Data Anomaly Detection and Predictive System for Energy Management

Wong, Jun Yong (2026) Data Anomaly Detection and Predictive System for Energy Management. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text WONG JUN YONG_Full Text.pdf Restricted to Registered users only Download (2MB)

Abstract

This research explores the use of advanced hybrid deep learning models for power consumption forecasting in smart grids, aiming to improve prediction accuracy. Traditional machine learning techniques, such as Random Forest, have limitations in handling complex power consumption data, especially in the context of modern smart grids. The study integrates Recurrent Neural Networks (RNNs), particularly Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs), with Convolutional Neural Networks (CNNs) by using python 3 to enhance forecasting performance. The research compares the performance of hybrid models with traditional machine learning methods and single deep learning models. Using data from residential power consumption, the models were evaluated based on Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The results reveal that LSTM outperformed other models, including attention-based models, across all evaluation metrics, achieving significant error reduction—approximately 90% improvement in RMSE and MAE compared to Random Forest. The study highlights the effectiveness of LSTM for forecasting power consumption, suggesting that simpler recurrent models, such as LSTM, offer better results than more complex hybrid models, particularly in smaller datasets. The findings support the adoption of deep learning for more accurate and reliable energy management in smart grid applications