Method for scale demonstration and optimization of inter-basin water transfer project based on artificial intelligence

Abstract

The application discloses a kind of based on artificial intelligence's cross-basin water diversion engineering scale demonstration and optimization method, belong to the water conservancy engineering optimization technical field based on deep learning;First, build water diversion scheme generation dataset and economic mapping dataset, provide basis for model training;Subsequently, a large number of candidate schemes are generated using conditional constraint type generation model, and risk quantification evaluation is carried out by extracting runoff variation and dry-wet ratio and other indexes through runoff regulation calculation and reservoir dispatching simulation;Then, through self-attention and gradient enhancement mechanism, economic indicators are predicted, and multidimensional economic evaluation is realized;Finally, based on hydrological variation and economic evaluation results, a multi-objective optimization function is constructed, and the optimal water diversion engineering scale parameters are selected by improved simulated annealing algorithm;The application realizes the scientific, systematic and intelligent demonstration and optimization of water diversion engineering scale through five links of multi-source dataset construction, conditional constraint scheme generation, hydrological variation index evaluation, engineering scale economic prediction and multi-objective optimization.

Description

Technical Field

[0001] This invention relates to the field of water project scale optimization technology, and is particularly applicable to artificial intelligence-based methods for the scale demonstration and optimization of inter-basin water transfer projects. Background Technology

[0002] Inter-basin water transfer projects, as an important means of national water resource regulation and regional water supply security, are of great significance in alleviating the uneven spatial and temporal distribution of water resources and ensuring water use for domestic, industrial, and agricultural purposes. With the impact of climate change and population growth, hydrological conditions are becoming increasingly complex, and the water supply and demand relationship between the supplying and receiving areas fluctuates significantly. Traditional planning and decision-making methods are insufficient to meet the ever-increasing demand for water transfer. Furthermore, water transfer projects involve substantial engineering investment and long-term operation and maintenance; any unreasonable scale decision may lead to significant economic, social, and ecological risks. Therefore, conducting research on the scientific demonstration of the scale and scheme of inter-basin water transfer projects is of great practical significance for ensuring basin water security, optimizing regional water resource allocation, reducing project implementation risks, and promoting coordinated regional economic and ecological development.

[0003] Currently, the main methods for inter-basin water transfer projects include the following: (1) Water diversion scheme design based on experience and rules: Traditional water diversion scheme design mainly relies on engineers' experience and existing planning specifications, and determines reservoir capacity, pumping station capacity and scheduling scheme through historical cases, empirical formulas and rules. This method has the advantages of being intuitive and easy to use and having low implementation costs, but it relies heavily on personal experience and historical data. It has poor adaptability to changes in hydrological conditions, extreme weather events and complex multi-parameter combinations, and it is difficult to conduct systematic optimization and quantitative risk assessment. The scientificity and repeatability of the scheme are limited.

[0004] (2) Numerical hydrological simulation and optimization methods: The numerical hydrological model-based method realizes the prediction and scheme evaluation of the operation process of water transfer projects by constructing water balance, runoff simulation and scheduling simulation. This technology can reflect the dynamic process of hydrology relatively accurately and support the comparison of multiple schemes. However, numerical simulation has a large computational load, is sensitive to parameters and highly dependent on model assumptions, and is difficult to handle the rapid generation and screening of massive candidate schemes. In addition, it usually lacks the comprehensive optimization capability for economic indicators and multi-objective constraints.

[0005] (3) Single-objective or multi-objective mathematical optimization methods: Some methods use optimization methods such as linear programming, nonlinear programming or genetic algorithms to optimize water transfer schemes with single or multiple objectives, such as minimizing investment costs or maximizing water supply security rate. This method can systematize the scheme selection process to a certain extent, but its disadvantages are that it is sensitive to input data and constraints, the optimization space is prone to getting trapped in local optima, and its scalability is insufficient when dealing with complex water transfer engineering problems with high dimensions, multiple parameters and multiple constraints. At the same time, it lacks the ability to dynamically assess the feasibility of the scheme and hydrological risks.

[0006] In summary, while existing inter-basin water transfer scheme design and optimization technologies each have their advantages, they generally suffer from poor adaptability to hydrological changes, low computational efficiency, difficulty in handling high-dimensional parameter combinations, and a lack of multi-objective comprehensive evaluation. These limitations make it difficult for traditional methods to achieve scientific, systematic, and efficient water transfer scheme demonstration and optimization while ensuring hydrological safety, economic rationality, and ecological sustainability. Summary of the Invention

[0007] The purpose of this invention is to provide a method for the scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence, which can be used to solve the problem.

[0008] To achieve the above objectives, the artificial intelligence-based method for the scale demonstration and optimization of inter-basin water transfer projects described in this invention includes the following steps: S1: Obtain hydrological environmental data and engineering constraints to describe the natural background and boundary conditions of the water diversion project, and obtain the combination of hydrological environment and constraint parameters. ; S2, will The input is fed into the trained conditional constraint-based scheme generation model to generate engineering design schemes covering reservoir capacity, tunnel cross-sectional flow, pump station installed capacity, water diversion period, and scheduling method; the conditional constraint-based scheme generation model adopts a conditional generative adversarial network model. S3 inputs the engineering design scheme generated in S2 into the hydrological variability index assessment module. Using water balance and reservoir simulation scheduling mechanisms, it obtains the complete time series of water transfer, discharge, and abandonment, and calculates the relevant hydrological variability indexes generated by the water transfer project. ; Simultaneously, the engineering design scheme generated by S2 is input into the trained engineering scale economy prediction model, which then predicts and outputs multidimensional economic indicators. This includes investment costs, operation and maintenance expenses, water and soil conservation costs, the impact of resettlement of migrants, and water loss due to water wastage. Based on the prediction results of relevant hydrological variability indicators and multidimensional economic indicators in S3, a multi-objective optimization function that comprehensively considers hydrological safety and engineering economic objectives is constructed in S4. The simulated annealing algorithm is improved to perform global search and iterative optimization, and the optimal combination of water diversion project scale parameters is selected.

[0009] Preferably, the combination of hydrological environment and constraint parameters The hydrological and environmental data includes: runoff from the water source area. Runoff in the water receiving area Regional rainfall Evaporation Water demand in the water-receiving area Engineering constraints include: upper limit of reservoir capacity in the water source area. Tunnel design flow range Ecological outflow .

[0010] Preferably, the relevant hydrological variability index Including the degree of runoff fluctuation , Abundance-Depletion Ratio and comprehensive indicators of hydrological variation .

[0011] Preferably, the condition-constrained scheme generation model includes a hydrological constraint perception generator and an engineering feasibility discriminator; The hydrological constraint sensing generator combines the hydrological environment and constraint parameters. Hydrological environmental parameters and Gaussian noise are input together into a fully connected layer, and nonlinear mapping is performed through the SiLU activation function to obtain hydrological constraint characterization features. ; After passing through four sets of KAN layers and layer standardization, the hydrological structure constraint characteristics were obtained. ; Combining hydrological environment and constraint parameters The engineering constraint parameters are input into a stacked structure of a fully connected layer and a SiLU activation function to obtain the engineering constraint representation features. ; Conditional sensitivity adjustment is performed through a FiLM modulation layer, that is, the engineering constraint features are scaled and offset modulated using hydrological environmental characteristics to generate constraint modulation features. ; Hydrological structural constraints With constraint modulation characteristics Channel splicing is performed to obtain fused candidate features. ; After passing through a bidirectional attention layer, hydrological features can guide the weight allocation of engineering constraint features, and engineering constraint features can also inversely correct the hydrological representation, resulting in bidirectional dependent fusion features. ; Bidirectional dependency fusion features After passing through the CrossNet feature interaction layer, higher-order interaction features are obtained. It is used to capture complex nonlinear interactions; By performing deep reconstruction using two KAN layers, followed by a fully connected layer and a ReLU activation function output, parameters for the water diversion project scale scheme that meet the constraints are generated. .

[0012] Preferably, the engineering feasibility discriminator is used to determine whether the input engineering design scheme is a generated scheme or a real scheme, forming an adversarial training mechanism with the hydrological constraint perception generator; specifically including: After inputting the parameters of the water diversion project scale scheme, preliminary feature extraction is first performed through a fully connected layer to obtain candidate project features. ; Global dependencies are extracted via a self-attention mechanism, and the nonlinear expression is enhanced using a ReLU activation function to obtain engineering feasibility features. ; Candidate project features Engineering feasibility characteristics Residual stitching is performed to obtain robustness enhancement features. This is to alleviate gradient vanishing and improve the model's discriminative performance; Five layers of "fully connected layer - self-attention layer - ReLU activation" units are stacked to obtain discriminative representation features, which are then passed through a fully connected layer and a softmax activation function to output the discriminative result. .

[0013] Preferably, the specific data processing procedure of the hydrological variability index assessment module includes: S3.1.1, Based on candidate engineering design schemes medium storage capacity Tunnel cross-sectional flow Pump station installed capacity Water diversion period Scheduling methods Based on the scheme parameters, an iterative simulation method for water balance is constructed; at each time step... Within, based on the current hydrological environment, the runoff of the water source area Evaporation and water demand in the water-receiving area The processes of water diversion, discharge, and wastewater disposal are obtained through runoff regulation calculations. S3.1.2, Based on the complete time series of water diversion, release, and abandonment after modeling. , as well as Extracting hydrological characteristics of candidate solutions to quantify changes in hydrological conditions and risks; specifically including: First, based on , ,as well as Calculate the average flow rates for the water diversion, discharge, and abandonment sequences respectively. Standard deviation This reflects the overall water volume level and fluctuation range; Secondly, the maxima of the three sequences are counted separately. and minimum value This reflects the characteristics of extreme water volume. Subsequently, the longest duration of continuous high or low flow rates for the three sets of sequences was recorded and analyzed. ; Finally, key events were defined according to preset thresholds, including water diversion flow exceeding the annual water diversion capacity threshold, discharge flow falling below the ecological discharge demand threshold, and water abandonment flow greater than zero. The frequency of key events occurring in the three sets of sequences was then counted. ; S3.1.3, Standard deviation of water diversion, discharge, and wastewater disposal. with the mean The ratios are then averaged and weighted to obtain the overall runoff fluctuation level of the candidate schemes. ; Calculate the maximum flow of the three sequences With minimum flow The average of the ratios is used to obtain the abundance-shortage ratio. The longest duration based on three sequences and frequency of key events Take respectively and The maximum value in the data was used as a salient feature and concatenated to form a comprehensive index of hydrological variability. .

[0014] Preferably, the specific data processing procedure for the project scale economy prediction model includes: S3.2.1, based on the parameters of the candidate engineering design scheme As input, the interdependencies between parameters are first extracted through a self-attention layer, highlighting the key factors that play a dominant role in economic indicators, forming an initial structured feature vector, which is denoted as the associated feature representation. ; S3.2.2, Representation of Association Features First, the basic gradient information vector is obtained by mapping to the potential economic indicator space through a fully connected layer. The input is then fed into the gradient enhancement layer to form a gradient enhancement feature vector. Finally, a nonlinear mapping is achieved using the SiLU activation function to obtain gradient-sensitive features. ; The “fully connected-gradient enhancement-SiLU” structural unit is stacked in three layers to achieve layer-by-layer feature extraction and gradient sensitivity enhancement, forming the final high-level economic representation feature. To obtain high-level economic representation characteristics ; S3.2.3, Characteristics of High-Level Economic Representation The input multi-task prediction head is used to predict economic performance indicators. Each task head consists of a fully connected layer and a ReLU activation function. Finally, the results include investment cost predictions. Operation and maintenance cost forecast results Prediction results of soil and water conservation costs Impact prediction results of immigrant resettlement Water loss prediction results The economic indicators are predicted and evaluated.

[0015] Preferably, the gradient enhancement layer specifically comprises: For the basic gradient information vector The backpropagation gradient vector in Take the absolute value and normalize it with Sigmoid, then map it to the 0-1 interval to obtain the gradient weights; If a gradient boosting method is designed, then the gradient boosting weights are: ,in, Let f be the gradient boosting function, whose parameters are learned during training, where f is the gradient boosting function. This is used to assign enhancement factors to features with large gradients to highlight their contributions. Used to assign a decay factor to features with weak gradient changes in order to reduce interference; The enhanced gradient weights are multiplied element-wise with the base features to form the gradient-enhanced feature vector. .

[0016] Preferably, the S4 process specifically includes: S4.1, Generation of a large number of candidate water diversion project scale schemes: based on hydrological environment and constraint parameters. Using the pre-trained conditional constraint-based scheme generation model as input, a large set of candidate water diversion project scale parameter schemes that satisfy basic hydrological constraints are generated. ,in, This represents the total number of candidate water diversion project scale parameter schemes; S4.2, Multi-objective optimization function design: Combining hydrological variability indicators and economic prediction results, a bi-objective optimization function is defined. ,in, and These represent the hydrological variability index and the economic index for the candidate schemes, respectively. Used to reflect the impact of candidate water diversion schemes on watershed hydrological processes; This is used to characterize the comprehensive economic burden of water diversion schemes in terms of investment and operation, and is obtained directly by summing the evaluation and forecast results of various economic indicators; S4.3, Improved Simulated Annealing Optimization Algorithm: Based on obtaining a large-scale candidate water diversion scheme set, the improved simulated annealing optimization algorithm is used to perform global search and iterative update on the candidate set as the initial solution space to obtain the optimal combination of water diversion project scale parameters.

[0017] Preferably, the improved simulated annealing optimization algorithm is as follows: Based on a bi-objective optimization function, a set of candidate water diversion project scale parameters is analyzed. An evaluation was conducted, and those with the best evaluations and wide distribution were selected. The group of solutions serves as the initial solution cluster, ensuring that the starting point of the optimization process covers diverse regions; During the iteration process, new solutions are generated in two ways: one is to randomly perturb the current solution, and the other is to select and combine schemes similar to the current solution from the candidate set, taking into account both global exploration and local fine-tuning. During the annealing process, the probability of accepting a new solution is obtained from the objective function calculation results and the data distribution of the candidate solutions: if the new solution belongs to the region with better evaluation results in the candidate set, its acceptance probability is increased, and vice versa, thus guiding the algorithm to converge faster in the region where the candidate solution has an advantage.

[0018] Compared with the prior art, the beneficial effects of the present invention are mainly reflected in the following aspects: (1) Intelligent generation of water transfer project schemes: A generative adversarial network model combining hydrological environment perception and engineering constraints is proposed to realize the automatic generation of a large number of high-dimensional water transfer schemes, which not only ensures the physical feasibility of the schemes, but also has diversity and exploratory nature, thus improving the efficiency and scientific nature of scheme design. (2) Quantitative assessment of the impact of hydrological situation: Through simulation of the entire process of water diversion, discharge and abandonment, multi-dimensional hydrological characteristics are extracted and runoff fluctuations, wet-dry ratio and comprehensive variation index are calculated to achieve quantitative assessment of the hydrological risk and adaptability of candidate schemes; (3) Economic prediction and interpretable analysis: Construct a multi-dimensional economic indicator prediction model, map the engineering scale parameters to investment costs, operation and maintenance costs and ecological costs, realize the quantitative assessment of economic efficiency, and reflect the contribution of parameters to economic indicators through gradient analysis, thereby improving the transparency and reliability of decision-making. (4) Multi-objective optimization and intelligent scheme selection: Establish an optimization framework that comprehensively considers the impact of hydrological conditions, economic costs and ecological sustainability, and achieve systematic optimization of the scale of water transfer projects through global search and Pareto frontier screening methods. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the overall technical route of the present invention.

[0020] Figure 2 This is the network structure diagram of the condition-constrained scheme generation model of the present invention.

[0021] Figure 3 This is a network structure diagram of the engineering scale economy prediction model of the present invention.

[0022] Figure 4 This is a schematic diagram illustrating the distribution of candidate water diversion schemes and their adaptability to hydrological constraints in the embodiments.

[0023] Figure 5 This is a diagram showing the quantification of hydrological variation indicators and risk distribution of candidate schemes in the example.

[0024] Figure 6 The example shows a radar chart illustrating the economic prediction of project scale and parameter sensitivity. Detailed Implementation

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] like Figure 1 As shown, the method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to the present invention includes the following steps: S1, Construction of Multidimensional Dataset for Inter-basin Water Transfer Projects: Collect and integrate hydrological environmental data and engineering constraints to describe the natural background and boundary conditions of water transfer projects, including multi-source data such as hydrological runoff, meteorological changes, and typical water transfer project cases. These data are used for generating condition-constrained schemes and predicting economic indicators of the project scale, respectively. The final constructed water transfer scheme generation dataset and engineering economic mapping dataset provide a unified foundation for subsequent model training and ensure that the data can simultaneously support automatic scheme generation and economic prediction modeling.

[0027] S2, Design and Training of Condition-Constrained Scheme Generation Model: Based on the dataset constructed in S1, a condition-constrained scheme generation adversarial network model for water diversion projects is designed and trained. This model is able to generate a large number of candidate water diversion schemes, including combinations of parameters such as reservoir capacity, tunnel cross-sectional flow, and pump station installed capacity, under the input hydrological background and engineering constraints. This forms a set of schemes that can be used for reasoning and screening, while ensuring that the generated schemes meet the hydrological constraints and engineering feasibility conditions.

[0028] S3, Hydrological Variation Index Assessment Module Construction: For the candidate schemes generated in S2, the reasoning mechanism based on the water balance iterative formula and numerical hydrological model is invoked to obtain the complete time series of water transfer, discharge and abandonment, and calculate hydrological variation indicators such as runoff variation rate, dry season guarantee rate and extreme climate adaptability. These are used as key constraint information to evaluate the feasibility of the schemes, so as to quantify the feasibility and risk level of the water transfer schemes under different hydrological conditions.

[0029] S4, Construction of the Engineering Scale and Economy Prediction Model: Based on the candidate schemes generated in S2 and the engineering economic mapping dataset constructed in S1, an engineering scale and economy prediction model combining self-attention mechanism and gradient enhancement mechanism is established to quantitatively predict the investment cost, operation and maintenance cost, soil and water conservation cost, resettlement impact and water abandonment loss of the candidate schemes, thereby realizing a comprehensive evaluation of the economics of the water transfer scheme.

[0030] S5, Multi-objective optimization and scheme selection: Based on the hydrological variability index in S3 and the economic prediction results in S4, a multi-objective optimization function integrating water supply security, economic cost and ecological sustainability is constructed; an improved simulated annealing algorithm is designed to perform global search and iterative optimization, and finally the optimal combination of water diversion project scale parameters is selected from the candidate scheme set to ensure a balance between watershed hydrological security and economic rationality.

[0031] The specific implementation process of the present invention will be described in detail below with reference to specific embodiments.

[0032] I. Construction of Multidimensional Datasets for Inter-basin Water Transfer Projects To address the issues of experience-dependent generation of water transfer project scale schemes and lack of training foundation for economic evaluation during the planning phase, this invention first constructs a multidimensional dataset for inter-basin water transfer projects. This dataset aims to simultaneously support automatic scheme generation and economic prediction modeling. The final dataset construction yields both a project scale scheme generation dataset and an economic indicator prediction dataset. Specifically, the process includes the following steps: Hydrological environment and constraint parameter selection: Hydrological environment data selection: Runoff in the water source area Runoff in the water receiving area Regional rainfall Evaporation Water demand in the water-receiving area Engineering constraint selection: Upper limit of reservoir capacity in the water source area Tunnel design flow range Ecological outflow .

[0033] The above data were obtained through actual measurements at hydrological monitoring stations, meteorological observation records, and planning boundary data, and are used to describe the natural background and boundary conditions of the water diversion project. Therefore, the complete combination of hydrological environment and constraint parameters is as follows: ; Engineering parameter selection for the design scheme: Selecting the storage capacity. Tunnel cross-sectional flow Pump station installed capacity Water diversion period Scheduling methods As parameters of the scheme, these parameters directly determine the scale and combination scheme of the water transfer project.

[0034] Therefore, the engineering parameters of the design scheme are combined as follows: Furthermore, the specific values ​​of the engineering parameters in the design scheme are determined in conjunction with historical water diversion project design data and planning specifications, providing output targets for subsequent scheme generation.

[0035] Selection of economic indicator parameters: Determining investment costs Operation and maintenance costs The cost of soil and water conservation Impact of immigrant resettlement Water loss As an economic indicator; Therefore, the combination of economic indicator parameters is as follows: Furthermore, this type of data is obtained through the final settlement data of completed water diversion projects and statistical yearbooks, and is used to reflect the economic benefits and costs under different scale combinations.

[0036] Dataset creation for water diversion schemes: Combination of hydrological environment and constraint parameters As input, the engineering parameters of the design scheme are combined. As output, a set of engineering-scale scheme generation datasets are formed; and based on this method, extensive data collection is carried out under different watersheds, different climate conditions and different supply and demand relationships, and finally a complete water diversion scheme generation dataset is formed, which is used to train the condition-constrained scheme generation model.

[0037] Dataset creation for economic indicator forecasting: based on a combination of engineering parameters from design schemes. As input, a combination of economic indicator parameters As output, a set of economic indicator prediction datasets is formed; and based on this method, cross-regional socio-economic statistics are introduced to collect a wide range of data, and finally a complete economic indicator prediction dataset is formed, which is used to train the engineering scale economy prediction model.

[0038] II. Design and Training of Constrained Solution Generation Model To achieve efficient generation of inter-basin water transfer schemes under hydrological conditions and ensure that the generated schemes meet physical constraints and engineering feasibility, this invention constructs a condition-constrained scheme generation model. This model includes a hydrological constraint-aware generator and an engineering feasibility discriminator, and is trained end-to-end using a constructed water transfer scheme generation dataset. The trained model, using the hydrological environment and constraint parameters as input, can generate a large number of engineering-scale schemes that meet the constraints. The condition-constrained scheme generation model is as follows: Figure 2 As shown.

[0039] 1. Design of Hydrological Constraint Sensing Generator: Based on the combination of hydrological environment and constraint parameters As input, it is used to generate engineering design schemes that meet hydrological conditions and physical constraints; the purpose of this design is to ensure that the generated schemes are feasible under hydrological variations and constraints, while enhancing the generator's ability to express complex nonlinear relationships. Specifically: (1) Combining hydrological environment and constraint parameters Hydrological environmental parameters and Gaussian noise are input together into a fully connected layer, and nonlinear mapping is performed through the SiLU activation function to obtain hydrological constraint characterization features. ; Gaussian noise is used to enhance the diversity of solutions and ensure the exploratory nature of the generated solutions. Subsequently, this characterization feature... After passing through four sets of KAN layers and layer standardization, the hydrological structure constraint characteristics were obtained. This feature can better capture the nonlinear laws and implicit constraints in the hydrological environment.

[0040] (2) Combining hydrological environment and constraint parameters The engineering constraint parameters are input into a stacked structure of a fully connected layer and a SiLU activation function to obtain the engineering constraint representation features. Subsequently, this feature undergoes conditional sensitivity adjustment through a FiLM modulation layer, that is, scaling and offset modulation of the engineering constraint feature using hydrological environmental features to generate constraint modulation features. This is to ensure that the generated scheme meets both the hydrological environment requirements and the engineering constraints. Specifically, the FiLM modulation layer generates a set of scaling and offset coefficients to linearly transform the engineering constraint features channel by channel. This mechanism enables the engineering constraint features to dynamically respond to changes in the hydrological environment in the numerical space, thereby ensuring stronger physical consistency of the generated scheme under different hydrological conditions.

[0041] (3) Constraining features of hydrological structure With constraint modulation characteristics Channel splicing is performed to obtain fused candidate features. This feature, after passing through a bidirectional attention layer, allows the hydrological feature to guide the weight allocation of the engineering constraint feature, while the engineering constraint feature can also inversely correct the hydrological representation, resulting in a bidirectional dependent fusion feature. This mechanism effectively avoids the problem of imbalance due to one-way constraints, and improves the physical rationality and diversity of the generated schemes.

[0042] Specifically, the bidirectional attention layer employs an interactive attention mechanism: on one hand, attention is calculated using hydrological environmental features as queries and engineering constraint features as keys to obtain environment-driven constraint responses; on the other hand, attention is calculated using engineering constraint features as queries and hydrological environmental features as keys to obtain constraint-driven environmental feedback. This bidirectional interaction captures the bidirectional dependency between hydrological conditions and engineering constraints, thereby ensuring the engineering feasibility of the generated scheme.

[0043] Subsequently, bidirectional dependency fusion features After passing through the CrossNet feature interaction layer, higher-order interaction features are obtained. This feature is used to capture complex nonlinear interactions. Subsequently, it undergoes deep reconstruction through two KAN layers, followed by a fully connected layer and a ReLU activation function to output the final parameters for a water diversion scheme that meets the given constraints. ; Among them, the CrossNet feature interaction layer explicitly captures the nonlinear correlation between multidimensional features through layer-by-layer vector inner product and weight superposition, thereby improving the expressive power and stability of the scheme under multi-parameter coupling conditions.

[0044] 2. Engineering Feasibility Discriminator Design: This device determines whether the input engineering design scheme is a generated scheme or a real scheme, thus forming an adversarial training mechanism with the hydrological constraint perception generator. Its structure includes: (1) Input scheme parameters ( or After including the design parameters of the actual water diversion scheme and the design parameters of the water diversion scheme generated by the condition-constrained scheme generation model, preliminary feature extraction is first performed through a fully connected layer to obtain the candidate project features. ; (2) Global dependencies are extracted via a self-attention mechanism and nonlinear expressions are enhanced by a ReLU activation function to obtain engineering feasibility features. ; (3) Select candidate project features Engineering feasibility characteristics Residual stitching is performed to obtain robustness enhancement features. This is to alleviate gradient vanishing and improve the model's discriminative performance; (4) Stack the “fully connected layer-self attention layer-ReLU activation” units into 5 layers to obtain the discriminative representation features, and then output the discriminative result through the fully connected layer and the Softmax activation function. .

[0045] 3. Loss Function Design and Model Training: The engineering feasibility discriminator and the hydrological constraint-aware generator are trained using a game-theoretic approach based on a conditional adversarial loss function to ensure the realism and diversity of the generated solutions under constraints. Simultaneously, the generated solutions are... Compared to the actual solution The mean squared error between the two is used as an auxiliary loss to further improve the similarity between the generated scheme and the real scheme at the distribution level; Through the above training mechanism, the model is trained using the constructed dataset of water diversion project scale schemes, and finally a conditional constraint scheme generation model that converges under Nash equilibrium conditions is obtained.

[0046] III. Construction of the Hydrological Variation Index Assessment Module Candidate schemes generated based on a condition-constrained scheme generation model are used to obtain complete time series of water transfer, discharge, and abandonment through the water balance iterative model formula and the reasoning mechanism of the numerical hydrological model. Based on this, relevant hydrological variability indicators such as runoff variability, drought relief, and comprehensive variability that may result from the implementation of the water transfer project are calculated, achieving a quantitative assessment of the hydrological variability of the candidate schemes. The specific process is as follows: S3.1.1, Simulation Modeling of Scheduling Process: Based on Candidate Project Design Scheme medium storage capacity Tunnel cross-sectional flow Pump station installed capacity Water diversion period Scheduling methods Based on the scheme parameters, an iterative simulation method for water balance is constructed. Specifically, at each time step... Within, based on the current hydrological environment, the runoff of the water source area Evaporation and leakage losses Water demand in the water-receiving area By combining the ecological water demand of the water source area, the processes of water regulation, discharge and abandonment are obtained through iterative water balance formula.

[0047] (1) Determine the amount of water entering the reservoir Inflow = Natural water supply - Water demand from upstream industries, domestic and agricultural sectors (2) Determine the storage capacity at the end of the time period End-of-period reservoir capacity = Beginning-of-period reservoir capacity + Inflow - Outflow - Seepage loss (3) Determine the outflow volume Outflow of water = water required for downstream river ecology + water diversion + water discarded.

[0048] The water diversion volume is determined by assessing the relationship between the available water volume and the required water volume in the current period, and the discharge volume is determined by the ecological water demand of the downstream river channel, the water demand of the water source area, and the abandoned water. The abandoned water volume is determined by determining the reservoir capacity at the end of the period based on the water balance, assessing its relationship with the normal water level, and determining whether abandoned water will be generated. If the reservoir capacity at the end of the period is greater than the reservoir capacity corresponding to the normal water level, abandoned water will be generated; otherwise, no abandoned water will be generated.

[0049] S3.1.2, Hydrological Feature Extraction: Based on the complete time series of water diversion, discharge, and abandonment after modeling. , ,as well as Extracting hydrological characteristics of candidate solutions to quantify changes in hydrological conditions and risks; specifically including: First, based on , ,as well as Calculate the average flow rates for the water diversion, discharge, and abandonment sequences respectively. Standard deviation This reflects the overall water volume level and fluctuation range; Secondly, the maxima of the three sequences are counted separately. and minimum value This reflects the characteristics of extreme water volume. Subsequently, the longest duration of continuous high or low flow rates for the three sets of sequences was recorded and analyzed. ; Finally, key events were defined according to preset thresholds: water diversion flow exceeding the annual water diversion capacity threshold, discharge flow falling below the ecological discharge demand threshold, and water abandonment flow greater than zero. The frequency of occurrence in the three sets of sequences was then statistically analyzed. .

[0050] S3.1.3, Calculation of hydrological variability index: using the extracted hydrological characteristics to calculate the average flow rate. Standard deviation Maximum value Minimum value Longest duration and frequency Calculate hydrological variability indices. Specifically: the standard deviation of water diversion, discharge, and wastewater discharge. with the mean The ratios are then averaged and weighted to obtain the overall runoff fluctuation level of the candidate schemes. ; Calculate the maximum flow of the three sequences With minimum flow The average of the ratios is used to obtain the abundance-shortage ratio. The longest duration based on three sequences and frequency of key events Take respectively and The maximum value in the data was used as a salient feature and concatenated to form a comprehensive index of hydrological variability. .

[0051] Finally, regarding the candidate project scale scheme This yields results including the degree of runoff fluctuation. , Abundance-Depletion Ratio and comprehensive indicators of hydrological variation The assessment results of hydrological variability indicators, namely hydrological variability indicators .

[0052] IV. Construction of the Project Scale Economy Prediction Model Based on the constructed engineering economic mapping dataset, this invention designs an economic prediction model combining a self-attention mechanism and a gradient enhancement mechanism to achieve mapping prediction of candidate schemes' multi-dimensional economic indicators. These economic indicators include investment costs, operation and maintenance expenses, soil and water conservation costs, resettlement impacts, and water loss, thereby providing a comprehensive economic evaluation of candidate project scale schemes. The network structure of the prediction model is as follows: Figure 3 As shown.

[0053] S3.2.1, Engineering parameter input and self-attention feature screening: based on candidate engineering design scheme parameters (Including reservoir capacity, tunnel cross-sectional flow, pumping station installed capacity, water diversion period, and scheduling mode parameters) as input, the interdependencies between parameters are first extracted through a self-attention layer to highlight the key factors that play a dominant role in economic indicators, forming an initial structured feature vector, which is denoted as the associated feature representation. .

[0054] S3.2.2, Gradient-Driven Feature Enhancement: Correlation Feature Representation First, the basic gradient information vector is obtained by mapping to the potential economic indicator space through a fully connected layer. The gradient enhancement layer is then input to selectively amplify important features and dynamically suppress weakly correlated features, specifically including: (1) For the basic gradient information vector The backpropagation gradient vector in (Represents the gradient value corresponding to the i-th neuron) Take the absolute value and normalize it with Sigmoid, then map it to the 0-1 interval to obtain the gradient weight; (2) Design a gradient enhancement method, then the gradient enhancement weights are: ,in, This is the gradient boosting function, whose parameters are learned during training. This is used to assign enhancement factors to features with large gradients to highlight their contribution. Used to assign a decay factor to features with weak gradient changes in order to reduce their interference; (3) Multiply the enhanced gradient weights element by element with the basic features to form the gradient-enhanced feature vector. .

[0055] Finally, the SiLU activation function is used to achieve nonlinear mapping, resulting in gradient-sensitive features. The "fully connected-gradient enhancement-SiLU" structural unit is stacked in three layers to achieve layer-by-layer feature extraction and gradient sensitivity enhancement, forming the final high-level economic representation feature. .

[0056] S3.2.3, Multidimensional Economic Indicator Forecasting and Output: Incorporating high-level economic characteristics The input multi-task prediction head is used to predict economic performance indicators. Each task head consists of a fully connected layer and a ReLU activation function. The final result includes: investment cost prediction results. Operation and maintenance cost forecast results Prediction results of soil and water conservation costs Impact prediction results of immigrant resettlement Water loss prediction results Together, they constitute the results of the economic indicator forecast and evaluation. .

[0057] Training and optimization of the engineering scale economic prediction model: Based on the constructed engineering economic mapping dataset, the Adam optimizer is used to train the economic prediction model, and the parameters are iteratively updated until the prediction error converges. By introducing the interpretability mechanism of the gradient enhancement layer, the engineering parameters that contribute significantly to the economic assessment can be identified and highlighted during the model training process, thereby improving the reliability of the prediction and finally obtaining the trained engineering scale economic prediction model.

[0058] V. Multi-objective optimization and optimal solution selection To ensure the hydrological safety of the watershed while achieving economic rationality in the scale of water diversion projects, this invention proposes a multi-objective optimization method that comprehensively considers both hydrological safety and economic objectives. This method is based on a trained, conditionally constrained scheme generation model, a constructed hydrological variability index evaluation module, and an established project scale economic prediction model. By constructing a joint objective function and designing an improved simulated annealing algorithm for global search and iterative updates, the optimal combination of water diversion project scale parameters is finally obtained. The specific steps include: S4.1, Generation of a large number of candidate water transfer schemes: based on the combination of hydrological environment and constraint parameters. Using the pre-trained conditionally constrained generative model as input, a large set of candidate water diversion project scale parameter schemes that satisfy basic hydrological constraints are generated. ,in, This represents the total number of candidate water diversion project scale parameter schemes.

[0059] S4.2, Multi-objective optimization function design: Combining the hydrological variability index assessment results and economic prediction results, a bi-objective optimization function is defined. ,in, and These represent the candidate solutions. The assessment results of hydrological variability indicators and the prediction and assessment results of economic indicators; specifically: This function is used to reflect the impact of candidate water diversion schemes on the hydrological processes of the watershed. It takes three hydrological variability indicators extracted from S3 as input: runoff fluctuation degree, wet-dry ratio, and comprehensive hydrological variability index. The specific calculations include: First, finding the maximum and minimum values ​​corresponding to the three indicators, and then normalizing the three indicators using the min-max method; subsequently, determining the importance of each indicator in the overall function according to a preset weighting method; finally, obtaining the result by weighted summation of the three normalized indicators. The larger the value, the stronger the disturbance to the hydrological process, and thus the worse the hydrological safety.

[0060] It is used to characterize the comprehensive economic burden of water diversion schemes in terms of investment and operation. It is obtained by summing the evaluation and prediction results of various economic indicators (including: investment costs, operation and maintenance costs, water and soil conservation costs, impact of resettlement of immigrants, and water loss). The larger the value, the higher the economic cost of the scheme and the worse its economic efficiency.

[0061] S4.3, Improved Simulated Annealing Optimization Algorithm Design: Based on obtaining a massive set of candidate water diversion schemes, this invention proposes an improved simulated annealing optimization algorithm. By leveraging the advantages of a large number and wide coverage of candidate solutions, the candidate set is used as the initial solution space to improve the diversity of the search starting point and the global exploration capability. Specifically, it includes: (1) Based on the dual-objective optimization function, the set of candidate water diversion project scale parameters is optimized. An evaluation was conducted, and those with the best evaluations and wide distribution were selected. The group of solutions serves as the initial solution cluster, ensuring that the starting point of the optimization process covers diverse regions; (2) During the iteration process, the generation of new solutions combines two methods: one is to randomly perturb the current solution, and the other is to select a scheme similar to the current solution from the candidate set and combine them, so as to take into account both global exploration and local fine-tuning. (3) During the annealing process, the probability of accepting a new solution is obtained from the objective function calculation results and the data distribution of the candidate solutions. Specifically, if the new solution belongs to the region with better evaluation results in the candidate set, its acceptance probability is increased, and vice versa, thereby effectively guiding the algorithm to converge faster in the region with the advantage of the candidate solution.

[0062] Optimal water diversion project scale parameter combination generation: After the temperature drops to a stable level, the algorithm outputs a set of Pareto front schemes, reflecting the balance between hydrological safety and economy under different trade-offs; finally, the user selects the final optimal water diversion project scale parameter combination from the front solution set according to actual needs and preferences, as a reference scheme for project implementation.

[0063] VI. Analysis of Experimental Results To verify the effectiveness of the proposed intelligent generation and multi-objective optimization method for inter-basin water transfer schemes, this experiment takes typical water transfer engineering cases as the object, covering multiple dimensions such as reservoir capacity, tunnel cross-sectional flow, pump station installed capacity, water transfer period and scheduling mode. The experiment focuses on scheme generation, hydrological impact of water transfer and economic indicators, and aims to evaluate the performance of the proposed method in intelligent generation of water transfer schemes, adaptability to constraints, risk quantification analysis and economic optimization.

[0064] 1. Generation and Constraint Adaptability Verification of Candidate Water Transfer Schemes This experiment first utilizes a condition-constrained scheme generation model to generate a large number of candidate water transfer schemes. Using combinations of hydrological and engineering constraint parameters as input, a set of schemes satisfying both hydrological and engineering constraints is generated. To visually demonstrate the diversity of schemes and their adaptability to constraints, this experiment employs a multi-dimensional scatter mapping and watershed hydrological heatmap projection visualization method. (1) Map the key parameters of the candidate schemes (reservoir capacity, pump station installed capacity, tunnel cross-sectional flow) to a three-dimensional scatter space; (2) Different colors represent different water diversion periods, and different point sizes represent differences in scheduling methods; (3) The hydrological constraint feasibility score is mapped to scattered transparency. The less transparent the score, the more the scheme conforms to the hydrological constraints.

[0065] Experimental results are as follows Figure 4 As shown in the figure, the results demonstrate that the generated model uniformly covers candidate schemes in the multi-parameter space, with low-transparency points mainly concentrated in the reasonable reservoir capacity and pump station capacity regions, verifying the model's high adaptability to constraints under complex nonlinear conditions. Furthermore, the bidirectional attention layer and CrossNet feature interaction mechanism effectively avoid unidirectional constraint shifts, ensuring both diversity and physical feasibility in the generated schemes.

[0066] 2. Quantification of hydrological variability indicators and risk analysis of candidate solutions In the hydrological variability quantification experiment, the hydrological variability index assessment module was used to model the time series of water diversion, discharge, and abandonment flows of candidate schemes, and to calculate the degree of runoff fluctuation, the ratio of abundant to dry periods, and comprehensive hydrological variability indices. The experiment introduced a three-dimensional time-space-scheme matrix visualization, combined with a watershed runoff heat map to illustrate the impact of the schemes on the hydrological process. The horizontal axis represents the time step (indicating the hydrological variability index assessment results at different times, with monthly time steps), the vertical axis represents the different candidate scheme numbers, and the color gradient represents the degree of deviation of the water diversion flow or runoff. By overlaying the matrix to display discharge and abandonment behaviors, a comprehensive hydrological impact map was formed, which can intuitively identify abnormal schemes and high-risk water diversion strategies.

[0067] Experimental results are as follows Figure 5 As shown, different candidate schemes exhibit significant differences during the dry season. Some high-reservoir-capacity schemes lead to higher comprehensive hydrological variability indices, verifying the method's ability to quantify hydrological risks and providing constraints and risk quantification basis for multi-objective optimization.

[0068] 3. Project scale economy prediction and key parameter sensitivity analysis For candidate schemes, an economic prediction model is used to predict multidimensional economic indicators (investment cost, operation and maintenance cost, soil and water conservation cost, impact of resettlement and water loss), and the sensitivity of key engineering parameters to economic efficiency is further analyzed.

[0069] like Figure 6 As shown, the economic prediction results of each scheme are displayed by filling the area, and the gradient enhancement weights are mapped to color gradients to show the contribution of the parameters to different indicators. The results show that the tunnel design flow rate and reservoir capacity contribute the most to investment costs and operation and maintenance expenses, while water discharge loss is highly sensitive to the scheduling method. This visualization method intuitively displays the distribution of economic indicators and also reflects the interpretability information provided by the gradient enhancement mechanism during model training.

[0070] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0071] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for the scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence, characterized in that, Includes the following steps: S1: Obtain hydrological environmental data and engineering constraints to describe the natural background and boundary conditions of the water diversion project, and obtain the combination of hydrological environment and constraint parameters. ; S2, will The input is fed into the trained conditional constraint-based scheme generation model to generate engineering design schemes covering reservoir capacity, tunnel cross-sectional flow, pump station installed capacity, water diversion period, and scheduling method; the conditional constraint-based scheme generation model adopts a conditional generative adversarial network model. S3 inputs the engineering design scheme generated in S2 into the hydrological variability index assessment module. Using water balance and reservoir simulation scheduling mechanisms, it obtains the complete time series of water transfer, discharge, and abandonment, and calculates the relevant hydrological variability indexes generated by the water transfer project. ; Simultaneously, the engineering design scheme generated by S2 is input into the trained engineering scale economy prediction model, which then predicts and outputs multidimensional economic indicators. This includes investment costs, operation and maintenance expenses, water and soil conservation costs, the impact of resettlement of migrants, and water loss due to water wastage. Based on the prediction results of relevant hydrological variability indicators and multidimensional economic indicators in S3, a multi-objective optimization function that comprehensively considers hydrological safety and engineering economic objectives is constructed in S4. The simulated annealing algorithm is improved to perform global search and iterative optimization, and the optimal combination of water diversion project scale parameters is selected.

2. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 1, characterized in that: The combination of hydrological environment and constraint parameters The hydrological and environmental data includes: runoff from the water source area. Runoff in the water receiving area Regional rainfall Evaporation Water demand in the water-receiving area Engineering constraints include: upper limit of reservoir capacity in the water source area. Tunnel design flow range Ecological outflow .

3. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 1, characterized in that: The relevant hydrological variation indicators Including the degree of runoff fluctuation , Abundance-Depletion Ratio and comprehensive indicators of hydrological variation .

4. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 1, characterized in that: The condition-constrained scheme generation model includes a hydrological constraint perception generator and an engineering feasibility discriminator. The hydrological constraint sensing generator combines the hydrological environment and constraint parameters. Hydrological environmental parameters and Gaussian noise are input together into a fully connected layer, and nonlinear mapping is performed through the SiLU activation function to obtain hydrological constraint characterization features. ; After passing through four sets of KAN layers and layer standardization, the hydrological structure constraint characteristics were obtained. ; Combining hydrological environment and constraint parameters The engineering constraint parameters are input into a stacked structure of a fully connected layer and a SiLU activation function to obtain the engineering constraint representation features. ; Conditional sensitivity adjustment is performed through a FiLM modulation layer, that is, the engineering constraint features are scaled and offset modulated using hydrological environmental characteristics to generate constraint modulation features. ; Hydrological structural constraints With constraint modulation characteristics Channel splicing is performed to obtain fused candidate features. ; After passing through a bidirectional attention layer, hydrological features can guide the weight allocation of engineering constraint features, and engineering constraint features can also inversely correct the hydrological representation, resulting in bidirectional dependent fusion features. ; Bidirectional dependency fusion features After passing through the CrossNet feature interaction layer, higher-order interaction features are obtained. It is used to capture complex nonlinear interactions; By performing deep reconstruction using two KAN layers, followed by a fully connected layer and a ReLU activation function output, parameters for the water diversion project scale scheme that meet the constraints are generated. .

5. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 4, characterized in that: The engineering feasibility discriminator is used to determine whether the input engineering design scheme is a generated scheme or a real scheme, forming an adversarial training mechanism with the hydrological constraint perception generator; specifically including: After inputting the parameters of the water diversion project scale scheme, preliminary feature extraction is first performed through a fully connected layer to obtain candidate project features. ; Global dependencies are extracted via a self-attention mechanism, and the nonlinear expression is enhanced using a ReLU activation function to obtain engineering feasibility features. ; Candidate project features Engineering feasibility characteristics Residual stitching is performed to obtain robustness enhancement features. This is to alleviate gradient vanishing and improve the model's discriminative performance; Five layers of "fully connected layer - self-attention layer - ReLU activation" units are stacked to obtain discriminative representation features, which are then passed through a fully connected layer and a softmax activation function to output the discriminative result. .

6. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 1, characterized in that: The specific data processing procedure of the hydrological variability index assessment module includes: S3.1.1, Based on candidate engineering design schemes medium storage capacity Tunnel cross-sectional flow Pump station installed capacity Water diversion period Scheduling methods Based on the scheme parameters, an iterative simulation method for water balance is constructed; at each time step... Within, based on the current hydrological environment, the runoff of the water source area Evaporation and water demand in the water-receiving area The processes of water diversion, discharge, and wastewater disposal are obtained through runoff regulation calculations. S3.1.2, Based on the complete time series of water diversion, release, and abandonment after modeling. , as well as Extracting hydrological characteristics of candidate solutions to quantify changes in hydrological conditions and risks; specifically including: First, based on , ,as well as Calculate the average flow rates for the water diversion, discharge, and abandonment sequences respectively. Standard deviation This reflects the overall water volume level and fluctuation range; Secondly, the maxima of the three sequences are counted separately. and minimum value This reflects the characteristics of extreme water volume. Subsequently, the longest duration of continuous high or low flow rates for the three sets of sequences was recorded and analyzed. ; Finally, key events were defined according to preset thresholds, including water diversion flow exceeding the annual water diversion capacity threshold, discharge flow falling below the ecological discharge demand threshold, and water abandonment flow greater than zero. The frequency of key events occurring in the three sets of sequences was then counted. ; S3.1.3, Standard deviation of water diversion, discharge, and wastewater disposal. with the mean The ratios are then averaged and weighted to obtain the overall runoff fluctuation level of the candidate schemes. ; Calculate the maximum flow of the three sequences With minimum flow The average of the ratios is used to obtain the abundance-shortage ratio. The longest duration based on three sequences and frequency of key events Take respectively and The maximum value in the data was used as a salient feature and concatenated to form a comprehensive index of hydrological variability. .

7. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 1, characterized in that: The specific data processing steps of the project scale economic prediction model include: S3.2.1, based on the parameters of the candidate engineering design scheme As input, the interdependencies between parameters are first extracted through a self-attention layer, highlighting the key factors that play a dominant role in economic indicators, forming an initial structured feature vector, which is denoted as the associated feature representation. ; S3.2.2, Representation of Association Features First, the basic gradient information vector is obtained by mapping to the potential economic indicator space through a fully connected layer. The input is then fed into the gradient enhancement layer to form a gradient enhancement feature vector. Finally, a nonlinear mapping is achieved using the SiLU activation function to obtain gradient-sensitive features. ; The "fully connected-gradient enhancement-SiLU" structural unit is stacked in three layers to achieve layer-by-layer feature extraction and gradient sensitivity enhancement, forming the final high-level economic representation feature. To obtain high-level economic representation characteristics ; S3.2.3, Characteristics of High-Level Economic Representation The input multi-task prediction head is used to predict economic performance indicators. Each task head consists of a fully connected layer and a ReLU activation function. Finally, the results include investment cost predictions. Operation and maintenance cost forecast results Prediction results of soil and water conservation costs Impact prediction results of immigrant resettlement Water loss prediction results The economic indicators are predicted and evaluated.

8. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 7, characterized in that: The gradient enhancement layer is specifically: For the basic gradient information vector The backpropagation gradient vector in Take the absolute value and normalize it with Sigmoid, then map it to the 0-1 interval to obtain the gradient weights; If a gradient boosting method is designed, then the gradient boosting weights are: ,in, Let f be the gradient boosting function, whose parameters are learned during training, where f is the gradient boosting function. This is used to assign enhancement factors to features with large gradients to highlight their contributions. Used to assign a decay factor to features with weak gradient changes in order to reduce interference; The enhanced gradient weights are multiplied element-wise with the base features to form the gradient-enhanced feature vector. .

9. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 1, characterized in that: The specific process of S4 includes: S4.1, Generation of a large number of candidate water diversion project scale schemes: based on hydrological environment and constraint parameters. Using the pre-trained conditional constraint-based scheme generation model as input, a large set of candidate water diversion project scale parameter schemes that satisfy basic hydrological constraints are generated. ,in, This represents the total number of candidate water diversion project scale parameter schemes; S4.2, Multi-objective optimization function design: Combining hydrological variability indicators and economic prediction results, a bi-objective optimization function is defined. ,in, and These represent the hydrological variability index and the economic index for the candidate schemes, respectively. Used to reflect the impact of candidate water diversion schemes on watershed hydrological processes; This is used to characterize the comprehensive economic burden of water diversion schemes in terms of investment and operation, and is obtained directly by summing the evaluation and forecast results of various economic indicators; S4.3, Improved Simulated Annealing Optimization Algorithm: Based on obtaining a large-scale candidate water diversion scheme set, the improved simulated annealing optimization algorithm is used to perform global search and iterative update on the candidate set as the initial solution space to obtain the optimal combination of water diversion project scale parameters.

10. The method for scale demonstration and optimization of inter-basin water transfer projects based on artificial intelligence according to claim 9, characterized in that: The improved simulated annealing optimization algorithm is specifically as follows: Based on a bi-objective optimization function, a set of candidate water diversion project scale parameters is analyzed. An evaluation was conducted, and those with the best evaluations and wide distribution were selected. The group of solutions serves as the initial solution cluster, ensuring that the starting point of the optimization process covers diverse regions; During the iteration process, new solutions are generated in two ways: one is to randomly perturb the current solution, and the other is to select and combine schemes similar to the current solution from the candidate set, taking into account both global exploration and local fine-tuning. During the annealing process, the probability of accepting a new solution is obtained from the objective function calculation results and the data distribution of the candidate solutions: if the new solution belongs to the region with better evaluation results in the candidate set, its acceptance probability is increased, and vice versa, thus guiding the algorithm to converge faster in the region where the candidate solution has an advantage.

Images