Short-term wind power prediction based on EMD-LSTM combined model

Abstract

Abstract In recent years, China is actively developing wind power generation. Wind energy is a natural factor with volatility, variability and uncontrollability, which will cause fluctuations in wind farm output. The accurate prediction of wind power is conducive to grid dispatchers deploying scheduling plans or doing ahead of schedule Adjustments to reduce losses to a certain extent are also conducive to improving wind power grid-connected capacity. In this paper, using the advantages of empirical mode decomposition EMD algorithm in nonlinear and non-stationary data processing, a wind farm power prediction model based on EMD-LSTM is established. First, the relevant data is preprocessed to obtain the ideal input sequence, and the LSTM network model is used. The NWP wind speed is corrected first, and the revised forecasted NWP wind speed sequence is closer to the actual wind speed. Then use EMD to decompose the wind power data sequence into data components of different scales, and then use the LSTM long and short-term memory network to model the decomposed IMF components and residual RES respectively, and sum the prediction results of each component and residual As the final prediction result. The research results show that the method described in this paper effectively improves the prediction.