A Comparison of Deep Learning vs Traditional Machine Learning for Electricity Price Forecasting

With the advent of Industry 4.0, smart technologies and an ever-increasing need for efficient energy management, more accurate forecasts of electricity prices can be exploited to increase the profits of those involved in real-time pricing or auction-based energy market contracts. This paper provides a state-of-the-art comparison of day-ahead forecasting models including both Deep Learning and traditional machine learning models such as Random Forest, Support Vector Machines, etc. This research also presents a novel price forecasting model: Capsule Networks (CapsNets) using dynamic routing. The examined models are appraised on their ability to predict the hourly electricity price schedule for the trading day period. Data from the Irish electricity market is used to demonstrate the feasibility of the proposed methods. It is demonstrated that a K-Nearest-Neighbors model produces forecasts that are more accurate than any of the Deep Learning models examined. CapsNet models are shown to outperform other CNNs but fail to match the traditional ML models examined. Furthermore, the deep learning methods including CapsNets require substantially greater training time.

Automated summary

With the rise of Industry 4.0 and the increasing demand for efficient energy management, more accurate electricity price forecasts can boost profits for those involved in real-time pricing or auction-based energy market contracts. This research compares various day-ahead forecasting models and finds that the K-Nearest-Neighbors model produces the most accurate forecasts, while deep learning methods including CapsNets require longer training time and do not perform as well as traditional machine learning models.