Improving accuracy of wind speed forecasted with power law and multiple layer perceptron: a case study of Aneho, a coastal area of Togo

Wind speed forecasting is critical for optimizing wind energy generation, especially in regions with complex and variable weather patterns. This study focuses on improving the accuracy of wind speed forecasting using Multiple Layer Perceptron (MLP) models, with specific application to Aneho, a coastal area in Togo. The Power Law (PL) extrapolation or interpolation method is commonly used to estimate wind speeds at different heights. However, PL tends to show significant performance variability, particularly in periods with complex wind conditions. This study aimed to assess the accuracy of the PL method and enhance it using MLP for bias correction. We used wind speed data at 50 m, 80 m and 100 m as validation points for both PL and MLP methods, while predictions for non-GFS gridded points at 55 m, 65 m, 75 m, and 85 m were generated and analyzed. The MLP model was trained using 50 m wind speed data and validated against the 80 m and 100 m heights, achieving significant error reductions compared to the PL method. For instance, in December, the MLP-corrected wind speeds showed substantial improvements, with RMSE values dropping below 0.14 m/s and R 2 values approaching 1.0, demonstrating a near-perfect fit between predicted and true values. On the other hand, August, the worst-performing month for PL, exhibited an RMSE as high as 2.95 m/s, which was significantly reduced after MLP correction. This study highlights the limitations of PL interpolation, particularly in months with chaotic wind patterns, and demonstrates the effectiveness of MLP in mitigating these errors. The combination of PL and MLP for bias correction offers a robust approach for wind speed forecasting in complex environments, providing more accurate data for heights not covered by GFS. These findings are essential for improving wind energy planning and operations in coastal regions in Togo.