Predictive Analytics and Machine Learning for Electricity Consumption Resilience in Wholesale Power Markets

This article presents the research results on creating prediction models using historical data on projected power usage in an area with many sectors. Given the constantly high energy intensity of any critical sector, it is imperative to prioritize the optimization of power use. A method to enhance the precision of managing energy expenses in the planning phase involves anticipating electrical loads. Although there is a wealth of scientific study on energy consumption prediction, it continues to be a significant problem because of the evolving demands of the wholesale electricity and power market, which require precise forecasts for resilience. This study aims to improve managerial decision-making through strategic power consumption planning. The approach involves constructing prognostic models based on historical data, including power consumption, system performance metrics, and meteorological data. The study achieves highly accurate short-term power consumption predictions using ensemble techniques like random forest, gradient boosting (XGBoost, CatBoost), and intelligent models. Incorporating gradient boosting with neural network models results in forecasts with minimal error rates, demonstrating the models' suitability for predicting integrated power system electricity consumption.