Research on boundary data state estimation techniques in electricity spot markets based on deep learning

Aiming to address the problems of knowledge conflict and insufficient physical constraints caused by the heterogeneity of boundary data in the electricity spot market, this paper presents a novel deep learning-based method for boundary data state estimation. Specifically, we propose a Power Estimation and Correction Engine (PEACE), which integrates three key components: a Power Data Multimodal Encoder (PDME), a Power System Mixture of Experts Module (PSMoE), and a Physical Constraint Guidance Module (PCGM). The PDME employs heterogeneous graph representation learning to unify multimodal data. The PSMoE uses an expert network to adaptively fuse information from different modalities based on the characteristics of the input data. The PCGM ensures that the model’s output adheres to the fundamental physical constraints of the power system. The proposed method effectively addresses knowledge conflicts caused by data heterogeneity and ensures the physical feasibility of the estimated states. Experimental results demonstrate the effectiveness of PEACE in providing accurate and reliable boundary data state estimates for the power spot market.