EURO 2025 Leeds
Abstract Submission

1361. An interpretable machine learning approach to day-ahead extreme electricity price forecast with imbalanced data

Invited abstract in session TC-34: Advancements of OR-analytics in statistics, machine learning and data science 4, stream Advancements of OR-analytics in statistics, machine learning and data science.

Tuesday, 12:30-14:00
Room: Michael Sadler LG10

Authors (first author is the speaker)

1. Jingchuan Ma
Alliance Manchester Business School, The University of Manchester
2. Yu-wang Chen
Alliance Manchester Business School, The University of Manchester
3. Fanlin Meng
University of Exeter Business School

Abstract

In recent years, the frequency and magnitude of extreme electricity prices have increased due to the growing share of renewable energy and external market dynamics, challenging market participants. Accurate forecasting of such extremes is crucial for market stability and risk management. To address this, we first introduce a Dynamic Weighted Threshold method for extreme price identification, enabling adaptive threshold adjustments based on market conditions.

Furthermore, we propose a Weighted-XGBoost model to predict extremely high and low electricity price occurrences in an imbalanced data context. Compared to baseline models, it demonstrates superior predictive performance. To enhance interpretability, SHAP (SHapley Additive exPlanations) is used to analyse feature contributions. Results indicate that extremely high prices are driven by complex interactions among supply-demand conditions, fossil fuel prices, and historical market behaviors. In contrast, extremely low prices are primarily influenced by forecasted residual load, highlighting a more deterministic supply-demand relationship.

By integrating dynamic extreme price identification, advanced forecasting, and explainability techniques, this study improves understanding of extreme price dynamics. The proposed approach is adaptable to various electricity markets, offering valuable insights for market operators and participants aiming to enhance risk management and forecasting accuracy.

Keywords

Status: accepted

Back to the list of papers