Short-term Wind Speed Prediction Based on GRU

Abstract

With the rapid increase of wind power grid-connected capacity, improving wind speed prediction accuracy is increasingly important to reduce wind power fluctuations and to the safe and stable operation of power systems. Due to the fluctuations and randomness of wind speed, traditional prediction methods are difficult to predict wind speed accurately. In this paper, a wind speed prediction model based on Gated Recurrent Unit network is built under the deep learning framework of Google-Tensorflow. The appropriate structure and other hyperparameter of the GRU network model are determined through experiments. The model has two hidden layers. The time series data of wind speed are modeled dynamically, and the network parameters are trained by time reverse error propagation algorithm. The experiment uses the measured wind speed data of a wind farm in Xinjiang Province to verify the simulation results. The results show that the proposed model has higher prediction accuracy than the ARMA model and the LSTM model.