Short-term Forecast of Price Spread between Day-Ahead and Real-Time Markets using Ensemble Learning and Global Sensitivity Analysis

With the continuous development of the highly competitive electricity sales market, short-term forecasting plays an increasingly important role. For market participants, forecasting the price spread between the day-ahead market and the real-time market can help them formulate trading strategies and better cope with uncertainty. Currently, various methods have been applied to the price spread forecasting with varying degrees of success, however, most existing price spread forecast methods rarely evaluate the impact of different feature quantities on the output results. This paper proposes a short-term forecast framework for the price spread between day-ahead and real-time markets, which is based on an ensemble learning technique combined with global sensitivity analysis. The categorical boosting (CatBoost) model is employed to implement the high-precision forecast of price spread between the day-ahead market and the real-time market. Then, the global sensitivity analysis based on Sobol method is leveraged to analyze the impact of the weight of each feature on the output results. Finally, the effectiveness of the proposed forecast framework is verified by using real-world electricity market data from a certain region in China.