Prediction Of Wind Speed Time Series Based On LSTM And ARIMA

With the over-exploitation of fossil fuels such as coal,oil and natural gas in recent years,the development of renewable energy has become an inevitable trend.Wind energy as a new environment-friendly energy,with its easy access,low cost and other advantages,has become the most potential of all new energy,the fastest development,relatively ma-ture technology of clean renewable energy.But as wind is a very unstable energy source and one of the most difficult meteorological factors to predict,it is of great significance to improve the accuracy of wind speed prediction.There are linear and nonlinear models in the traditional wind speed prediction mod-els,most of the nonlinear models are better than the linear models in predicting the wind speed data,which is a kind of nonlinear time series data,however,the single nonlinear model has its own limitations,so this paper considers the nonlinearity and linearity of the wind speed time series,and proposes a new hybrid modeling and forecasting method:LSTM-ARIMA.Using the strong non-linear fitting ability and fast learning ability of the short-term memory neural network LSTM to model the non-linear features of wind speed data,and then using ARIMA to model and predict the error series,finally,the final wind speed prediction results are obtained by adding the predicted results of the two models.Because the ARIMA model is used to fit the data linearly,the sum of the two models can be regarded as a small linear correction of the neural network's nonlinear prediction of wind speed with the prediction error.