Combination of Metaheuristic Algorithm and Artificial Neural Networks Model to Forecast Wind Energy

The transition to renewable energies, in particular wind power, is essential to meeting environmental challenges and ensuring a sustainable future. Precise estimation of wind energy production is essential for maximizing its incorporation into electrical systems, therefore enhancing overall dependability. In this context, this study focuses on wind power forecasting at the Scotland Penmanshiel wind farm. The wind power forecasting process with the proposed particle swarm optimization- artificial neural network (PSO-ANN) method begins with preprocessing a dataset containing wind speed, wind direction, and energy production data to handle anomalies and normalize the features. Next, an artificial neural network (ANN) model architecture is defined, and key hyperparameters are optimized using particle swarm optimization (PSO) to minimize forecasting error. The performance of the PSO-optimized ANN model is evaluated using metrics such as root mean square error (RMSE), coefficient of determination (R²), and weighted index (WI) to assess its predictive accuracy. By comparing forecasted wind power values with actual observations and analyzing forecast errors, insights are gained into the model's ability to capture wind patterns and variability. Evaluation metrics, including root mean square error (RMSE=94.98), coefficient of determination (R²=0.98) and weighted index (WI=0.03), demonstrate the effectiveness of the PSO-ANN model in capturing diverse wind patterns. The results highlight the significance of this hybrid approach in achieving precise short-term wind power forecasts, emphasizing the complementary strengths of PSO in optimizing the intricate architecture of ANN for improved predictive performance. As a perspective, this paper aims to be a starting point for further studies that combine several meta-heuristic algorithms in order to make accurate medium- and long-term forecasts.