Forecasting volatile electricity prices in the Australian National Electricity market using machine learning
Abstract The growing integration of variable renewable energy Sources have intensified price volatility in the Australian National Electricity Market (NEM), creating operational challenges for market operators, traders, and battery energy storage systems (BESS). Short-term electricity price forecasting is important for arbitrage, market operation, and participation in ancillary services such as Frequency Control Ancillary Services (FCAS). The evaluation presents a Comparative evaluation of eight forecasting models: Autoregressive Integrated Moving Average (ARIMA), Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX), linear regression, exponential smoothing, XGBoost, Long Short-Term Memory (LSTM) networks, and Prophet with and without an exogenous demand regressor, using historical price data from New South Wales (NSW). Five-minute-ahead forecasting performance is assessed across selected stable, volatile, and rapidly fluctuating periods. XGBoost achieves the lowest mean validation MAE (0.040) and highest mean validation $$R^2$$ (0.949) among the evaluated models, and retains the best performance on the held-out test set. The findings support the use of market information in the evaluated forecasting models.
Authors
- Jason M. West (ORCID: https://orcid.org/0000-0003-3271-3155)
- Zeyang Zhou (ORCID: https://orcid.org/0009-0007-7788-6339)
Institutions
- University of Technology Sydney (AU)
- Bureau of Meteorology (AU)
Publication Details
- Journal
- Energy Systems
- Published
- 2026-11-14
- DOI
- https://doi.org/10.1007/s12667-026-00853-9
- Primary Topic
- Energy Load and Power Forecasting
- Type
- article
- Field-Weighted Citation Impact
- 0.00