Comparison of multilayer perceptron and nonlinear autoregressive models for wind speed prediction

Abstract

Wind energy is a critical component of the global shift to renewable energy sources, with significant growth driven by the need to reduce carbon emissions. Accurate wind speed prediction is crucial for increasing wind energy output since it directly influences wind farm design and performance. The purpose of this study is to compare two artificial neural network (ANN) models for predicting wind speed in Dakhla City, a place with a high wind energy potential. The first model is a multilayer perceptron (MLP) trained with the backpropagation algorithm, while the second is a nonlinear autoregressive with exogenous inputs (NARX) model, a form of recurrent neural network (RNN) noted for its ability to handle time-series data more well. The comparative analysis results show that the NARX model outperforms the MLP model in terms of wind speed forecast accuracy. The NARX model achieved a near-perfect regression coefficient (R) of 0.9998 and a root mean square error (RMSE) of 1.02899, indicating that it can represent complex, nonlinear wind speed patterns. These findings indicate that the NARX model could be a beneficial tool for increasing the efficiency of Dakhla City’s wind energy infrastructure, assisting the region in meeting its renewable energy ambitions.