Optimization of Price Difference Arbitrage in Electricity Spot Market

Overview

Developed an XGBoost-DRO framework for optimizing price difference arbitrage strategies in the electricity spot market for power retail companies.

Price Forecast Modeling

Constructed an XGBoost regression model to predict day-ahead and real-time electricity prices in the spot market.

• Feature engineering: Incorporated hourly and weekly data, day-ahead prices, 45-day rolling statistics (mean and standard deviation), and historical price difference lags.
• Results: MAE and RMSE for day-ahead price prediction were 51.75 and 70.62 respectively; real-time prediction achieved 107.76 and 154.89, significantly outperforming Prophet and Seasonal Naive baseline models.

Distribution Robust Optimization

Designed a Wasserstein-1 fuzzy set (radius ε=3.09) using historical price spread samples, with the quotation coefficient α as the decision variable. Introduced a variance penalty term to avoid overly conservative solutions.

A dual-layer quotation structure (premium of +20 yuan/MWh and discount of -30 yuan/MWh) was designed to lock in base load while achieving partial real-time market exposure, balancing returns and tail risks.

Strategy Backtesting

Using Q1 2026 data, calculated total returns, annualized Sharpe ratios, and maximum drawdowns:

• Three-month cumulative return: RMB 11.3184 million
• Annualized Sharpe ratios: 3.67 (January), 4.56 (March)
• Validated the effectiveness of the XGBoost-DRO framework in practical order execution mechanisms