water supplied to power plant i by external water transfer; The maximum total amount of reclaimed water utilized is: in, The amount of reclaimed water supplied to power plant i.

3. The method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ according to claim 1, characterized in that, The water balance constraint is: in, The water demand of thermal power plant i; The maximum available water supply capacity of the water source is subject to the following constraints: in, The maximum available water volume of water source j; The non-negativity constraint on the available water supply from the water source: in, The amount of water supplied from water source j to power plant i.

4. The method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ according to claim 1, characterized in that, The weighting calculation based on the water supply cost of each water source unit includes: Cost weights are calculated based on the reciprocal of the unit water supply cost for each water source: ； in, The unit cost of water supply from the water source; The allocation weights are obtained by normalizing the cost weights of each water source: 。 5. The method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ according to claim 1, characterized in that, Calculating the scaling factor and proportionally scaling the distribution of this water source across various thermal power plants includes: The scaling factor is calculated based on the ratio of the maximum available water volume to the total allocation of each water source in the initial allocation scheme: in, The maximum available water volume of water source j; The allocation of water sources exceeding the maximum available water volume in the initial allocation scheme to each thermal power plant is scaled and corrected according to a scaling factor.

6. The method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ according to claim 1, characterized in that, Calculate the water volume difference, and allocate the difference among water sources with remaining capacity according to the allocation weight to obtain the second correction scheme, which is as follows: Calculate the water consumption difference for thermal power plants: ;in, The difference In water sources with remaining capacity, water is allocated according to the allocation weight; if it cannot be fully replenished, the constraint violation degree of the population is recorded as greater than zero.

7. The method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ according to claim 1, characterized in that, The second modified scheme, after applying a controllable random perturbation, also includes: Traverse each water source and calculate the amount by which the total allocation of any water source to all thermal power plants in the first generation population exceeds its maximum available water capacity. If the excess is greater than zero, then compress the allocation of that water source to all thermal power plants. Determine whether the total water supply of each thermal power plant after compression is equal to its water demand. If they are not equal, calculate the water difference and allocate the difference to water sources with remaining capacity according to the allocation weight. After repair, recalculate the constraint violation degree of the individual. If the constraint violation degree is still greater than zero, it is retained as an infeasible solution.

8. The method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ according to claim 7, characterized in that, The degree of constraint violation is calculated using the following formula: 。 9. The method for optimizing the multi-water source dispatching of thermal power plants based on NSGA-Ⅲ according to claim 1, characterized in that, The feasible solution screening and optimization correction of the optimal solution set specifically includes: Individuals with a constraint violation rate of zero are retained as final candidates; if the constraint violation rate of all individuals is greater than zero, the individuals with the lowest constraint violation rate are selected for constraint repair, and their constraint violation rate is reduced before they are selected as backup candidates. The final or backup candidates are simplified, the optimal compromise is identified, and they are sorted or clustered according to the decision-maker's preference weight to obtain representative solutions. The representative schemes are fine-tuned locally using linear programming or quadratic programming to optimize the main objective or improve the utilization rate of reclaimed water while satisfying the constraints. For the locally optimal fine-tuned scheme, multiple scenarios are tested under the fluctuation of the maximum water supply from the water source and the water demand of the thermal power plant. The fluctuation range of water distribution cost, total external water transfer and reclaimed water utilization is statistically analyzed. Schemes with fluctuation ranges exceeding the preset threshold are eliminated to determine the final water distribution scheme.

10. A thermal power plant multi-water source dispatch optimization system based on NSGA-Ⅲ, characterized in that, The system includes: Water distribution cost calculation module: used to obtain the water supply cost of each water source unit, and calculate the shortest reachable distance between each water source and each water user based on GIS spatial analysis, so as to calculate the water distribution cost from each water source to each water user. Water distribution scheme update trigger module: used to obtain the real-time maximum available water volume of each water source, the real-time water demand information of each thermal power plant, and the water distribution cost from each water source to each water user in real time, and to determine whether the changes of the real-time maximum available water volume, real-time water demand information, or water distribution cost of each water source relative to the previous period exceed the preset threshold. If they exceed the threshold, the multi-objective optimization model construction module is executed. Multi-objective optimization model construction module: used to establish a multi-objective optimization model with N thermal power plants in the water receiving area as water users and M types of water sources as water supply sources; the multi-objective optimization model takes the minimum total water distribution cost, the minimum total dependence on external water transfer, and the maximum total utilization of reclaimed water as optimization objectives, and takes the water balance constraints of each thermal power plant and the upper limit of the maximum available water volume of each water source as constraints. The initial population generation module is used to calculate the allocation weight based on the unit water supply cost of each water source, and to perform initial water allocation according to the water demand of each thermal power plant and the allocation weight to obtain an initial allocation scheme; it determines whether the total allocation of each water source in the initial allocation scheme exceeds its maximum available water capacity. If it does, it calculates a scaling factor and proportionally scales the allocation of that water source to each thermal power plant to obtain a first correction scheme; it determines whether the total water supply of each thermal power plant in the first correction scheme is equal to its water demand. If they are not equal, it calculates the water difference and allocates the difference to water sources with remaining capacity according to the allocation weight to obtain a second correction scheme; it applies a controllable random perturbation to the second correction scheme to generate the initial population. The solution module is used to input the initial population into the improved NSGA-Ⅲ algorithm for iterative evolution and solve the multi-objective optimization model. Each iteration performs the following operations: crossover and mutation of the parent population to generate the offspring population; the mutation is applied differently based on the individual constraint violation degree, with individuals having a constraint violation degree greater than zero exhibiting a greater mutation magnitude than those with a constraint violation degree equal to zero; merging the parent and offspring populations into a mixed population, calculating the objective vector and constraint violation degree of each individual in the mixed population; stratifying the mixed population based on the constraint violation degree, forming a priority layer with a constraint violation degree equal to zero and a secondary layer with constraint violation degrees greater than zero, sorted by numerical value; projecting the normalized objective vectors of the stratified population individuals onto a reference direction, where the reference direction densifies the dimension of the reclaimed water utilization objective; and selecting individuals of equal size from the stratified population based on the individual distribution density under each reference direction to form a new generation population. Optimal solution set output module: used to terminate the iteration and output the optimal solution set when the preset number of iterations or the rate of change of the target value of the population for multiple consecutive generations is lower than a preset threshold. Correction module: used to screen and optimize the optimal solution set to obtain the final water distribution scheme.

Images