Active Power Prediction And Optimal Scheduling Of Wind Turbine Based On Integrated Learning Algorithm

As a clean energy without pollution and renewable,wind energy has been effectively used in wind power generation in various countries.With the wide application of wind turbines,the variability of operating conditions and environmental factors of wind turbines,as well as the uncertainty of wind power output,put forward higher requirements for grid connected wind power generation and effective scheduling of wind turbines.Therefore,it is of great practical significance to realize the accurate prediction of wind turbine power generation and the optimal scheduling of wind turbine for improving the efficiency of grid connected generation of wind turbine and promoting the stable and safe operation of power system.In this thesis,aiming at the uncertainty of wind power output and the variability of wind turbine operating conditions,based on the SCADA(supervisory control and data)of wind turbine Acquisition)system data,taking the active power prediction of wind turbine and the power scheduling strategy of wind turbine as the research object,respectively,using the random forest algorithm to build the importance analysis and abnormal value correction of relevant factors,using the integrated learning algorithm to build the active power prediction model of wind turbine,using the evolutionary multi-objective optimization algorithm to build the multi-objective optimization of wind turbine The scheduling model is verified by some examples.The main research work of this paper is as follows:(1)Determine the relevant factors of active power prediction.Based on the study of the principle of wind power generation and operation,this paper analyzes the relevant influencing factors of the active power of wind turbine.Considering the mechanical loss,electrical loss,magnetic loss and environmental factors in the power factors,the air density,real-time wind speed,pitch angle,rotor speed,instantaneous wind direction,yaw angle,wind direction and shaft angle,coil voltage,coil current are selected respectively,coil temperature,engine room temperature,gearbox oil pool temperature,main bearing temperature,rotor coil temperature,gearbox oil pressure,actual torque and other parameters,and data are derived from the fan SCADA system.(2)The preprocessing model of related factors of active power prediction is constructed.First of all,analyze the data source and the cause of bad data,test the data rationality and deal with the abnormal value;secondly,take the load level determined by the active power as the classification index,based on random forest(random Forest,RF)classification algorithm,carries out attribute importance analysis,and solves the weight coefficient λ I of related attributes according to the obtained importance results;then,based on case-based reasoning(CBR)technology and k-nearest neighbor classification(k-nearest neighbor Based on the idea of KNN algorithm,this paper proposes a λ-3NN missing value completion model,which is verified by an example.Finally,according to the threshold setting of weight coefficient,the irrelevant factors and abnormal data are deleted,and the sample database of power prediction is constructed.(3)The active power prediction model based on the integrated algorithm is constructed.Based on the idea of integrated learning bagging algorithm,support vector machine(SVM)and extreme learning are selected respectively Machine,ELM)and Stochastic Forest regression algorithm are the basic learners;a dynamic weighted integrated learning strategy based on particle swarm optimization algorithm is proposed,which uses the good parameter optimization performance of particle swarm optimization algorithm to dynamically select the optimal individual model with the largest difference between them,and realizes the integrated learning of dynamic optimization;according to the normal power and limited power sample database,the integrated learning is carried out The example verification and performance analysis of the model are compared.(4)A multi-objective optimal scheduling model of wind turbine based on the improved non dominated sorting genetic algorithms(NSGA)is constructed.Taking the overall power deviation of the wind farm,the fluctuation amplitude of the output of the single wind turbine and the real-time health index of the whole unit as the optimization objectives,the output constraints of the single wind turbine,the total output constraints of the wind farm and the maximum output range constraints of the wind farm as the constraints,the multi-objective optimization scheduling model of the wind turbine is constructed.The evolutionary multi-objective optimization algorithm is used to solve the model,and 10 wind turbines in a wind farm in Hebei Province are taken as the verification object.Under the condition of the set scheduling task requirements and the health state index of each unit,the Pareto optimal frontier boundary solution is solved,which is proved to be highly accurate and practical by an example.