Exploring optimization strategies for support vector regression networks in predicting power consumption

The present examination introduces a promising approach for efficiently predicting power consumption using empirical data from a well-established related study. Utilizing support vector regression (SVR) networks, the training process is enhanced by employing various optimizers, including the salp swarm model, PSO, GWO algorithm, and simplified memory bounded a star. Additionally, K-fold cross-validation is used to mitigate overfitting. Multiple statistical metrics are computed to evaluate the effectiveness of different metaheuristic optimization algorithms. Results indicate that the performance of the selected optimizers, as well as SVM without optimization, is satisfactory. However, the SVR–SMA approach emerges as the most accurate and expedient method for power consumption prediction, demonstrating superior performance according to numerical and empirical data. Specifically, the SVR–SMA model achieves maximum R² index scores of 0.9902 and 0.9904 for training and test data. The key innovation of the utilized research lies in the proposed prediction model, which rivals other contemporary approaches in the literature and, at times, surpasses them.