Electricity Spot Market Risk Identification and Multi-Dimensional Early Warning Based on Improved Transformer
Reliable risk identification and early warning in electricity spot markets are increasingly important for intelligent power communication infrastructures and electromagnetic information systems that support real-time monitoring and dispatching. To address the challenges posed by high-frequency, heterogeneous, and strongly coupled market data, this study proposes an improved Transformer-based framework for electricity spot market risk identification and multidimensional early warning. The proposed method integrates multi-source operational data through temporal embedding and multi-channel feature encoding, enabling effective representation of electricity price fluctuations, dispatch instructions, transaction behaviors, and load responses. An enhanced self-attention mechanism is employed to capture long-range temporal dependencies and dynamic interactions, while a lightweight classifier performs hierarchical risk categorization. Furthermore, a multi-dimensional warning mechanism incorporating spatial location, subsystem characteristics, and temporal risk evolution is established to support adaptive operational decision-making. Experimental results demonstrate an overall risk identification accuracy of 93.2%, an average F1-score of 0.88, a dispatch subsystem localization accuracy of 97.3%, and a trend identification accuracy of 94.2% with only 3.1 minutes of response delay during risk escalation. The proposed framework provides an effective solution for intelligent risk perception and dynamic monitoring in complex electricity markets and offers methodological insights for communication-enabled energy systems and electromagnetic sensing environments requiring reliable real-time information processing.