Wind Speed Forecasting using Artificial Neural Networks: A Comparative Study

This paper describes the design and implementation of an artificial neural network model based on a Levenberg-Marquardt algorithm with Nonlinear Autoregressive (NAR) model and Nonlinear Autoregressive with Exogeneous inputs (NARX) model, for the purpose of predicting wind speed values. The dataset used in this paper was collected from NSRDB (National Solar Radiation database) from Jaisalmer India. Total of 8760 datasets consisting of ten, time and meteorological parameters were used to train and validate the networks. The results obtained show that a NARX neural network with a Levenberg-Marquardt (LM) algorithm is an effective technique for predicting wind speed. The results also demonstrate that the number of neurons in the hidden layer, the number of input and output lags, and the learning rate had a positive effect on the performance of the neural network, although increasing them did not increase the performance beyond a certain threshold.