Research On Short-term Prediction Method Of Fan Power Based On Machine Learning

With the large-scale application of wind power generation,the wind power will directly impact the load of the whole power grid,and the problems of randomness,intermittence and volatility are increasingly obvious,which adds a lot of pressure to the power system control and power grid dispatching of the whole power grid.Therefore,in order to make sure the grid operating safely and stably after the wind energy connection,it is urgent to study the wind power prediction.Therefore,according to the problem of short-term prediction of fan power,taking advantage of deep learning in machine learning,and based on the analysis and processing of fan power data,an in-depth study of combined prediction model is carried out.This paper firstly studies the development status of wind power forecasting technique.According to the study background and status quo of wind power forecast both here and abroad,the existing technologies are analyzed and summarized.To solve the problem of instability of current prediction models,the research direction based on deep learning is determined,the structure flow of combined model prediction is put forward,and the evaluation indexes of model prediction results are introduced.Secondly,the data of fan power prediction is preprocessed.Aiming at the problem of random abnormal fluctuations of data,through comparison,this paper puts forward an integrated empirical mode decomposition algorithm with adaptive noise to decompose the data,extract the characteristic components in different time scales,complete the processing of abnormal data,and make sure the validity of the fan power prediction figures.Through experimental analysis and study,the feasibility of the data preprocessing method is verified,which provides a reliable data basis for subsequent prediction.Finally,the combination forecasting model is trained and verified by an example.Back propagation neural network and long-term and short-term memory neural network are selected to make a comparative study of prediction based on these two networks.On the basis of data preprocessing,aiming at the problems of low fitting degree,poor convergence ability and optimization of the prediction results of a single model,an improved Sparrow Search Algorithm is proposed to optimize the model,and the prediction models of CEEMDAN-ISSA-BPNN and CEEMDAN-ISSALSTM are established.Different data samples are selected to verify and analyze the prediction performance of the combination model,which proves the prediction advantages of the prediction model.Based on the historical data of a wind farm in Shanghai,two combined algorithm prediction models with different advantages are obtained.The CEMDAN-ISSA-BPNN model has better prediction effect when the data fluctuates greatly;When the data fluctuation is relatively stable,CEEMDAN-ISSA-LSTM model has higher prediction fitting degree.