Research On Middle And Long-term Generation Dispatching Rules Of Cascade Hydropower Stations Based On Machine Learning

With the planning,construction and commissioning of a large number of hydropower stations in my country,a group of cascade hydropower stations represented by the Three Gorges-Gezhouba cascade,Xiluodu-Xiangjiaba cascade,etc.have been preliminarily formed,and the development and utilization of hydropower energy in the basin has moved from planning and construction to cascade hydropower stations.The key transition period of joint optimization dispatch management transfer.The middle and long-term optimal dispatching of cascade hydropower stations can effectively promote the efficient use of water resources in the river basin and significantly improve the power generation efficiency of cascade hydropower stations.The dispatching rules or dispatching functions of hydropower stations are an important basis for guiding the middle and long-term dispatching operation of hydropower stations.Generally,they can be inferred according to the optimal operation process of hydropower stations under long-term deterministic incoming water conditions.At present,most of the research on dispatching rules focuses on the extraction of optimal dispatching rules for single hydropower stations.Although there has been some progress in the research of dispatching rules for cascade hydropower stations,there are still insufficient deterministic optimization dispatching models for cascade hydropower stations,low solution efficiency,and dispatching rule decision-making.The factor determination method is single and empirical,the model fitting and generalization ability of the dispatching rule extraction model is not balanced,and the actual application effect is not good.To this end,based on the in-depth development of the project of "Research and System Development of Water Resources Management Decision Support Models for the Lower Jinsha River-Three Gorges Cascade Hydropower Stations",this paper focuses on the core problem of extracting middle and long-term power generation dispatching rules for cascade hydropower stations,and proposes a BP Neural network calculation method for water consumption rate of hydropower station,established a middle and long-term power generation dispatching model for cascade power station based on water consumption rate output calculation;Combined with random forest algorithm and Pearson correlation analysis method,on the basis of fully considering the interconnection between cascade reservoirs,a "primary selection-optimization-de-duplication" decision factor screening method is proposed;Gaussian process regression model is introduced,and a BOA-GPR scheduling rule extraction model based on Bayesian optimization algorithm hyperparameter optimization is constructed.The research results have been integrated into the "Decision Support System for Water Resources Management of Three Gorges Cascade Power Stations on the Lower Reaches of the Jinsha River" and formally deployed in the Three Gorges Cascade Dispatching Center of the Yangtze Power Company.The main work content and innovations of the thesis are as follows:(1)Aiming at the problem of insufficient accuracy of the traditional K-value output calculation method in the middle and long-term power generation optimal dispatching,based on the actual operating data of the hydropower station,the BP neural network is used to mine the nonlinear relationship between the water consumption rate and the operating head of the hydropower station,and the relationship between the two is obtained.The simulation calculation verifies the accuracy of the water consumption rate curve applied to the output calculation;On this basis,a medium and long-term power generation dispatch model for cascade hydropower stations based on the water consumption rate curve is established,and the DP algorithm,the POA algorithm and the DPSA algorithm are used in an organic combination for the solution of the model.By comparing with the DP algorithm,the effectiveness and engineering practicability of the combined algorithm are verified.(2)In view of the strong subjectivity in the selection of decision factors in the power generation dispatching rules of cascade hydropower stations and the insufficient consideration of the correlation relationship of cascade hydropower stations,based on the random forest algorithm and Pearson correlation analysis,based on the full consideration of the interconnection and influence between cascade reservoirs Law proposed a set of "primary selection-optimization-de-duplication" decision factor determination method.In turn,through the preliminary selection of decision factors,priority ranking and optimization of decision factors,and relevance analysis of decision factors,the decision factors are screened layer by layer,and finally the decision factors of cascade hydropower generation dispatching rules are obtained.The relevant method is applied to the Xiluodu-Xiangjiaba cascade hydropower stations,and the conventional dispatching rule extraction model is used to verify the effectiveness of the method.The results show that the method can more accurately analyze the complex hydraulic connections of the cascade hydropower stations and effectively identify the decision factors of each hydropower station.It can provide a new idea for the selection of decision-making factors in power generation dispatching rules of cascade hydropower stations.(3)Aiming at the problems of too simple models,insufficient optimization of hyperparameters,and poor actual application effects of dispatching rules in the traditional cascade hydropower station dispatching rules extraction,based on the Gaussian process regression model and Bayesian optimization algorithm,it is proposed to be used for the middle and long-term power dispatching of cascade hydropower stations.BOA-GPR model of rule extraction.The model is used to extract the dispatching rules of Xiluodu-Xiangjiaba cascade hydropower stations,and the dispatching rules are used in the simulated dispatching of cascade hydropower stations.At the same time,the conventional dispatching rule extraction model is introduced for comparative analysis.The results show that the BOAGPR model established in this paper can be used for cascade hydropower stations.Optimizing the dispatching rules for more adequate learning.The extracted cascade hydropower generation dispatching rules used to guide dispatching can significantly increase the cascade power generation and have practical application value.