Multi-Step wind power forecasting model Using LSTM networks, Similar Time Series and LightGBM

Intermittent and fluctuating wind forces are detrimental to the grid. A multivariate model was proposed to improve the accuracy of wind power generation prediction in order to induce system operators to reduce risks. The model consists of three steps. First, the meteorological data such as wind speed are predicted by LSTM networks on the basis of traditional time series approaches. Then a method of similar time series matching with hierarchical search is proposed to highlight the main factors and save computing time. We use similar disparity as a criterion to select similar meteorological series and power data as training sets. Finally, similar data are inputted into LightGBM for modeling, training, and prediction. Industrial data of the wind power plant is examined case. The results are clearly display that the proposed method can effectively predict wind power in the next 6 hours and achieve high precision, which has certain engineering practical value.