Wind Speed Forecasting Using Wavelet Analysis and Recurrent Artificial Neural Networks Based on Local Measurements in Singida Region, Tanzania

Abstract

High accuracy wind speed forecasting is essential for wind energy harvest and plays a significant role in wind farm management and grid integration. Wind speed is intermittent in nature, which makes the forecasting to be a big challenge. In the present study, three hybrid single-step wind speed forecasting techniques are proposed and tested by local measurement data in Singida region, Tanzania. The three techniques are based on Wavelet Analysis (WA), Back Propagation (BP) optimization algorithm, and Recurrent Neural Network (RNN). They are referred to as WA-RNN, BP-RNN, and WA-BP-RNN. The model results showed that WA-BP-RNN outperforms the other two proposed techniques, with minimum statistical errors of 0.56 m/s (BIAS), 6.89% (MAPE) and 0.53 m/s (RMSE). Furthermore, the WA-BP-RNN technique has shown highest correlation value of 0.95, which indicates that, the strength of a linear association between the observed and forecasted dataset of the wind speed. In addition, the deployment of the BP optimization algorithm in the proposed technique showed improvements of the model results.