Air pollution impact on forecasting electricity demand utilizing CNN-PSO hyper-parameter optimization

Abstract

Electricity consumption is expected to increase considerably in the next few years, so forecasting and planning will become more important. A new method of forecasting electricity loads based on air pollution is presented in this paper. Air pollution indirect effects are not incorporated in current evaluations since they rely primarily on weather conditions, historical load data, and seasonality. The accuracy of electricity load forecasting improved by incorporating air pollution data and its potential effects, especially in regions where air quality heavily impacts energy consumption and generation patterns. This robust prediction model is capable of capturing the complex interactions between air pollution and electricity load by integrating innovative environmental factors with historical load data, weather forecasts, and other features. As part of the second contribution, we use metaheuristic algorithms to optimize hyper parameters, which provide advantages such as exploration capability, global optimization, robustness, parallelization, and adaptability making them valuable tools to improve machine learning models' performance and efficiency. The study found that the correlation coefficient (R) between predicted and real electricity demand and supply was high, at 0.9911. Beyond that this approach reduces MAPE by up to 19.5% when CNN and particle swarm optimization are combined with utilizing innovative air pollution variables. As a result, the optimization results were robust compared to state-of-the-art results based on statistical metrics such as RMSE and MAPE. Lastly, we emphasize the importance of factoring in air pollution effects when forecasting and managing electricity loads; future research directions include developing integrated modeling frameworks that reflect the dynamic interaction between air quality, energy consumption, and renewable energy generation.