Comparison of Statistical Time Series and Machine Learning Modelling in Forecasting Electricity Consumption and Production

Chin, Wei Lun (2025) Comparison of Statistical Time Series and Machine Learning Modelling in Forecasting Electricity Consumption and Production. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text RMM_Chin Wei Lun_Full Text.pdf Restricted to Registered users only Download (2MB)

Abstract

The increasing demand for efficient energy management necessitates accurate forecasting of electricity consumption and production. This study compares the performance of statistical time series analysis and machine learning models in predicting electricity consumption and production in Romania from 2019 to 2024. The research aims to evaluate the strengths and limitations of each approach, focusing on their ability to capture trends, seasonality, and long-term dependencies. The study employs SARIMA, Holt Winter, and Kalman Filter for time series forecasting, while LSTM, XGBoost, Random Forest for machine learning modelling, and also a hybrid LSTM-SARIMA approach to compare performance. Historical electricity data is collected, preprocessed, and analysed to identify key patterns. The performance of models is assessed using evaluation metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Results indicate that while SARIMA effectively captures seasonality and short-term trends, LSTM demonstrates superior performance in modelling long-term dependencies and handling non-linear patterns. However, LSTM models tend to smooth out fluctuations, making them less effective in capturing sudden demand spikes. The findings provide insights into the practical applications of each methodologies, offering recommendations for improving energy forecasting and supporting sustainable energy management strategies. Keywords: Electricity Forecasting, Energy Management, Time Series Analysis, Machine Learning, LSTM, SARIMA, Hybrid LSTM-SARIMA, Holt Winter, XGBoost, Random Forest, Kalman Filter.