Energy Demand Forecasting for Extreme Weather Conditions with Artificial Intelligence

Abstract

Accurate energy demand forecasting is essential for ensuring the stability and efficiency of power systems, particularly under extreme weather conditions. This research develops a dynamic, multi-layered forecasting approach that integrates Isolation Forest for anomaly detection and XGBoost for demand prediction and continuously learns from incoming data.By providing a more responsive and context-aware system, the proposed approach mitigates forecasting errors during extreme weather events, addressing key limitations in traditional forecasting methodologies.The system operates through a layered, multi-model forecasting structure, where Isolation Forest detects weather anomalies and informs the demand prediction model, allowing it to adjust its forecasts based on the evolving weather conditions. The system is designed to capture complex relationships between weather patterns and energy consumption,improving forecast accuracy across multiple time horizons, from one day ahead to fourteen days ahead.Beyond methodological advancements, this study addresses key gaps in existing forecasting literature, particularly the lack of models explicitly designed to incorporate weather anomalies into energy demand prediction. Utilizing datasets from the Austin, TX, USA area, the findings demonstrate that this approach outperforms the current modelling practices at the Electric Reliability Council of Texas (ERCOT), the organization responsible for managing the electrical grid in Texas. This study provides a scalable forecasting system, designed to be applicable, adaptable, and extendable across different regions and energy forecasting applications.

Date of Award	Mar 2025
Original language	English
Awarding Institution	Aston University
Supervisor	Leonidas Anastasakis (Supervisor) & Roberto Alamino (Supervisor)