Recurrent Neural Network for Chaotic Wind Speed Time Series Prediction

Accurate wind speed forecasting is essential for wind energy systems, but it's difficult due to the wind's uncertainty. In this study, short-time wind speed prediction is done on the basis of phase space reconstruction and using neural network techniques. The NREL (National Renewable Energy Laboratory) site collects the wind speed data. Firstly, the chaotic nature of the time series is ascertained to verify its short time predictability. Then training data is created by using the phase space reconstruction technique, i.e., using knowledge of embedding delay and embedding dimension. Using this data, two models of Artificial neural networks, i.e., Feedforward neural network (FFNN) and Recurrent Neural network (RNN) are trained to predict wind speed. Training accuracy of each model is compared on the basis of Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE) & Symmetric Mean Absolute Percentage Error (SMAPE) metrics. Simulation results show the superiority of RNN (Recurrent Neural Network) over FFNN (Feed Forward Neural Network) for the prediction of wind speed.