Improving Energy Forecasting and Anomaly Detection in Multivariate Environments

Vavilala, Parashuram (2025) Improving Energy Forecasting and Anomaly Detection in Multivariate Environments. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (910kB) | Preview

Abstract

This research presents a hybrid deep learning approach for multivariate energy time series forecasting and anomaly detection using an Attention-Based GRU Autoencoder. The model is designed to effectively capture temporal dependencies while prioritizing important features through attention mechanisms, making it robust under data imbalance conditions. A comprehensive comparison was conducted across classical machine learning classifiers and deep learning autoencoder models. Experimental results on a real-world imbalanced energy dataset demonstrate that the proposed model significantly outperforms baseline models in detecting rare anomalies, achieving the highest recall (0.85) and F1 score (0.85) for the anomaly class. The results confirm the feasibility of attention-enhanced GRU networks as a robust way of detecting anomalies in complex energy systems.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Rustam, Furqan UNSPECIFIED
Subjects:	H Social Sciences > HC Economic History and Conditions > Natural resources > Power resources > Energy consumption Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	27 Aug 2026 09:14
Last Modified:	27 Aug 2026 09:14
URI:	https://norma.ncirl.ie/id/eprint/9680

Actions (login required)

View Item