Investigating an Ensemble of ARIMA Models for Accurate Short-Term Electricity Demand Forecasting

Abstract

This paper investigates the effectiveness of ensemble modelling for time series forecasting using Autoregressive Integrated Moving Average (ARIMA) models. In recent years, ensemble modelling has become a popular approach for improving forecasting accuracy by combining multiple models to achieve better performance than individual models. However, there is still limited research on the effectiveness of ensemble models for time series forecasting using ARIMA models. In this paper, we tested simple averaging of ARIMA models and investigate their performance in comparison to individual models. We conducted experiments using real-world datasets and evaluated the models' performance using metrics such as Mean Absolute Percentage Error, and Root Mean Squared Error. In this paper, we provided experiments on both short and long datasets to evaluate the performance of ensembled models compared to individual models. For the short datasets, our results clearly demonstrated the advantages of using ensembled models over individual models. The ensemble of models consistently outperformed the individual models in terms of accuracy. Our findings suggest that ensemble modelling can be a tool for time series forecasting and can provide improvements in accuracy. By leveraging the strengths of different models, ensemble models can effectively capture the underlying patterns in the data and make more accurate predictions.