A Deployment-Oriented Extreme Learning Machine for Electricity Price Forecasting

In electricity price forecasting research, deployment is often an afterthought. The literature has largely emphasized computationally expensive deep learning architectures, creating a disconnect between the needs of smart grid operators and resource-intensive approaches. In addition, prior studies often depend on non-public datasets or suffer from data leakage, which limits reproducibility and claims of real-time applicability. We address this gap by demonstrating that lightweight models can exceed both operational baselines and a long short-term memory (LSTM) alternative, while enabling practical edge deployment. Using the public archives of the Independent Electricity System Operator (IESO), we develop an Extreme Learning Machine (ELM) for multi-horizon Hourly Ontario Energy Price (HOEP) forecasting that predicts <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1-3$</tex> hours ahead simultaneously. Our model outperforms IESO's predispatch forecasts by an average of 21% across the three horizons and the LSTM baseline by 2.4%. Additionally, our ELM trains <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$10 \times$</tex> faster than the LSTM baseline and achieves sub-5 ms inference on low-cost edge hardware (Raspberry Pi 4). These results demonstrate that deployment considerations need not compromise forecasting accuracy, enabling adoption in edge device environments.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.