Improving short-term wind power forecasting in Senegal’s flagship wind farm: a deep learning approach with attention mechanism

Abstract

Accurate wind power forecasting assumes an important role in power system operation and economic planning, particularly in Senegal’s flagship wind farm, the largest in West Africa. The fundamental volatility, intermittent nature, and unexpected character of wind power make it difficult to maintain power system stability. To address these challenges, an attention mechanism-based deep learning model is proposed to anticipate wind power in the short term with the goal of improving forecasting accuracy. The dynamic shifts in the wind power dataset are first processed by convolutional neural networks to extract multi-dimensional features. After being extracted, the feature vectors are placed into a long short-term memory (LSTM) network by being transformed into a series structure. Next, to optimize and improve the forecast accuracy of the model, an attention mechanism is included by assigning distinct weights to each hidden layer in the LSTM network. Real operational wind power generation data from the wind farm is utilized to verify the effectiveness of the proposed method. The results show that the proposed method can successfully boost the forecasting accuracy of wind power with better performance compared to other machine learning and deep learning models. This study not only contributes to improving wind power generation management and power system operations in Senegal but also serves as a valuable reference for promoting renewable energy transitions across sub-Saharan Africa.