Short-term Wind Power Prediction Based on LSTM Considering Power Fluctuation Characteristics

Good wind power predictions of wind farms can provide data support for system planning and operation scheduling. In this paper, a short-term wind power prediction method is proposed based on long short-term memory (LSTM) neural network, considering power fluctuation characteristics to meet this demand. First, this paper takes the historical data of numerical weather prediction (NWP) including wind speed, wind direction, temperature, humidity and air pressure into consideration. The correlation coefficient between meteorological factors and wind power is calculated to reduce the dimension of high-dimensional meteorological data. Then based on the fluctuation and trend of wind power output, the 90% confidence interval is adapted to describe the volatility of wind power output. It plays a preliminary screening role in predicting training data to ensure prediction accuracy. Finally, based on the LSTM neural network prediction model, the filtered data is trained. Further, the mapping relationship between meteorological factors and wind power output is established. Through the prediction and analysis of the power output data of a measured wind farm, a better prediction effect can be attained.