Research On Data-driven Reservoir Dynamic Scheduling Decision Algorithm Based On Feature Selection

With the development of the economy,large-scale reservoirs and hydropower stations have been gradually built,also,the hydrological data platform has been gradually improved.The dispatch of reservoirs and hydropower stations has shifted from traditional manual dispatching to intelligent optimal dispatching.Reservoir dispatching not only involves various objectives such as flood control,irrigation,power generation,and water supply,but also comprehensively coordinates the comprehensive needs and interests between various departments.Therefore,it is a multi-objective decision-making problem with complex constraints.At present,the models and methods for dispatching decision-making in reservoirs and hydropower stations are basically still handled according to the general multi-objective decision-making problem.The decision-makers need to choose the appropriate scheduling scheme from the alternatives according to the current specific situation.In this paper,a hydropower station scheduling decision algorithm based on real-time situation classification is proposed.Firstly,through the two feature selection algorithms proposed in the second and third chapters,the feature selection of hydrological and meteorological data is used to select the key factors affecting the reservoir scheduling mode.Secondly,based on the real-time hydrological and meteorological data,the current scheduling period is used.The scheduling decision is made,that is,the different scheduling targets are adjusted according to the real-time situation to provide more intelligent and real-time reservoir scheduling decision support.Finally,the current scheduling model is solved to obtain the scheduling scheme.main tasks as follows:(1)A wrapper feature selection algorithm(Dragonfly Algorithm-Partial Sequential Backward Floating Selection DA-PSBFS)based on the dragonfly algorithm and the sequence floating backward selection algorithm is proposed.The experimental results on the standard data sets show that the algorithm has higher classification accuracy.A smaller feature subset size was obtained,which was used to screen important reservoir parameters,hydrometeorological and other important features.(2)In order to overcome the shortcomings of high computational time consumption of DA-PSBFS algorithm,a hybrid feature selection algorithm HDAPSBFS(Hybrid Dragonfly Algorithm-Partial Sequential Backward Floating Selection)based on mutual information filtering method and DA-PSBFS is proposed.The experimental results of the proposed feature selection algorithm on the standard dataset show that it can have not only higher classification accuracy but also less computation time than other similar hybrid feature selection algorithms.(3)Introducing the idea of individual and group learning in particle swarm optimization algorithm and the idea of mutation in genetic algorithm in the basic algorithm.A single-object IDA(Improved Dragonfly Algorithm)algorithm is proposed and used to solve the three-objective model of flood control scheduling.(4)Taking the Lishui River Basin in Jiangxi Province as an example,using the feature selection algorithm proposed in the second and third chapter to analyze the characteristics of hydrology,meteorology,reservoir parameters,runoff,and downstream ecological environment data in the basin.The important characteristics of the basin scheduling model are divided into two categories: flood control and power generation irrigation.The flood control modes are divided into three sub-categories according to the flood level,and different scheduling models are applied.The proposed IDA algorithm is used for scheduling and solving,and the power generation irrigation is the daily scheduling mode of the reservoir group.The power generation and ecological two-objective optimization model is established and the model is solved by MOEA/D algorithm.