Performance Comparison of Machine Learning Approaches for Predicting Wind Power Generation of Thambapavani Wind Farm in Sri Lanka

In recent years, wind energy has become an alternative eco-friendly source of power generation in the Sri Lankan energy landscape. The present study compares machine learning techniques to identify the most suitable wind power prediction model for the Thambapavani Wind Farm, the largest wind farm in Sri Lanka, constructed in 2021. After a comprehensive literature review, three models, namely Multiple Linear Regression (MLR), Artificial Neural Network (ANN), and Long Short-Term Memory (LSTM) were selected for this purpose. Wind speed, direction and generation hours with a 10 -minute interval serve as inputs, while wind energy production is the output. The dataset, from May 2021 to December 2023, was collected from the Department of Meteorology and Ceylon Electricity Board, Sri Lanka. The efficiencies of the models were computed using $R^{2}$ and RMSE (Root Mean Square Error) metrics. The accuracies of ANN and MLR models were found 48.35% and $\mathbf{8 2. 7 6 \%}$ respectively, and the experimental results demonstrated that the proposed LSTM model yields the best performance, achieving an accuracy of $\mathbf{9 9. 9 9 \%}$. These results provide important clues for improving energy generation methods and boosting the long-term development of energy requirements in Sri Lanka.