耦合信号分解和机器学习的防洪调度规则提取研究OA

Existing methods for extracting reservoir operation rules often struggle to fully capture the operational information em-bedded in optimization outputs.To address this limitation,a series of flood-control operation rule extraction schemes integrating signal decomposition and machine learning were developed and systematically evaluated.Taking the Three Gorges Reservoir(TGR)as a case study,a multi-objective flood-control operation optimization model was first established by jointly considering upstream and downstream flood-control requirements.The Non-dominated Sorting Genetic Algorithm Ⅱ(NSGA-Ⅱ)was em-ployed to derive the Pareto-optimal solutions,from which a preferred operation scheme for each year was identified using the en-tropy-weighted Technique for Order Preference by Similarity to Ideal Solution(TOPSIS).Subsequently,four signal decomposi-tion techniques,namely Ensemble Empirical Mode Decomposition(EEMD),Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN),Variational Mode Decomposition(VMD),and Empirical Wavelet Transform(EWT),were integrated with four machine-learning algorithms,including Extreme Gradient Boosting(XGBoost),Light Gradient Boosting Ma-chine(LightGBM),Radial Basis Function Neural Network(RBFNN),and Convolutional Neural Network(CNN),yielding 16 hybrid configurations.The decomposed inflow components served as model inputs,while reservoir outflow was used as the predic-tion target for extracting flood-control operation rules.Based on daily flood-season data from the TGR for 2011~2024,the com-parative results indicate that:① integrating signal decomposition with machine learning generally enhances the accuracy of reser-voir outflow simulation,increasing the coefficient of determination(R2)by more than 0.1 compared with their corresponding stan-dalone machine-learning models;② VMD and CEEMDAN showed greater effectiveness when combined with tree-based algo-rithms,whereas CEEMDAN and EWT were more advantageous for neural-network-based models,with maximum improvements in R2 of approximately 0.3;③ EWT-based hybrid models achieved the best overall performance in reproducing optimized opera-tion schemes,with EWT-RBFNN exhibiting the highest overall performance,followed by EWT-CNN.These findings demon-strate the potential of combining signal decomposition with machine to extract flood-control operation rules derived from optimiza-tion and provide a systematic basis for selecting appropriate hybrid modeling strategies for complex reservoir systems.