Regional sluice system scheduling optimization method and system

By using multimodal data fusion and deep learning technologies, a multi-scale hydrodynamic model is constructed to collaboratively optimize sluice gate scheduling. This solves the problems of hydrological data processing and scheduling optimization in existing technologies, and realizes high-precision and intelligent water resource management and ecological protection.

CN119005064APending Publication Date: 2024-11-22NANJING HYDRAULIC RES INST
View PDF 0 Cites 47 Cited by

Patent Information

Application Number
CN202411225048.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing technologies struggle to capture complex dynamic features when processing high-dimensional, nonlinear, and non-stationary hydrological time series data. They also fail to adequately utilize information on river system topology, provide inaccurate simulations of extreme hydrological events, and simplify ecological water demand assessments, making it difficult to achieve a balance between global optimality and real-time performance.

Method used

An adaptive multimodal data fusion algorithm is used to process multi-source heterogeneous data. Hydrological features are extracted by combining deep convolutional long short-term memory networks and dynamic graph neural networks. Multi-scale wavelet decomposition and reconstruction techniques are applied, combined with nonlinear dynamic analysis, and dynamic Bayesian networks are used to simulate hydrological processes to generate high-resolution hydrological scenarios. A multi-scale coupled hydrodynamic model is constructed, and sluice gate scheduling is coordinated through deep reinforcement learning and game theory. The scheduling strategy is optimized by combining robust optimization and eco-hydrological demand assessment.

Benefits of technology

It has improved the accuracy and adaptability of sluice gate system scheduling, enhanced the ability to warn and respond to extreme hydrological events, achieved a multi-objective balance of flood control and disaster reduction, water resource utilization and ecological protection, and improved the level of automation and intelligence of scheduling.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005064A_ABST
    Figure CN119005064A_ABST
Patent Text Reader

Abstract

The invention discloses a regional sluice system scheduling optimization method and system, and the method comprises the following steps: carrying out the multi-source heterogeneous data fusion and feature extraction through the deep learning and graph neural network technology, and achieving the multi-scale analysis of hydrodynamic features; generating a high-resolution hydrological scene set by using a dynamic Bayesian network and a conditional generative adversarial network; a multi-scale hydrodynamic model is constructed, and the precision and timeliness of the model are improved in combination with an adaptive grid technology and a data assimilation method; and a deep reinforcement learning and game theory method is applied to optimize water gate group collaborative scheduling, and multi-target intelligent scheduling is realized. According to the method, the scientificity, efficiency and adaptability of water gate dispatching are remarkably improved, and the requirements of flood control and disaster reduction, water resource utilization and ecological protection are effectively balanced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to hydrological and hydrodynamic technology, in particular to a method and system for optimizing the scheduling of a regional sluice system. Background Art

[0002] The optimization of regional sluice system scheduling plays a vital role in water resources management, flood prevention and disaster reduction, and ecological environment protection. With the intensification of climate change and the acceleration of urbanization, problems such as uneven temporal and spatial distribution of water resources, frequent floods and waterlogging disasters, and degradation of water ecosystems have become increasingly prominent, which puts higher requirements on the scheduling of sluice systems. Optimized sluice scheduling can not only improve the efficiency of water resource utilization and reduce the losses caused by floods and waterlogging disasters, but also maintain the ecological water demand of rivers and protect aquatic habitats. In addition, accurate sluice scheduling is also of great significance for improving water quality, promoting water circulation, and maintaining a reasonable water level. Therefore, studying advanced regional sluice system scheduling optimization methods has great practical significance and long-term strategic value for achieving sustainable utilization of water resources and building ecological civilization.

[0003] At present, the research on regional sluice system scheduling optimization has made significant progress. Traditional methods are mainly based on empirical rules and simplified models, such as linear programming and dynamic programming. In recent years, with the development of artificial intelligence and big data technology, new scheduling optimization methods have emerged. Despite the significant progress made in research, regional sluice system scheduling optimization still faces many technical challenges. For example, when dealing with high-dimensional, nonlinear, and non-stationary hydrological time series data, existing methods are difficult to fully capture the complex dynamic characteristics. In the construction of multi-scale hydrodynamic models, the dynamic coupling mechanism between the one-dimensional river model and the two-dimensional floodplain model is not perfect enough, and it is difficult to accurately simulate the water flow exchange process under complex terrain conditions. In addition, the existing deep learning model is not effective in processing graph structure data, and it is difficult to fully utilize the water system topological structure information to improve the prediction accuracy. In addition, in the simulation and response of extreme hydrological events, traditional probabilistic statistical methods are difficult to accurately describe the occurrence and impact of new extreme events. In terms of scheduling optimization algorithms, how to improve the real-time performance of the algorithm while ensuring global optimality is still a challenge. Finally, existing ecological water demand assessment models are often oversimplified and it is difficult to accurately quantify the impact of different hydrological regimes on the habitats of specific species, thus affecting the formulation of eco-friendly scheduling plans. Solving these specific technical problems will greatly improve the scientificity and effectiveness of regional sluice system scheduling.

[0004] Hence the need for research and innovation. Summary of the invention

[0005] The purpose of the invention is to provide a method and system for optimizing the scheduling of a regional sluice system, in order to solve one of the above-mentioned problems.

[0006] Technical solution: A method for optimizing the scheduling of a regional sluice system, comprising the following steps: Step S1, acquiring multi-source heterogeneous data including remote sensing images, ground observation station data, meteorological radar data, and Internet of Things sensor data, and processing and standardizing them using an adaptive multimodal data fusion algorithm to generate a standardized data set aligned in time and space; using a deep convolutional long short-term memory network and a multimodal feature extraction method enhanced by an attention mechanism to extract spatiotemporal features from the standardized data set to form a preliminary feature data set; constructing a dynamic graph neural network based on the water system topological structure, further processing the preliminary feature data set to generate a dynamic feature data set containing the mutual influence of water system elements; applying multiscale wavelet decomposition and reconstruction technology, combined with a nonlinear dynamic analysis method, extracting multiscale hydrological dynamic features from the dynamic feature data set, and outputting a fused multi-dimensional, multi-scale hydrological dynamic feature data set; based on the self-organizing criticality theory, performing uncertainty analysis and critical state detection on the fused data set, and outputting a comprehensive data set with uncertainty estimation and system critical state indicators; Step S2, obtain a comprehensive data set with uncertainty estimates and system critical state indicators, apply a dynamic Bayesian network model to simulate the hydrological process, and generate a preliminary hydrological scenario set; use a conditional generative adversarial network to enhance and refine the preliminary scenario to generate a high-resolution hydrological scenario; use extreme value theory and coupled Markov chain methods to simulate extreme hydrological events and form an extended scenario set containing extreme situations; use dynamic pattern decomposition technology to extract the dominant mode of the hydrological system, and generate long-term hydrological trend forecasts based on these modes; integrate all generated scenarios, apply fuzzy comprehensive evaluation methods to sort and screen the scenarios, and output the final adaptive hydrological scenario set; Step S3, obtain an adaptive hydrological scenario set and a multi-dimensional hydrodynamic characteristic data set, and construct a multi-scale coupled hydrodynamic model including a one-dimensional river model, a two-dimensional floodplain model, and an urban drainage model; use adaptive grid refinement technology to optimize the model's spatial resolution and improve the simulation accuracy of key areas; use data assimilation methods to integrate real-time observation data and model prediction results to improve the timeliness and accuracy of the model; use a multi-objective optimization algorithm to automatically calibrate model parameters and balance the simulation effects of different scales and different evaluation indicators; evaluate the uncertainty of the model through scenario-driven Monte Carlo simulation, and output the calibrated multi-scale hydrodynamic model and its uncertainty estimate.

[0007] Step S4, obtain the calibrated multi-scale hydrodynamic model and its uncertainty estimation, as well as the adaptive hydrological scenario set, and construct a multi-objective sluice group collaborative scheduling optimization model; apply the deep reinforcement learning method to train the sluice group collaborative control strategy to achieve unified scheduling of multiple sluices; use the game theory method to coordinate the conflicts of interest between upstream and downstream sluices to form a balanced scheduling plan; use robust optimization technology to consider model uncertainty and scenario changes to generate a highly adaptable scheduling strategy; combine the results of eco-hydrological demand assessment to optimize the scheduling strategy to maintain the ecological health of the river; evaluate the long-term effect of the scheduling strategy through large-scale parallel simulation, and output the final sluice group collaborative optimization scheduling plan.

[0008] According to one aspect of the present application, a regional sluice system scheduling optimization system includes: at least one processor; and, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement a regional sluice system scheduling optimization method described in any of the above technical solutions.

[0009] Beneficial effects: significantly improve the accuracy of sluice system dispatching, so that the dispatching scheme can better adapt to complex and changeable hydrological conditions; improve the real-time and foresight of dispatching, enhance the early warning and response capabilities for extreme hydrological events; achieve a multi-objective balance of flood control and disaster reduction, water resources utilization and ecological protection, and promote the sustainable use of water resources; improve the automation and intelligence level of dispatching, reduce the burden of manual operation, and reduce the risk of human error. In general, this method provides strong technical support for the scientific management and efficient operation of regional sluice systems, and has important practical significance for improving the level of water resources management, enhancing flood control and disaster reduction capabilities, and protecting the water ecological environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flow chart of the present invention.

[0011] Figure 2 It is a flow chart of step S1 of the present invention.

[0012] Figure 3 It is a flow chart of step S2 of the present invention.

[0013] Figure 4 It is a flow chart of step S3 of the present invention.

[0014] Figure 5 It is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0015] like Figure 1As shown, a method for optimizing regional sluice system scheduling is provided, comprising the following steps: Step S1, acquiring multi-source heterogeneous data including remote sensing images, ground observation station data, meteorological radar data, and Internet of Things sensor data, and processing and standardizing them using an adaptive multimodal data fusion algorithm to generate a standardized data set aligned in time and space; using a deep convolutional long short-term memory network and a multimodal feature extraction method enhanced by an attention mechanism to extract spatiotemporal features from the standardized data set to form a preliminary feature data set; constructing a dynamic graph neural network based on the water system topological structure, further processing the preliminary feature data set to generate a dynamic feature data set containing the mutual influence of water system elements; applying multiscale wavelet decomposition and reconstruction technology, combined with a nonlinear dynamic analysis method, extracting multiscale hydrological dynamic features from the dynamic feature data set, and outputting a fused multi-dimensional, multi-scale hydrological dynamic feature data set; based on the self-organizing criticality theory, performing uncertainty analysis and critical state detection on the fused data set, and outputting a comprehensive data set with uncertainty estimation and system critical state indicators; Step S2, obtain a comprehensive data set with uncertainty estimates and system critical state indicators, apply a dynamic Bayesian network model to simulate the hydrological process, and generate a preliminary hydrological scenario set; use a conditional generative adversarial network to enhance and refine the preliminary scenario to generate a high-resolution hydrological scenario; use extreme value theory and coupled Markov chain methods to simulate extreme hydrological events and form an extended scenario set containing extreme situations; use dynamic pattern decomposition technology to extract the dominant mode of the hydrological system, and generate long-term hydrological trend forecasts based on these modes; integrate all generated scenarios, apply fuzzy comprehensive evaluation methods to sort and screen the scenarios, and output the final adaptive hydrological scenario set; Step S3, obtain an adaptive hydrological scenario set and a multi-dimensional hydrodynamic characteristic data set, and construct a multi-scale coupled hydrodynamic model including a one-dimensional river model, a two-dimensional floodplain model, and an urban drainage model; use adaptive grid refinement technology to optimize the model's spatial resolution and improve the simulation accuracy of key areas; use data assimilation methods to integrate real-time observation data and model prediction results to improve the timeliness and accuracy of the model; use a multi-objective optimization algorithm to automatically calibrate model parameters and balance the simulation effects of different scales and different evaluation indicators; evaluate the uncertainty of the model through scenario-driven Monte Carlo simulation, and output the calibrated multi-scale hydrodynamic model and its uncertainty estimate.

[0016] Step S4, obtain the calibrated multi-scale hydrodynamic model and its uncertainty estimation, as well as the adaptive hydrological scenario set, and construct a multi-objective sluice group collaborative scheduling optimization model; apply the deep reinforcement learning method to train the sluice group collaborative control strategy to achieve unified scheduling of multiple sluices; use the game theory method to coordinate the conflicts of interest between upstream and downstream sluices to form a balanced scheduling plan; use robust optimization technology to consider model uncertainty and scenario changes to generate a highly adaptable scheduling strategy; combine the results of eco-hydrological demand assessment to optimize the scheduling strategy to maintain the ecological health of the river; evaluate the long-term effect of the scheduling strategy through large-scale parallel simulation, and output the final sluice group collaborative optimization scheduling plan.

[0017] In this embodiment, firstly, the inconsistency problem of remote sensing images, ground observation station data, meteorological radar data and IoT sensor data in spatiotemporal scales is solved through an adaptive multimodal data fusion algorithm, and data quality and availability are improved. The application of deep convolutional long short-term memory network effectively captures the long-term dependence and short-term fluctuation characteristics of hydrological time series, and enhances the model's ability to express complex hydrological processes. The construction of dynamic graph neural network makes full use of the topological structure information of the water system and improves the understanding of the dynamic changes of the overall water system. The combination of multiscale wavelet decomposition and nonlinear dynamic analysis method realizes the accurate characterization of the multiscale characteristics and nonlinear behavior of the hydrological system, and provides theoretical support for predicting extreme events. Secondly, the application of dynamic Bayesian network model improves the accuracy and interpretability of hydrological process simulation. The introduction of conditional generative adversarial network breaks through the limitations of traditional hydrological scenario generation methods and can generate high-resolution and more realistic hydrological scenarios. The combination of extreme value theory and coupled Markov chain method significantly improves the model's simulation ability for extreme hydrological events. The application of dynamic mode decomposition technology enables the model to extract dominant dynamic patterns from complex hydrological time series, which helps to identify long-term hydrological trends. Secondly, a high-precision, multi-scale hydrodynamic model was constructed. The construction of a multi-scale coupled hydrodynamic model realizes the unified simulation of one-dimensional river channels, two-dimensional floodplains and urban drainage systems. The application of adaptive grid refinement technology significantly improves the simulation accuracy of key areas. The introduction of data assimilation methods, especially the combination of ensemble Kalman filtering and variational assimilation methods, realizes the dynamic fusion of models and real-time observation data, and improves the timeliness and accuracy of the model. The application of multi-objective optimization algorithms realizes the automatic calibration of model parameters and balances the simulation effects under different scales and evaluation indicators. Finally, the construction of a multi-objective sluice group collaborative scheduling optimization model realizes the unified optimization of multiple objectives such as flood control and disaster reduction, water resources utilization and ecological water demand. The application of deep reinforcement learning methods enables the model to adaptively learn complex hydrological environments and generate flexible and efficient scheduling strategies. The introduction of game theory methods effectively coordinates the conflicts of interest between upstream and downstream sluices. The application of robust optimization technology significantly enhances the adaptability of the scheduling strategy to uncertainties and extreme situations. The integration of eco-hydrological demand assessment ensures that the scheduling scheme can maintain the ecological health of the river.

[0018] Through the organic combination of these four steps, we have achieved: deep fusion and feature extraction of multi-source heterogeneous data, which has improved the perception and understanding of complex hydrological systems and provided high-quality data support for subsequent analysis. The hydrological scenario generation method based on deep learning has broken through the limitations of traditional methods, and can generate more realistic and diverse hydrological scenarios, providing a more comprehensive decision-making basis for scheduling decisions. The construction of multi-scale coupled hydrodynamic models and the application of data assimilation technology have significantly improved the accuracy and timeliness of hydrodynamic simulation, providing a reliable simulation platform for scheduling optimization. Finally, the coordinated scheduling optimization method of sluice groups based on deep reinforcement learning and game theory has realized multi-objective and multi-scale intelligent scheduling, and improved the scientificity, adaptability and eco-friendliness of the scheduling scheme.

[0019] According to one aspect of the present application, step S1 specifically comprises: Step S11, acquiring multi-source heterogeneous data, including remote sensing images, ground observation station data, meteorological radar data and Internet of Things sensor data, applying an adaptive data cleaning algorithm to remove noise and detect outliers; using a dynamic time warping algorithm to achieve time alignment of data with different sampling frequencies; using spatial interpolation technology to perform spatial alignment of data to generate a preliminary data set that is consistent in time and space; using a Z-score normalization method to normalize the preliminary data set, and output a standardized multi-source data set; Step S12, receiving the standardized multi-source data set output by S11, constructing and training a deep convolutional long short-term memory network; the convolution layer is used to extract spatial features, and the LSTM layer captures time series features; batch normalization and residual connection are used to optimize network performance; the cross-validation method is used to select the optimal model parameters, and the trained DC-LSTM model and the initially extracted spatiotemporal feature data set are output; Step S13, obtaining the preliminary spatiotemporal feature data set output by S12, calling the multi-head self-attention mechanism; calculating the attention weights of the features of different data sources; adaptively fusing the features based on the attention weights; applying the feature importance analysis method to screen the key features; and outputting the enhanced multimodal feature data set; Step S14, receiving the enhanced multimodal feature dataset output by S13, combining it with the pre-stored water system topology data; constructing a dynamic graph reflecting the water system structure; constructing a graph convolution layer and a gated recurrent unit to capture the dynamic interaction between nodes; applying a graph attention mechanism to enhance the influence of key nodes; training a DGNN model, and outputting a graph embedding representation containing dynamic features of the water system; Step S15, obtain the graph embedding representation output by S14, select the data-rich region as the source domain; use the adversarial training method to reduce the feature distribution difference between the source domain and the target domain; adopt the progressive migration strategy to gradually adapt to the target domain; apply the active learning method to select the most informative samples for annotation; fine-tune the model parameters and output the general feature representation adapted to different regions; Step S16, receiving the general feature representation output by S15, applying discrete wavelet transform to perform multi-scale decomposition; performing threshold processing on wavelet coefficients of different scales to remove noise; using wavelet packet analysis method to extract more refined frequency features; constructing an adaptive weighting scheme to weight different scales based on feature importance; reconstructing the signal by inverse wavelet transform, and outputting a multi-scale feature data set; Step S17, obtaining the multi-scale feature data set output by S16, applying Takens embedding theorem to reconstruct the phase space; calculating the correlation dimension and information dimension to characterize the complexity of the system; estimating the maximum Lyapunov exponent to quantify the degree of chaos of the system; using the recursive quantitative analysis method to extract the periodicity and deterministic characteristics of the system; constructing a nonlinear prediction model to evaluate the predictability of the system; outputting an enhanced data set containing nonlinear dynamic characteristics; Step S18, receiving the enhanced data set output by S17, applying the power law distribution fitting method to detect the self-organized critical state of the system; calculating the Fisher information to quantify the order change of the system; using the permutation entropy method to evaluate the complexity and uncertainty of the system; using the ensemble learning method to integrate the results of multiple detectors to improve the reliability of detection; applying Monte Carlo simulation to evaluate the uncertainty of feature extraction and state detection; outputting the final comprehensive data set with uncertainty estimation and system critical state indicators.

[0020] In this embodiment, comprehensive perception and accurate characterization of complex hydrological systems are achieved through the fusion of multi-source heterogeneous data and deep feature extraction. First, an adaptive multimodal data fusion algorithm is used to process remote sensing images, ground observation station data, meteorological radar data, and IoT sensor data, which solves the problem of spatiotemporal alignment of data from different sources and scales, and improves data quality and consistency. The application of deep convolutional long short-term memory network effectively captures the long-term dependence and short-term fluctuation characteristics of hydrological time series, and enhances the model's ability to express complex hydrological processes. The introduction of multi-head self-attention mechanism enables the model to adaptively focus on key information in different data sources, improving the pertinence and effectiveness of feature extraction. The dynamic graph neural network constructed based on the topological structure of the water system makes full use of the spatial correlation between water system elements and enhances the model's ability to understand the dynamic changes of the overall water system. The application of multi-scale wavelet decomposition and reconstruction technology realizes the separation and analysis of different frequency components of hydrological signals, which helps to capture multi-scale hydrological dynamic characteristics. The introduction of nonlinear dynamic analysis methods enables the model to identify chaotic behavior and critical states in hydrological systems, providing a theoretical basis for predicting extreme events. The application of the theory of self-organized criticality enhances the model's ability to warn of sudden changes in the hydrological system, providing an important reference for sluice scheduling decisions. Through the integration of this series of innovative technologies, this step not only improves the accuracy and efficiency of data processing, but also enhances the model's ability to understand and predict complex hydrological systems, laying a solid data foundation for subsequent hydrological scenario generation and scheduling optimization.

[0021] According to one aspect of the present application, step S2 further comprises: Step S21, obtaining the comprehensive data set outputted from step S1, and combining it with the pre-stored hydrological model parameter library; constructing a dynamic Bayesian network structure reflecting the interrelationships between hydrological elements; using the expectation maximization (EM) algorithm to learn network parameters; applying the particle filtering method to perform online inference of sequence data; generating multiple groups of hydrological processes through Monte Carlo simulation, and outputting a preliminary hydrological scenario set; Step S22, receiving the preliminary hydrological scenario set output by S21, constructing and training the CGAN model; the generator network uses the U-Net architecture, and the discriminator adopts the Patch GAN structure; calling the Wasserstein distance as the loss function to improve the training stability; using the spectral normalization technology to avoid the gradient vanishing problem; through iterative optimization of the preliminary scenario, generating a high-resolution and more realistic hydrological scenario, and outputting an enhanced hydrological scenario set; Step S23, obtaining the enhanced hydrological scenario set output by S22, applying the generalized extreme value distribution (GEV) to fit extreme hydrological events; using the time-varying copula function to simulate the time dependence of extreme events; using the piecewise Markov chain method to simulate the duration and intensity of extreme events; combining the climate change prediction data to adjust the probability of occurrence of extreme events; generating extended scenarios containing various extreme conditions, and outputting the extended scenario set containing extreme hydrological events; Step S24, receiving the extended scenario set outputted by S23, applying the dynamic mode decomposition (DMD) technology to extract the dominant dynamic mode of the hydrological system; using the sparse recognition algorithm to optimize the DMD model to improve the stability of long-term prediction; combining with wavelet transform to capture multi-scale periodic changes; applying the Mann-Kendall test to identify significant long-term trends; generating long-term hydrological predictions based on the extracted dominant modes and trends, and outputting the long-term hydrological trend prediction results; Step S25, obtain the output results of S22, S23 and S24, and construct a multi-criteria evaluation system, including the rationality, diversity and representativeness of the scenarios; apply the fuzzy comprehensive evaluation method to calculate the comprehensive score of each scenario; use the principal component analysis method to reduce the redundancy of the scenarios; use the K-means++ algorithm to group the scenarios; based on the system critical state index, give priority to scenarios close to the critical state; use the Monte Carlo cross-validation method to evaluate the robustness of the scenarios; and finally output the evaluated and screened adaptive hydrological scenario set.

[0022] In this embodiment, high-quality and diversified hydrological scenario generation is achieved, providing comprehensive decision support for sluice scheduling optimization. The application of the dynamic Bayesian network model enables the hydrological process simulation to consider the complex dependencies and temporal dynamics between variables, improving the accuracy and interpretability of the simulation. The introduction of the conditional generative adversarial network (CGAN) breaks through the limitations of traditional hydrological scenario generation methods, can generate high-resolution and more realistic hydrological scenarios, and enhances the model's predictability of possible future hydrological conditions. The combination of extreme value theory and coupled Markov chain method significantly improves the model's simulation ability for extreme hydrological events, providing a more reliable decision-making basis for flood control and disaster reduction. The application of dynamic mode decomposition (DMD) technology enables the model to extract dominant dynamic patterns from complex hydrological time series, which helps to identify long-term hydrological trends and provide a scientific basis for long-term water resources planning. The use of fuzzy comprehensive evaluation methods realizes multi-dimensional evaluation and screening of generated scenarios, ensuring the representativeness and reliability of the final scenario set. Through the comprehensive application of these technologies, this step not only improves the accuracy and diversity of hydrological scenario generation, but also enhances the model's ability to predict extreme events and long-term trends. This provides a more comprehensive and reliable decision-making basis for the optimization of sluice scheduling, helps to formulate more scientific and reasonable scheduling plans, and improves the efficiency of water resources management and risk response capabilities.

[0023] According to one aspect of the present application, step S3 further comprises: Step S31, obtaining the hydrodynamic characteristic data set outputted in step S1 and the pre-stored terrain data, constructing a one-dimensional river channel Saint-Venant equation model; applying the finite volume method to discretize the two-dimensional shallow water equation and establish a floodplain model; using a hyperbolic model to describe the water flow in the urban pipe network and constructing an urban drainage model; constructing a dynamic coupling interface between models to realize water volume exchange; using OpenMP parallel computing technology to optimize the model operation efficiency, and outputting a preliminary multi-scale hydrodynamic model framework; Step S32, receiving the preliminary model framework outputted by S31, applying the quadtree grid division method to realize adaptive grid refinement; dynamically adjusting the grid resolution based on the water flow gradient and terrain characteristics; using the boundary fitting technology to improve the simulation accuracy of complex terrain; calling the sub-grid parameterization method to consider the influence of small-scale terrain characteristics on water flow; using GPU acceleration technology to improve the computational efficiency of the refined grid, and outputting the optimized high-resolution hydrodynamic model; Step S33, obtain the optimization model output by S32 and the real-time observation data of step S1, apply the ensemble Kalman filter method to perform data assimilation; use the particle filter algorithm to process the observation data with strong nonlinearity; use the variational assimilation method to optimize the initial conditions and boundary conditions of the model; call the multi-scale data assimilation strategy to coordinate the influence of data of different scales; construct a sliding time window assimilation scheme to achieve continuous correction of the model; output a dynamic hydrodynamic model with real-time correction capability; Step S34, receiving the dynamic model output by S33 and part of the hydrological scenario of step S2, constructing a multi-objective evaluation system, including water level, flow and inundation range; applying the NSGA-III algorithm to perform multi-objective optimization and find the Pareto optimal solution set; calling the ε-constraint method to deal with the trade-off relationship between objectives; using the alternating direction multiplier method to accelerate the optimization process; using the Latin hypercube sampling method to generate an initial parameter set to improve the optimization efficiency; evaluating the stability of the parameters through cross-validation, and outputting multiple sets of Pareto optimal parameter sets; Step S35, obtain the Pareto optimal parameter set output by S34 and the complete hydrological scenario set of step S2, apply the GLUE method to evaluate parameter uncertainty; use the Sobol sensitivity analysis method to identify key parameters; use the Bootstrap method to evaluate the model structure uncertainty; call fuzzy mathematics theory to quantify the uncertainty of boundary conditions; through scenario-driven Monte Carlo simulation, comprehensively evaluate the performance of the model under different conditions; finally output the calibrated multi-scale hydrodynamic model and its comprehensive uncertainty estimate.

[0024] In this embodiment, a high-precision, multi-scale hydrodynamic model is constructed, providing a reliable simulation platform for sluice scheduling optimization. The construction of a multi-scale coupled hydrodynamic model realizes the unified simulation of one-dimensional river channels, two-dimensional floodplains and urban drainage systems, and improves the simulation accuracy of complex water systems. The application of adaptive grid refinement technology solves the problem that traditional fixed grids are difficult to accurately depict local complex terrain, significantly improves the simulation accuracy of key areas, and maintains computational efficiency. The introduction of data assimilation methods, especially the combination of ensemble Kalman filtering and variational assimilation methods, realizes the dynamic fusion of models and real-time observation data, and improves the timeliness and accuracy of the model. The application of the multi-objective optimization algorithm NSGA-III realizes the automatic calibration of model parameters, balances the simulation effects under different scales and evaluation indicators, and improves the overall performance and applicability of the model. The introduction of Monte Carlo simulation enables the model to comprehensively evaluate multi-source uncertainties such as parameters, structures and boundary conditions, and provides a reliable risk assessment basis for scheduling decisions. Through the comprehensive application of these technologies, this step not only improves the accuracy and reliability of the hydrodynamic model, but also enhances the adaptability and robustness of the model. This provides a more accurate and comprehensive simulation environment for sluice scheduling optimization, helps to evaluate the effects of different scheduling schemes, predict possible hydrodynamic responses, and thus formulate more scientific and reasonable scheduling strategies to improve the operating efficiency and safety of the sluice system.

[0025] According to one aspect of the present application, step S4 is further: Step S41, obtaining the hydrodynamic model of step S3 and the pre-stored sluice characteristic data, establishing a state transfer equation reflecting the relationship between sluice operation and hydrodynamic response; constructing a multi-objective evaluation function, including flood prevention and disaster reduction, water resource utilization and ecological water demand; calling constraints, including sluice scheduling rules and safe operation restrictions; using a multi-agent system framework to describe the interaction of sluice groups; using a dynamic programming method to construct a basic solution algorithm; outputting a preliminary multi-objective sluice group scheduling optimization model; Step S42, receiving the scheduling optimization model output by S41, constructing a deep reinforcement learning framework based on the actor-critic architecture; using the long short-term memory network (LSTM) as the state encoder to capture the hydrological time series characteristics; using the attention mechanism to enhance the information interaction between the sluices; calling the priority experience replay mechanism to improve the learning efficiency; applying the double Q learning algorithm to alleviate the over-estimation problem; gradually increasing the task difficulty through the course learning strategy; using the distributed training technology to accelerate the learning process; outputting the preliminary trained sluice group collaborative control strategy; Step S43, obtain the preliminary control strategy output by S42, construct a non-cooperative game model that reflects the interest relationship between the upstream and downstream sluice gates; apply the Nash equilibrium theory to find the balance point of the interests of all parties; call the Shapley value method to fairly distribute the overall benefits brought by the dispatch; use the evolutionary game theory to simulate the evolution of the long-term dispatch strategy; use the multi-level bargaining model to handle the complex interest negotiation process; explore the global optimal solution through the cooperative game method; output a coordinated dispatch plan that balances the interests of all parties; Step S44, receiving the coordinated scheduling scheme output by S43 and the model uncertainty estimation of step S3, applying the scenario tree method to construct the uncertainty set; using the conditional value at risk (CVaR) measurement method to evaluate the risk in extreme cases; using the robust adaptive optimization algorithm to generate a scheduling strategy with strong adaptability to uncertainty; calling dynamic decision rules to improve the real-time adjustment capability of the strategy; using the rolling time domain optimization method to regularly update the long-term scheduling plan; evaluating the robustness of the strategy through multi-scenario response capability testing; outputting a robust scheduling strategy with strong adaptability; Step S45, obtain the robust scheduling strategy output by S44 and the ecosystem data of step S1, establish a river ecological water demand assessment model; apply the ecohydraulics method to quantify the impact of hydrological conditions on the habitats of key species; call on the environmental flow concept to determine the minimum flow requirement to maintain the health of the ecosystem; use HSI to evaluate the ecological effects of different scheduling schemes; adopt a multi-objective trade-off analysis method to balance flood control, water supply and ecological needs; optimize the scheduling strategy through ecological response simulation to improve the health of the ecosystem; output an optimized scheduling strategy that comprehensively considers ecological needs; Step S46, receive the optimized scheduling strategy output by S45 and the complete hydrological scenario set of step S2, build a long-term simulation experiment plan, covering a variety of hydrological and climate scenarios; use high-performance computing clusters to achieve large-scale parallel hydrodynamic simulation; apply the Monte Carlo method to evaluate the performance of the strategy under different scenarios; call extreme event response evaluation to test the emergency handling capability of the strategy; use long-term trend analysis methods to evaluate the adaptability of the strategy to climate change; identify potential risk points of the scheduling strategy through system vulnerability analysis; and comprehensively analyze the evaluation results to output the final sluice group collaborative optimization scheduling plan and its long-term effect evaluation report.

[0026] In this embodiment, the intelligent coordinated dispatching of the sluice group is realized, and the scientificity, adaptability and eco-friendliness of the dispatching scheme are improved. The construction of the multi-objective sluice group coordinated dispatching optimization model realizes the unified optimization of multiple objectives such as flood control and disaster reduction, water resource utilization and ecological water demand, and improves the comprehensive benefits of the dispatching scheme. The application of deep reinforcement learning methods, especially the framework based on the actor-critic architecture, enables the model to adaptively learn complex hydrological environments and generate flexible and efficient dispatching strategies. The introduction of game theory methods effectively coordinates the conflicts of interest between upstream and downstream sluices and realizes the global optimal dispatching scheme. The application of robust optimization technology significantly enhances the adaptability of the dispatching strategy to uncertainty and extreme situations, and improves the reliability and stability of the scheme. The integration of eco-hydrological demand assessment ensures that the dispatching scheme can maintain the ecological health of the river channel and achieve a balance between water resource utilization and ecological protection. The implementation of large-scale parallel simulation enables the model to comprehensively evaluate the long-term effects of the dispatching strategy and provides a reliable basis for decision-making. Through the comprehensive application of these technologies, this step not only improves the intelligence level and coordinated efficiency of sluice dispatching, but also enhances the adaptability and eco-friendliness of the dispatching scheme. This provides an advanced decision support tool for the management of regional sluice systems, helps achieve efficient use of water resources, improve flood prevention and disaster reduction capabilities, and protect the water ecological environment, thereby promoting the sustainable development of water resources.

[0027] According to one aspect of the present application, step S16 is specifically: Step S161, obtaining the general feature representation data output by S15, applying the information entropy criterion to evaluate the performance of different wavelet basis functions; using a genetic algorithm to optimize wavelet basis parameters, including support length and vanishing moment; selecting the most suitable wavelet basis function by a cross-validation method; outputting the preferred wavelet basis function and its parameters; Step S162, receiving the preferred wavelet basis function outputted by S161, performing discrete wavelet transform on the general feature representation data; using an adaptive scale selection algorithm, dynamically determining the number of decomposition layers according to the local characteristics of the data; using the Best Basis algorithm to select the optimal basis in the wavelet packet; calculating the wavelet coefficients of each scale, and outputting a multi-scale wavelet coefficient set; Step S163, obtaining the multi-scale wavelet coefficient set outputted by S162, applying the SURE criterion to automatically select the optimal threshold; using the soft threshold method to denoise the wavelet coefficients; using an adaptive threshold strategy, using different thresholds for different scales and different features; outputting the denoised multi-scale wavelet coefficient set; Step S164, receiving the denoised wavelet coefficient set output by S163, calculating the energy distribution and statistical characteristics of each scale; applying principal component analysis to reduce feature dimensions; using random forest algorithm to evaluate feature importance; selecting the most representative feature subset based on mutual information criterion; outputting a multi-scale feature set and its importance weight; Step S165, obtain the multi-scale feature set and importance weights output by S164, and construct a weighted reconstruction strategy based on feature importance; use inverse discrete wavelet transform to reconstruct the signal; use an adaptive fusion algorithm to integrate reconstruction results of different scales; apply an integrated learning method to improve the stability and generalization ability of the reconstruction results; output the final multi-scale feature data set.

[0028] In this embodiment, multi-scale decomposition and reconstruction of hydrological features are achieved, and the accuracy and effectiveness of feature extraction are improved. First, the wavelet basis function is optimized by information entropy criterion and genetic algorithm, which solves the problem that traditional fixed wavelet basis is difficult to adapt to different hydrological features, and improves the adaptability and accuracy of wavelet transform. The application of adaptive scale selection algorithm enables the model to dynamically determine the number of decomposition layers according to the local characteristics of the data, and enhances the ability to capture hydrological features of different scales. The Best Basis algorithm selects the optimal basis in the wavelet packet, further improving the efficiency and accuracy of feature extraction. The combination of the SURE criterion automatically selects the optimal threshold and the soft threshold denoising method effectively removes the noise in the hydrological signal and improves the reliability of subsequent analysis. The introduction of the adaptive threshold strategy enables the model to use different thresholds for different scales and different features, which enhances the refinement and pertinence of the denoising process. The application of principal component analysis and random forest algorithm realizes the effective reduction of feature dimensions and importance evaluation, and improves the scientificity and efficiency of feature selection. The weighted reconstruction strategy based on feature importance ensures that important features are fully expressed in the reconstruction process, and improves the representativeness and reliability of the reconstruction results. The use of adaptive fusion algorithms and integrated learning methods enhances the stability and generalization of the reconstruction results. Through the comprehensive application of this series of technologies, this step not only improves the accuracy and efficiency of hydrological feature extraction, but also enhances the model's ability to express multi-scale hydrological dynamics characteristics, providing more reliable and rich feature information for subsequent hydrological scenario generation and scheduling optimization, thereby improving the prediction and decision-making capabilities of the entire system.

[0029] According to one aspect of the present application, step S22 is specifically: Step S221, obtain the preliminary hydrological scenario set output by S21, build a U-Net-based generator network architecture; build a Patch GAN discriminator network; call the conditional embedding layer and use the hydrological scenario features as conditional input; build a multi-scale feature extraction module to capture hydrological features at different spatial scales; output the initial architecture of the CGAN model; Step S222, receiving the CGAN initial architecture outputted by S221, constructing a composite loss function, including adversarial loss, content loss and feature matching loss; calling Wasserstein distance as a measure of adversarial loss; using a gradient penalty strategy to improve training stability; constructing a progressive training scheme to gradually improve the resolution of the generated image; and outputting the optimized loss function and training strategy; Step S223, obtain the loss function and training strategy output by S222, use the Adam optimizer to train the model; apply the learning rate decay strategy; call the spectral normalization technology to alleviate the gradient vanishing problem; use the batch normalization method to accelerate the training process; determine the optimal hyperparameter combination through cross-validation; output the trained CGAN model; Step S224, receiving the trained CGAN model output by S223 and the preliminary hydrological scenario set of S21, using the sliding window technology to process the large-scale hydrological scene; applying the conditional control strategy to generate corresponding high-resolution scenarios according to different hydrological conditions; using the integration method to fuse multiple generated results to improve stability; using the post-processing algorithm to ensure the physical consistency of the generated results; outputting the high-resolution hydrological scenario set; Step S225, obtain the high-resolution hydrological scenario set output by S224, construct multidimensional evaluation indicators, including authenticity, diversity and conditional consistency; use FID to evaluate the quality of the generated results; apply MMD to measure the difference between the generated distribution and the true distribution; use the expert scoring system to comprehensively evaluate the rationality of the scenario; based on the evaluation results, screen out the optimal enhanced hydrological scenario set; output the final high-quality, high-resolution hydrological scenario set.

[0030] In this embodiment, the generation of high-quality and high-resolution hydrological scenarios is achieved, which significantly improves the accuracy and diversity of hydrological scenario prediction. The generator network design of the U-Net architecture enables the model to effectively capture the multi-scale characteristics of the hydrological scene, improves the detail performance and overall consistency of the generated results. The application of the Patch GAN discriminator enhances the model's ability to identify local details and improves the authenticity of the generated hydrological scenarios. The introduction of the conditional embedding layer enables the model to generate corresponding scenarios according to different hydrological conditions, enhancing the pertinence and controllability of the generated results. The design of the composite loss function, especially the introduction of the Wasserstein distance, effectively solves the problem of instability in traditional GAN ​​training and improves the convergence and generation quality of the model. The application of the gradient penalty strategy and spectral normalization technology further enhances the training stability of the model and avoids problems such as mode collapse. The implementation of the progressive training scheme enables the model to gradually improve the resolution of the generated image, effectively balancing the computational efficiency and generation quality. The application of the sliding window technology solves the problem of processing large-scale hydrological scenes, enabling the model to generate high-resolution hydrological scenarios covering a wide area. The introduction of conditional control strategies enhances the model's ability to generate corresponding scenarios according to different hydrological conditions, and improves the diversity and adaptability of the generated results. The use of integrated methods and post-processing algorithms ensures the stability and physical consistency of the generated results. The construction of multidimensional evaluation indicators, especially the application of FID and MMD, provides an objective and comprehensive standard for the quality evaluation of the generated results. Through the comprehensive application of these technologies, this step not only improves the quality and resolution of hydrological scenario generation, but also enhances the model's ability to generate diversified and realistic hydrological scenarios, providing a richer and more reliable decision-making basis for sluice scheduling optimization, and helping to improve the scientificity and foresight of water resources management.

[0031] According to one aspect of the present application, step S32 is further: Step S321, obtaining the preliminary model framework outputted from S31 and the pre-stored terrain data, applying the Delaunay triangulation method to generate an initial unstructured grid; calculating grid quality indicators, including grid distortion and aspect ratio; identifying key hydrological and geomorphological features, including river channels, levees, and depressions; outputting the initial grid and its quality assessment report; Step S322, receiving the initial grid outputted by S321, constructing refinement criteria based on water flow gradient, terrain relief and hydrodynamic characteristics; calling the curvature adaptive method to increase the grid density in the high curvature area; using the error estimation technology to identify the high error area that needs to be refined; constructing a dynamic threshold strategy to automatically adjust the refinement criteria according to the simulation process; outputting the adaptive refinement criteria and threshold set; Step S323, obtaining the refinement criteria output by S322 and the initial grid of S321, implementing a quadtree or octree grid refinement algorithm; applying a local refinement strategy to maintain a gradual change in the size of adjacent cells; implementing dynamic load balancing to optimize parallel computing efficiency; performing a recursive refinement process until all refinement criteria are met; and outputting a multi-resolution adaptive grid; Step S324, receiving the multi-resolution grid output from S323, using isogeometric analysis method to accurately fit complex boundaries; using moving grid technology to deal with time-varying boundary problems; applying Laplace smoothing algorithm to improve grid quality; implementing local reconstruction technology to eliminate degenerate units; balancing grid orthogonality and smoothness through iterative optimization process; outputting high-quality boundary fitting adaptive grid; Step S325, obtain the high-quality grid output by S324, call the sub-grid terrain parameterization method to capture the small-scale terrain characteristics; construct the concept of effective hydraulic radius to characterize the water flow resistance at the sub-grid scale; establish a porous medium model to simulate the flow characteristics at the sub-grid scale; integrate the sub-grid information into the computational grid through the scale conversion function; output the optimized hydrodynamic model grid containing the sub-grid parameterization.

[0032] In this embodiment, high-precision, adaptive grid generation of the hydrodynamic model is achieved, which significantly improves the simulation accuracy and computational efficiency of the model. The application of the Delaunay triangulation method solves the problem of initial grid generation under complex terrain and provides a good foundation for subsequent refinement. The calculation of grid quality indicators and the identification of key hydrological and geomorphological features provide a clear direction and basis for grid optimization. The refinement criteria based on water flow gradient, terrain undulation and hydrodynamic characteristics enable the model to specifically improve the grid density in key areas, while ensuring accuracy and optimizing computing resources. The combination of curvature adaptive method and error estimation technology further improves the accuracy and efficiency of grid refinement. The introduction of dynamic threshold strategy enables the model to automatically adjust the refinement standard according to the simulation process, enhancing the adaptability of grid generation. The implementation of quadtree or octree grid refinement algorithm ensures the gradual change of adjacent unit sizes and improves grid quality. The application of dynamic load balancing strategy optimizes parallel computing efficiency and speeds up large-scale simulation. The introduction of geometric analysis methods such as , achieves accurate fitting of complex boundaries and improves the simulation accuracy at the boundaries. The application of moving grid technology solves the time-varying boundary problem and enhances the model's adaptability to dynamic hydrological processes. The use of Laplace smoothing algorithm and local reconstruction technology further improves the grid quality, eliminates degenerate units, and improves the stability of the simulation. The call of the sub-grid terrain parameterization method enables the model to capture the impact of small-scale terrain features on water flow and improves the refinement of the simulation. The introduction of the effective hydraulic radius concept and porous media model realizes the simulation of sub-grid scale flow characteristics and enhances the model's ability to express complex hydrological processes. Through the comprehensive application of these technologies, this step not only improves the spatial resolution and simulation accuracy of the hydrodynamic model, but also enhances the model's adaptability and computational efficiency, providing a more accurate and efficient simulation platform for sluice scheduling optimization, which helps to improve the scientificity and reliability of scheduling decisions.

[0033] According to one aspect of the present application, step S33 is specifically: Step S331, obtain the real-time observation data and pre-stored historical data of step S1, apply the outlier detection algorithm to identify and mark suspicious data; use the wavelet analysis method to remove high-frequency noise; use the multivariate regression technology to interpolate missing data; implement consistency checks to ensure the temporal and spatial consistency of data from different sources; output the observation data set that has undergone quality control; Step S332, receiving the post-quality control observation data output by S331, generating initial set members to reflect the uncertainty of the model state; constructing a covariance localization scheme to reduce the impact of false correlation; implementing covariance inflation technology to avoid filter divergence problems; calling an adaptive inflation factor algorithm to dynamically adjust the set discreteness; outputting a well-configured EnKF assimilation framework; Step S333, obtain the EnKF framework output by S332 and the observation data of S331, construct an importance sampling strategy to improve the representativeness of particles; implement a resampling algorithm to alleviate the problem of particle degradation; use the Markov chain Monte Carlo method to enhance particle diversity; call smoothing technology to improve the processing capability of non-Gaussian distribution; output the optimized particle filter assimilation scheme; Step S334, receiving the observation data of S331 and the optimized hydrodynamic model output by S32, constructing an adjoint model, and calculating the gradient of the cost function; implementing the incremental 4D-Var method to improve the computational efficiency; adopting a multi-scale variational assimilation strategy to process observations of different temporal and spatial resolutions; calling the observation operator to map the model variables to the observation space; minimizing the cost function through an iterative optimization process; and outputting the 4D-Var assimilation system and its optimal initial conditions and boundary conditions; Step S335, obtain various assimilation schemes output by S332, S333 and S334, build a multi-scale data assimilation framework, coordinate the influence of data of different scales; implement the scale conversion algorithm to achieve seamless integration of data of different resolutions; call the dynamic weight allocation mechanism to adjust the weights of different assimilation methods according to data quality and model performance; evaluate the overall performance of the multi-scale assimilation strategy through integrated testing; output a comprehensive multi-method, multi-scale data assimilation system.

[0034] In this embodiment, high-precision, multi-source data assimilation is achieved, which significantly improves the prediction accuracy and timeliness of the hydrodynamic model. First, the application of outlier detection algorithm and wavelet analysis method effectively identifies and removes outliers and high-frequency noise in the observed data, and improves the quality of input data. The use of multivariate regression technology solves the problem of missing data interpolation and ensures the integrity and continuity of the data. The construction of the ensemble Kalman filter (EnKF) framework, especially the introduction of covariance localization and adaptive inflation factor algorithm, effectively solves the filter divergence and false correlation problems faced by traditional EnKF when dealing with high-dimensional nonlinear systems, and improves the stability and accuracy of the assimilation effect. The combination of importance sampling strategy and resampling algorithm in the particle filter assimilation scheme effectively alleviates the particle degradation problem and enhances the processing ability of non-Gaussian distribution. The application of Markov chain Monte Carlo method further improves the diversity of particles and enhances the model's ability to express complex hydrological processes. The implementation of the four-dimensional variational assimilation (4D-Var) method, especially the application of incremental 4D-Var and multi-scale variational assimilation strategies, has achieved the effective use of observational data with different spatiotemporal resolutions, and improved the accuracy and computational efficiency of assimilation. The construction of the multi-scale data assimilation framework has solved the problem of data fusion at different scales and achieved seamless data assimilation from local to global. The introduction of the scale conversion algorithm ensures the consistency of data with different resolutions during the assimilation process. The application of the dynamic weight allocation mechanism enables the model to adaptively adjust the weights of different assimilation methods according to data quality and model performance, thereby improving the overall assimilation effect. Through the comprehensive application of these technologies, this step not only improves the prediction accuracy and timeliness of the hydrodynamic model, but also enhances the model's ability to utilize multi-source and multi-scale observational data, providing more accurate and timely hydrological and hydrodynamic information for sluice scheduling optimization, which helps to improve the scientificity and real-time nature of scheduling decisions, especially playing a key role in responding to extreme hydrological events.

[0035] According to one aspect of the present application, the process of deep reinforcement learning sluice control strategy training is specifically as follows: S421. Obtain the multi-objective sluice group scheduling optimization model output by S41 and build a reinforcement learning environment that meets the OpenAI Gym standard; define the state space, including variables such as water level, flow, and gate opening; design the action space to represent the possible options for sluice operation; develop a reward function to quantify the immediate effect of scheduling decisions; implement the state transfer function to simulate the impact of sluice operation on the hydrodynamic system; and output a standardized reinforcement learning environment.

[0036] S422, receive the reinforcement learning environment output by S421, design a LSTM-based state encoder to capture the characteristics of hydrological time series; build a multi-head attention layer to enhance the information interaction between sluices; develop a graph convolutional network module to process the water system topological structure information; design a strategy network (Actor) and a value network (Critic) to implement the Actor-Critic architecture; output a complete deep neural network model structure.

[0037] S423. Obtain the neural network model output by S422, implement the Proximal Policy Optimization (PPO) algorithm, and improve training stability; adopt the Generalized Advantage Estimation (GAE) method to balance the bias and variance; introduce a curiosity-driven exploration strategy to enhance the exploration of the environment; implement curriculum learning to gradually increase the difficulty of tasks; develop a multi-agent collaborative learning framework to promote the overall optimization of the sluice group; output the optimized reinforcement learning training algorithm.

[0038] S424: Receive the optimization algorithm output by S423 and design a distributed training architecture based on Ray; implement the experience replay buffer pool to improve sample utilization efficiency; develop an asynchronous parameter update mechanism to accelerate the training process; adopt a priority experience replay strategy to focus on key samples; implement periodic model evaluation and checkpoint saving; and output an efficient distributed reinforcement learning training system.

[0039] S425. Obtain the training system and preliminary training strategy output by S424, design a multi-scenario test environment to cover various hydrological conditions; implement Monte Carlo simulation to evaluate the stability and robustness of the strategy; apply sensitivity analysis to identify the key influencing factors of the strategy; fine-tune the reinforcement learning strategy through expert knowledge guidance; conduct comparative experiments to compare performance with traditional scheduling methods; output a fully evaluated and fine-tuned coordinated control strategy for the sluice group.

[0040] According to one aspect of the present application, the process of robust optimization and adaptive scheduling strategy generation is specifically as follows: S441. Obtain the coordinated scheduling plan output by S43 and the model uncertainty estimate of step S3, and apply the Latin hypercube sampling method to generate representative scenarios; construct a multi-level scenario tree to reflect the evolution of uncertainty at different time scales; use the copula function to simulate the correlation structure between hydrological variables; generate a large number of possible future scenarios through Monte Carlo simulation; and output a structured uncertainty scenario set.

[0041] S442. Receive the scenario set output by S441, implement the conditional value at risk (CVaR) calculation module, and evaluate the risks under extreme situations; develop a dynamic risk budget algorithm to allocate risk tolerance in the time dimension; design multi-dimensional risk assessment indicators, comprehensively considering flood control, water supply, ecology and other goals; apply fuzzy comprehensive evaluation methods to quantify the comprehensive risks under different scenarios; and output multi-dimensional risk assessment results.

[0042] S443. Obtain the risk assessment results of S442, implement the robust dual dynamic programming algorithm, and handle multi-stage decision-making problems; develop scenario decomposition technology to improve the efficiency of solving large-scale problems; design an adaptive weight adjustment mechanism to balance the trade-offs between different objectives; introduce a flexible constraint processing method to enhance the adaptability of the optimization model; accelerate the solution process through the alternating direction multiplier method (ADMM); and output a robust adaptive optimization algorithm framework.

[0043] S444, receive the optimization algorithm framework output by S443, construct dynamic decision rules based on decision trees; introduce fuzzy logic controllers to handle the uncertainty of decision boundaries; develop neural network approximators to learn complex nonlinear decision mappings; design multi-level decision structures to achieve coordination of long-term and short-term decisions; optimize the parameters of decision rules through reinforcement learning methods; and output flexible dynamic decision rule sets.

[0044] S445. Obtain the dynamic decision rule set output by S444, design a rolling time domain optimization framework, and regularly update the long-term scheduling plan; implement a fast re-optimization algorithm to respond to emergencies and new information; develop a prediction-correction mechanism to continuously improve the model prediction accuracy; design a multi-time scale coordination method to balance short-term operations and long-term planning; improve the real-time performance of rolling optimization through parallel computing technology; and output a complete rolling time domain optimization system.

[0045] S446, receives the rolling optimization system output by S445 and the uncertainty scenario set of S441, designs a comprehensive scenario testing plan covering normal, extreme and emergency situations; implements large-scale parallel simulation to evaluate the performance of the strategy under different scenarios; applies statistical analysis methods to quantify the robustness and adaptability of the strategy; develops visualization tools to intuitively display the performance of the strategy in multiple dimensions; identifies the advantages and potential risks of the strategy through sensitivity analysis; outputs a detailed multi-scenario response capability assessment report and the final robust adaptive scheduling strategy.

[0046] According to another aspect of the present application, a regional sluice system scheduling optimization system is also provided, comprising: at least one processor; and, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement a regional sluice system scheduling optimization method described in any of the above technical solutions.

[0047] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for optimizing regional sluice system scheduling, characterized in that: The steps include: Step S1, acquiring multi-source heterogeneous data including remote sensing images, ground observation station data, meteorological radar data, and Internet of Things sensor data, and processing and standardizing them using an adaptive multimodal data fusion algorithm to generate a standardized data set aligned in time and space; using a deep convolutional long short-term memory network and a multimodal feature extraction method enhanced by an attention mechanism to extract spatiotemporal features from the standardized data set to form a preliminary feature data set; constructing a dynamic graph neural network based on the water system topological structure, further processing the preliminary feature data set to generate a dynamic feature data set containing the mutual influence of water system elements; applying multiscale wavelet decomposition and reconstruction technology, combined with a nonlinear dynamic analysis method, extracting multiscale hydrological dynamic features from the dynamic feature data set, and outputting a fused multi-dimensional, multi-scale hydrological dynamic feature data set; based on the self-organizing criticality theory, performing uncertainty analysis and critical state detection on the fused data set, and outputting a comprehensive data set with uncertainty estimation and system critical state indicators; Step S2, obtain a comprehensive data set with uncertainty estimates and system critical state indicators, apply a dynamic Bayesian network model to simulate the hydrological process, and generate a preliminary hydrological scenario set; use a conditional generative adversarial network to enhance and refine the preliminary scenario to generate a high-resolution hydrological scenario; use extreme value theory and coupled Markov chain methods to simulate extreme hydrological events and form an extended scenario set containing extreme situations; use dynamic pattern decomposition technology to extract the dominant mode of the hydrological system, and generate long-term hydrological trend forecasts based on these modes; integrate all generated scenarios, apply fuzzy comprehensive evaluation methods to sort and screen the scenarios, and output the final adaptive hydrological scenario set; Step S3, obtaining an adaptive hydrological scenario set and a multi-dimensional hydrodynamic characteristic data set, constructing a multi-scale coupled hydrodynamic model including a one-dimensional river channel model, a two-dimensional floodplain model and an urban drainage model; applying adaptive grid refinement technology to optimize the model spatial resolution and improve the simulation accuracy of key areas; using data assimilation methods to integrate real-time observation data and model prediction results to improve the timeliness and accuracy of the model; using a multi-objective optimization algorithm to automatically calibrate model parameters and balance the simulation effects of different scales and different evaluation indicators; through scenario-driven Monte Carlo simulation, evaluating the uncertainty of the model, and outputting the calibrated multi-scale hydrodynamic model and its uncertainty estimation; Step S4, obtaining a calibrated multi-scale hydrodynamic model and its uncertainty estimation, as well as an adaptive hydrological scenario set, and constructing a multi-objective sluice group coordinated dispatch optimization model; Deep reinforcement learning methods are applied to train collaborative control strategies for sluice groups to achieve unified scheduling of multiple sluices. Game theory methods are used to coordinate conflicts of interest between upstream and downstream sluices to form a balanced scheduling plan. Robust optimization techniques are used to consider model uncertainty and scenario changes to generate highly adaptable scheduling strategies. Combined with the results of eco-hydrological demand assessment, scheduling strategies are optimized to maintain the ecological health of the river. The long-term effects of scheduling strategies are evaluated through large-scale parallel simulations, and the final collaborative optimization scheduling plan for the sluice group is output.

2. A method for optimizing regional sluice system scheduling as claimed in claim 1, characterized in that: The step S1 is specifically as follows: Step S11, acquiring multi-source heterogeneous data, including remote sensing images, ground observation station data, meteorological radar data, and IoT sensor data, applying an adaptive data cleaning algorithm to remove noise and detect outliers; using a dynamic time warping algorithm to achieve time alignment of data with different sampling frequencies; Use spatial interpolation technology to align the data spatially and generate a preliminary dataset that is consistent in time and space. Use the Z-score standardization method to normalize the preliminary dataset and output a standardized multi-source dataset. Step S12, receiving the standardized multi-source data set output by S11, constructing and training a deep convolutional long short-term memory network; the convolution layer is used to extract spatial features, and the LSTM layer captures time series features; Batch normalization and residual connection are used to optimize network performance; cross-validation method is used to select the optimal model parameters, and the trained DC-LSTM model and the initially extracted spatiotemporal feature dataset are output; Step S13, obtaining the preliminary spatiotemporal feature data set output by S12, calling the multi-head self-attention mechanism; calculating the attention weights of the features of different data sources; Adaptively fuse features based on attention weights; Apply feature importance analysis method to screen key features; Output enhanced multimodal feature dataset; Step S14, receiving the enhanced multimodal feature dataset output by S13, and combining it with the pre-stored water system topology data; constructing a dynamic graph reflecting the water system structure; constructing a graph convolution layer and a gated recurrent unit to capture the dynamic interaction between nodes; and applying a graph attention mechanism to enhance the influence of key nodes; Train the DGNN model to output a graph embedding representation containing the dynamic characteristics of the water system; Step S15, obtaining the graph embedding representation output by S14, selecting a data-rich region as the source domain; using an adversarial training method to reduce the feature distribution difference between the source domain and the target domain; adopting a progressive migration strategy to gradually adapt to the target domain; Apply active learning methods to select the most informative samples for annotation; fine-tune model parameters to output universal feature representations that are suitable for different regions; Step S16, receiving the general feature representation output by S15, applying discrete wavelet transform to perform multi-scale decomposition; performing threshold processing on wavelet coefficients of different scales to remove noise; and using wavelet packet analysis method to extract more refined frequency features; Construct an adaptive weighting scheme to weight different scales based on feature importance; reconstruct the signal through inverse wavelet transform and output a multi-scale feature dataset; Step S17, obtaining the multi-scale feature data set output by S16, and applying Takens embedding theorem to reconstruct the phase space; Calculate the correlation dimension and information dimension to characterize the complexity of the system; estimate the maximum Lyapunov exponent to quantify the degree of chaos in the system; use recursive quantitative analysis methods to extract the periodic and deterministic characteristics of the system; build a nonlinear prediction model to evaluate the predictability of the system; output an enhanced data set containing nonlinear dynamic characteristics; Step S18, receiving the enhanced data set output by S17, applying the power law distribution fitting method to detect the self-organized critical state of the system; calculating the Fisher information amount to quantify the order change of the system; The permutation entropy method is used to evaluate the complexity and uncertainty of the system. The ensemble learning method is used to integrate the results of multiple detectors to improve the reliability of detection. Monte Carlo simulation is applied to evaluate the uncertainty of feature extraction and state detection. The final comprehensive data set with uncertainty estimation and system critical state indicators is output.

3. A method for optimizing regional sluice system scheduling as claimed in claim 1, characterized in that: The step S2 is further as follows: Step S21, obtaining the comprehensive data set outputted from step S1, and combining it with the pre-stored hydrological model parameter library; constructing a dynamic Bayesian network structure reflecting the interrelationships between hydrological elements; and using the expectation maximization (EM) algorithm to learn the network parameters; Apply particle filtering method to perform online inference of sequence data; generate multiple groups of hydrological processes through Monte Carlo simulation and output a preliminary hydrological scenario set; Step S22, receiving the preliminary hydrological scenario set output by S21, constructing and training the CGAN model; the generator network uses the U-Net architecture, and the discriminator adopts the Patch GAN structure; calling the Wasserstein distance as the loss function to improve the training stability; using the spectral normalization technology to avoid the gradient vanishing problem; through iterative optimization of the preliminary scenario, generating a high-resolution and more realistic hydrological scenario, and outputting an enhanced hydrological scenario set; Step S23, obtaining the enhanced hydrological scenario set output by S22, applying the generalized extreme value distribution (GEV) to fit extreme hydrological events; using the time-varying copula function to simulate the time dependence of extreme events; using the piecewise Markov chain method to simulate the duration and intensity of extreme events; combining the climate change prediction data to adjust the probability of occurrence of extreme events; generating extended scenarios containing various extreme conditions, and outputting the extended scenario set containing extreme hydrological events; Step S24, receiving the extended scenario set outputted by S23, and applying the dynamic mode decomposition (DMD) technique to extract the dominant dynamic mode of the hydrological system; Use sparse recognition algorithm to optimize DMD model and improve the stability of long-term prediction; Combined with wavelet transform, it can capture multi-scale periodic changes; apply Mann-Kendall test to identify significant long-term trends; generate long-term hydrological forecasts based on the extracted dominant patterns and trends, and output long-term hydrological trend forecast results; Step S25, obtaining the output results of S22, S23 and S24, and constructing a multi-criteria evaluation system, including the rationality, diversity and representativeness of the scenarios; The fuzzy comprehensive evaluation method was applied to calculate the comprehensive score of each scenario; the principal component analysis method was used to reduce the redundancy of the scenarios; The K-means++ algorithm was used to group the scenarios; Based on the system critical state indicators, scenarios close to the critical state are given priority; the Monte Carlo cross-validation method is used to evaluate the robustness of the scenarios; and finally an adaptive hydrological scenario set that has been evaluated and screened is output.

4. A method for optimizing regional sluice system scheduling as claimed in claim 1, characterized in that: The step S3 is further as follows: Step S31, obtaining the hydrodynamic characteristic data set outputted in step S1 and the pre-stored terrain data, constructing a one-dimensional river channel Saint-Venant equation model; applying the finite volume method to discretize the two-dimensional shallow water equation and establish a floodplain model; using a hyperbolic model to describe the water flow in the urban pipe network and constructing an urban drainage model; constructing a dynamic coupling interface between models to realize water volume exchange; using OpenMP parallel computing technology to optimize the model operation efficiency, and outputting a preliminary multi-scale hydrodynamic model framework; Step S32, receiving the preliminary model framework outputted by S31, applying the quadtree grid partitioning method to realize adaptive grid refinement; dynamically adjusting the grid resolution based on the water flow gradient and terrain characteristics; and using the boundary fitting technology to improve the simulation accuracy of complex terrain; The sub-grid parameterization method is used to consider the impact of small-scale terrain features on water flow; GPU acceleration technology is used to improve the computational efficiency of refined grids and output an optimized high-resolution hydrodynamic model; Step S33, obtaining the optimization model outputted by S32 and the real-time observation data of step S1, applying the ensemble Kalman filter method for data assimilation; using the particle filter algorithm to process the observation data with strong nonlinearity; Use variational assimilation methods to optimize the model's initial and boundary conditions; call multi-scale data assimilation strategies to coordinate the impact of data at different scales; build a sliding time window assimilation scheme to achieve continuous correction of the model; output a dynamic hydrodynamic model with real-time correction capabilities; Step S34, receiving the dynamic model output by S33 and part of the hydrological scenario of step S2, constructing a multi-objective evaluation system, including water level, flow and flooding range; applying the NSGA-III algorithm to perform multi-objective optimization and find the Pareto optimal solution set; calling the ε-constraint method to deal with the trade-off relationship between objectives; using the alternating direction multiplier method to accelerate the optimization process; using the Latin hypercube sampling method to generate an initial parameter set to improve the optimization efficiency; Evaluate the stability of parameters through cross-validation and output multiple sets of Pareto optimal parameter sets; Step S35, obtaining the Pareto optimal parameter set outputted by S34 and the complete hydrological scenario set of step S2, and applying the GLUE method to evaluate parameter uncertainty; Use Sobol sensitivity analysis method to identify key parameters; The Bootstrap method is used to evaluate the structural uncertainty of the model; fuzzy mathematics theory is used to quantify the uncertainty of boundary conditions; scenario-driven Monte Carlo simulation is used to comprehensively evaluate the performance of the model under different situations; and finally the calibrated multi-scale hydrodynamic model and its comprehensive uncertainty estimate are output.

5. A method for optimizing regional sluice system scheduling as claimed in claim 1, characterized in that: The step S4 is further as follows: Step S41, obtaining the hydrodynamic model of step S3 and the pre-stored sluice characteristic data, establishing a state transfer equation reflecting the relationship between sluice operation and hydrodynamic response; constructing a multi-objective evaluation function, including flood prevention and disaster reduction, water resource utilization and ecological water demand; Invoke constraints, including sluice scheduling rules and safe operation restrictions; adopt a multi-agent system framework to describe the interaction of sluice groups; use dynamic programming methods to build basic solution algorithms; output a preliminary multi-objective sluice group scheduling optimization model; Step S42, receiving the scheduling optimization model output by S41, constructing a deep reinforcement learning framework based on the actor-critic architecture; using a long short-term memory network (LSTM) as a state encoder to capture hydrological time series characteristics; Adopting attention mechanism to enhance information interaction between sluice gates; Call the priority experience playback mechanism to improve learning efficiency; The double Q-learning algorithm is applied to alleviate the over-estimation problem; the task difficulty is gradually increased through the curriculum learning strategy; Use distributed training technology to accelerate the learning process; Output the initially trained coordinated control strategy for the sluice group; Step S43, obtaining the preliminary control strategy outputted by S42, constructing a non-cooperative game model reflecting the interest relationship between the upstream and downstream sluice gates; applying the Nash equilibrium theory to find the balance point of the interests of all parties; The Shapley value method is used to fairly distribute the overall benefits brought by scheduling; the evolutionary game theory is used to simulate the evolution of long-term scheduling strategies; and the multi-level bargaining model is used to handle complex interest negotiation processes; Explore the global optimal solution through cooperative game methods; output a coordinated scheduling plan that balances the interests of all parties; Step S44, receiving the coordinated scheduling plan outputted from S43 and the model uncertainty estimation from step S3, and constructing an uncertainty set by applying a scenario tree method; Use the conditional value at risk (CVaR) metric to assess risks in extreme situations; use a robust adaptive optimization algorithm to generate a scheduling strategy that is highly adaptable to uncertainty; Invoke dynamic decision rules to improve the real-time adjustment capability of strategies; use rolling time domain optimization methods to regularly update long-term scheduling plans; evaluate the robustness of strategies through multi-scenario response capability tests; and output robust scheduling strategies with strong adaptability; Step S45, obtain the robust scheduling strategy output by S44 and the ecosystem data of step S1, establish a river ecological water demand assessment model; apply the ecohydraulics method to quantify the impact of hydrological conditions on the habitats of key species; call on the environmental flow concept to determine the minimum flow requirement to maintain the health of the ecosystem; use HSI to evaluate the ecological effects of different scheduling schemes; adopt a multi-objective trade-off analysis method to balance flood control, water supply and ecological needs; optimize the scheduling strategy through ecological response simulation to improve the health of the ecosystem; output an optimized scheduling strategy that comprehensively considers ecological needs; Step S46, receive the optimized scheduling strategy output by S45 and the complete hydrological scenario set of step S2, build a long-term simulation experiment plan, covering a variety of hydrological and climate scenarios; use high-performance computing clusters to achieve large-scale parallel hydrodynamic simulation; apply the Monte Carlo method to evaluate the performance of the strategy under different scenarios; call extreme event response evaluation to test the emergency handling capability of the strategy; use long-term trend analysis methods to evaluate the adaptability of the strategy to climate change; identify potential risk points of the scheduling strategy through system vulnerability analysis; and comprehensively analyze the evaluation results to output the final sluice group collaborative optimization scheduling plan and its long-term effect evaluation report.

6. A method for optimizing regional sluice system scheduling as claimed in claim 1, characterized in that: The step S16 is specifically as follows: Step S161, obtaining the general feature representation data output by S15, applying the information entropy criterion to evaluate the performance of different wavelet basis functions; using a genetic algorithm to optimize wavelet basis parameters, including support length and vanishing moment; selecting the most suitable wavelet basis function by a cross-validation method; outputting the preferred wavelet basis function and its parameters; Step S162, receiving the preferred wavelet basis function outputted by S161, performing discrete wavelet transform on the general feature representation data; using an adaptive scale selection algorithm, dynamically determining the number of decomposition layers according to the local characteristics of the data; using the Best Basis algorithm to select the optimal basis in the wavelet packet; calculating the wavelet coefficients of each scale, and outputting a multi-scale wavelet coefficient set; Step S163, obtaining the multi-scale wavelet coefficient set outputted by S162, applying the SURE criterion to automatically select the optimal threshold; using the soft threshold method to perform denoising on the wavelet coefficients; Adopting adaptive threshold strategy, using different thresholds for different scales and different features; outputting denoised multi-scale wavelet coefficient sets; Step S164, receiving the denoised wavelet coefficient set outputted by S163, and calculating the energy distribution and statistical characteristics of each scale; Apply principal component analysis to reduce feature dimensionality; use random forest algorithm to evaluate feature importance; Select the most representative feature subset based on the mutual information criterion; Output multi-scale feature sets and their importance weights; Step S165, obtaining the multi-scale feature set and importance weights output by S164, constructing a weighted reconstruction strategy based on feature importance; using inverse discrete wavelet transform to reconstruct the signal; using an adaptive fusion algorithm to integrate reconstruction results of different scales; Apply ensemble learning methods to improve the stability and generalization ability of the reconstruction results; output the final multi-scale feature dataset.

7. A method for optimizing regional sluice system scheduling as claimed in claim 1, characterized in that: The step S22 is specifically as follows: Step S221, obtain the preliminary hydrological scenario set output by S21, build a U-Net-based generator network architecture; build a Patch GAN discriminator network; call the conditional embedding layer, and use the hydrological scenario features as conditional input; build a multi-scale feature extraction module to capture hydrological features at different spatial scales; Output the initial architecture of the CGAN model; Step S222, receiving the CGAN initial architecture outputted by S221, constructing a composite loss function, including adversarial loss, content loss, and feature matching loss; calling Wasserstein distance as a measure of adversarial loss; using a gradient penalty strategy to improve training stability; constructing a progressive training scheme to gradually improve the resolution of the generated image; Output optimized loss function and training strategy; Step S223, obtaining the loss function and training strategy output by S222, using the Adam optimizer for model training; applying the learning rate decay strategy; Call spectral normalization technology to alleviate the gradient vanishing problem; use batch normalization method to accelerate the training process; determine the optimal hyperparameter combination through cross-validation; output the trained CGAN model; Step S224, receiving the trained CGAN model output by S223 and the preliminary hydrological scenario set of S21, and using the sliding window technology to process the large-scale hydrological scene; Apply conditional control strategies to generate corresponding high-resolution scenarios according to different hydrological conditions; adopt integrated methods to fuse multiple generated results to improve stability; use post-processing algorithms to ensure the physical consistency of generated results; output high-resolution hydrological scenario sets; Step S225, obtain the high-resolution hydrological scenario set output by S224, construct multidimensional evaluation indicators, including authenticity, diversity and conditional consistency; use FID to evaluate the quality of the generated results; apply MMD to measure the difference between the generated distribution and the true distribution; use the expert scoring system to comprehensively evaluate the rationality of the scenario; based on the evaluation results, screen out the optimal enhanced hydrological scenario set; output the final high-quality, high-resolution hydrological scenario set.

8. A method for optimizing regional sluice system scheduling as claimed in claim 1, characterized in that: The step S32 is further as follows: Step S321, obtaining the preliminary model framework outputted in S31 and pre-stored terrain data, and applying the Delaunay triangulation method to generate an initial unstructured grid; Calculate mesh quality metrics, including mesh distortion and aspect ratio; Identify key hydrogeomorphic features, including river channels, levees, and depressions; Output the initial mesh and its quality assessment report; Step S322, receiving the initial grid outputted in S321, and constructing a refinement criterion based on water flow gradient, terrain relief and hydrodynamic characteristics; Call the curvature adaptive method to increase the mesh density in high curvature areas; Use error estimation techniques to identify high error areas that need to be refined; Build dynamic threshold strategies to automatically adjust refinement criteria based on the simulation process; Output adaptive refinement criteria and threshold sets; Step S323, obtaining the refinement criteria outputted by S322 and the initial grid of S321, implementing a quadtree or octree grid refinement algorithm; applying a local refinement strategy to maintain a gradual change in the size of adjacent cells; Implement dynamic load balancing to optimize parallel computing efficiency; go through a recursive refinement process until all refinement criteria are met; output a multi-resolution adaptive grid; Step S324, receiving the multi-resolution grid outputted in S323, using isogeometric analysis method to accurately fit the complex boundary; using moving grid technology to handle time-varying boundary problems; applying Laplace smoothing algorithm to improve grid quality; Implement local reconstruction technology to eliminate degenerate cells; balance mesh orthogonality and smoothness through iterative optimization process; output high-quality boundary-fitting adaptive mesh; Step S325, obtain the high-quality grid output by S324, call the sub-grid terrain parameterization method to capture the small-scale terrain features; construct the concept of effective hydraulic radius to characterize the water flow resistance at the sub-grid scale; Establish a porous media model to simulate the flow characteristics at the sub-grid scale; integrate the sub-grid information into the computational grid through the scale conversion function; Output the optimized hydrodynamic model mesh including subgrid parameterization.

9. A method for optimizing regional sluice system scheduling as claimed in claim 1, characterized in that: The step S33 is specifically as follows: Step S331, obtaining the real-time observation data of step S1 and the pre-stored historical data, applying an outlier detection algorithm to identify and mark suspicious data; using a wavelet analysis method to remove high-frequency noise; and using a multivariate regression technique to interpolate missing data; Implement consistency checks to ensure the temporal and spatial consistency of data from different sources; output quality-controlled observational datasets; Step S332, receiving the post-quality control observation data output by S331, generating initial set members to reflect the uncertainty of the model state; constructing a covariance localization scheme to reduce the impact of false correlation; implementing covariance inflation technology to avoid filter divergence problems; calling an adaptive inflation factor algorithm to dynamically adjust the set discreteness; outputting a well-configured EnKF assimilation framework; Step S333, obtaining the EnKF framework output by S332 and the observation data of S331, constructing an importance sampling strategy, and improving the representativeness of particles; Implement resampling algorithm to alleviate particle degradation problem; use Markov chain Monte Carlo method to enhance particle diversity; call smoothing technology to improve the processing ability of non-Gaussian distribution; output optimized particle filter assimilation scheme; Step S334, receiving the observation data of S331 and the optimized hydrodynamic model output by S32, constructing an adjoint model, and calculating the gradient of the cost function; implementing the incremental 4D-Var method to improve the computational efficiency; adopting a multi-scale variational assimilation strategy to process observations of different temporal and spatial resolutions; Call the observation operator to map the model variables to the observation space; Through an iterative optimization process, the cost function is minimized; Output 4D-Var assimilation system and its optimal initial and boundary conditions; Step S335, obtaining various assimilation schemes output by S332, S333 and S334, constructing a multi-scale data assimilation framework, and coordinating the influence of data at different scales; Implement the scale conversion algorithm to achieve seamless integration of data of different resolutions; call the dynamic weight allocation mechanism to adjust the weights of different assimilation methods according to data quality and model performance; evaluate the overall performance of the multi-scale assimilation strategy through integrated testing; and output a comprehensive multi-method, multi-scale data assimilation system.

10. A regional sluice system scheduling optimization system, characterized in that: include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement a regional sluice system scheduling optimization method as described in any one of claims 1 to 9.

Citation Information

Cited By

  • Construction method and application of variable-working-condition bearing fault diagnosis model

    CN115310495A

  • A method for constructing a variable working condition bearing fault diagnosis model and application thereof

    CN115310495B

  • Urban flood control toughness critical state identification method and system considering multiple pressure coupling

    CN119226782A

  • College campus planning and design multi-objective optimization method based on cGAN and NSGA-II

    CN119293936A

  • Method and system for dynamically updating spatio-temporal data elements based on GIS (Geographic Information System) support

    CN119311707A