基于并行异构变换器与多模态融合的电力市场异常操纵识别方法OA

Aimed at the regulatory challenges in electricity markets arising from the energy transition and the entry of di⁃verse emerging entities,the limitations of existing approaches in accurately identifying covert violations are highlight⁃ed,and an abnormal manipulation behavior identification method in electricity markets based on a parallel heteroge⁃neous transformer architecture and multi-modal fusion is proposed to ensuring fair market operations.First,the Grami⁃an angular field coding technology is employed to transform the 1D electricity price and load series into 2D Gramian an⁃gular summation fields images,so as to realize spatial modal mapping of features.Second,a dual-branch parallel frame⁃work is constructed,which uses a 1D transformer and a vision transformer to extract temporal dependencies and spatial texture features,respectively.In addition,a dynamic gated fusion mechanism is used to realize the dynamic alignment and fusion of the heterogeneous data.Furthermore,an improved focal loss function is introduced to mitigate the class imbalance caused by abnormal sample sparsity.Experimental results based on the real data from Shandong Province demonstrated that the proposed method achieved a significantly higher F1 score than the baseline model,effectively im⁃proving the accuracy of identifying the covert manipulation behaviors.Specifically,its decision logic highly aligns with the market physical mechanisms,providing robust intelligent decision support for the regulation of next-generation pow⁃er systems in electricity markets.