A LSTM Based Wind Power Forecasting Method Considering Wind Frequency Components and the Wind Turbine States

The existing ultra-short-term wind power prediction methods do not fully consider the influence of wind speed frequency components and wind turbine states. This leads the prediction accuracy is difficult to improve. An ultra-short-term wind power prediction method is proposed. This method is based on long short-term memory (LSTM) network with wind speed components and wind turbines states. Based on the filtering characteristics of LSTM, the wind speed components with different resolutions are screened automatically. These components were generated by wavelet packet decomposition (WPD) and ensemble empirical mode decomposition (EEMD). The forecasting results show that wind speed components and wind turbines states be used as the network input. The prediction accuracy can be improved. The error index will be smaller when wind turbine states was considered. Considering yaw error and rotor speed of wind turbine states, the mean absolute percentage error at the 5 s time interval is 2.72%