A whale optimization algorithm-based ensemble model for power consumption prediction

As one of the indispensable energies, electricity plays an essential role in modern society. Accurate prediction of power consumption is crucial for managing energy generation and purchasing, preventing overloading, and ensuring energy generation and purchase for preventing overloading and having efficient energy storage. In this paper, we propose a whale optimization algorithm-based ensemble learning algorithm (WOA-ELA) for multi-output power consumption prediction using a weighted ensemble approach. Firstly, we optimize the hyper-parameters of the base estimators using the WOA. Each base estimator's performance is evaluated through fivefold cross-validation to mitigate the influence of data splitting during model training. After obtaining the trained estimators, we further optimize the weights that use to ensemble these base estimators for the final prediction. These weights are designed to emphasize the strengths of high-performing estimators while mitigating the impact of weaker ones. Given the three outputs of power consumption, our target is to minimize their mean absolute error (MAE) using the WOA. Experimental results show that our proposed WOA-ELA outperforms the comparative models in predicting the power consumption. The average MAE achieved by our proposed WOA-ELA surpasses the four comparison models by 19.14%, 17.17%, 10.53%, and 2.15%, respectively. Additionally, its average root mean squared error is better by 15.12%, 18.50%, 7.16%, and 1.16% compared to the four models. This indicates that the use of WOA is effective for improving the prediction performance by optimizing the hyper-parameters of each model and the weights for the ensemble operation. With the promising results, our WOA-ELA can be a useful tool for energy management systems in multi-output power control scenarios.