Short-Term Prediction of Wind Speed Using Machine-Learning Algorithms

The demand for electrical energy has sharply increased with the development of industry. However, supplying this demand through fossil fuels leads to problems such as global warming and environmental pollution. Considering the limitations and depletion of fossil fuel resources, finding sustainable alternatives has become essential. Among these alternatives, wind energy stands out as a viable renewable source for electricity generation due to its low cost and lack of pollution. However, to achieve stable power generation from wind farms, accurate information about future wind speed is essential. Predicting wind speed is highly challenging due to its random and intermittent nature. In this paper, a hybrid model combining a Temporal Convolutional Network and Bidirectional Long Short-Term Memory (TCN-BiLSTM) is employed to address this challenge and achieve accurate predictions. First, the hyperparameters of the Variational Mode Decomposition (VMD) algorithm were optimized using the powerful Optuna method. Next, the original wind speed data were normalized to enhance the performance of the hybrid model (TCN-BiLSTM) and then fed into the VMD algorithm to be decomposed into Intrinsic Mode Functions (IMFs). Each IMF is then individually fed into the hybrid model for prediction. Finally, the outputs are denormalized and combined to obtain the final result. Based on the evaluation of the hybrid model using statistical metrics, the results indicate that the proposed model achieves high accuracy. In this evaluation, the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE) were found to be 99.1%, 0.36, and 0.48, respectively.