A Two-Step Locational Marginal Price Forecasting Method in Spot Market with Poor Data

Accurate real-time electricity price forecasting is crucial for market regulators to maintain stable market operations, guide market participants in formulating bidding strategies and mitigate market participation risks. Recently, many data-driven artificial neural network algorithms have been widely applied to real-time electricity price forecasting due to their excellent capability of feature extraction and nonlinear approximation. However, their requirements for high-quality and large-scale data also pose new challenges in emerging electricity markets such as China's. For example, in China's Hubei Province, the electricity spot market is almost in the trial phase and frequent policy adjustments have led to widespread gaps and interruptions in historical information data. As a result, price forecasting models based on artificial neural networks struggle to be effectively trained, limiting their ability to provide robust decision support for bidding quantity and pricing. To address this problem, we propose a two-step method combining the similar moment method with machine learning algorithms for real-time clearing locational marginal price forecasting, which is less susceptible to the impact of low-quality data. First, the raw features are compressed with an autoencoder to improve the effectiveness and efficiency of similar moments selection. Subsequently, a multiple-output support vector regression model is introduced to derive locational marginal price forecasts based on information from both the day-ahead clearing and similar moments. Finally, an extensive experiment based on real-world data from the Hubei province's electricity spot market demonstrates the effectiveness of the proposed method.