GridTransformer: transformer-based credit rating with power grid data

With the rapid development of the digital economy and digital finance, enterprise credit risk management faces unprecedented challenges and opportunities. Traditional credit models rely on static financial data, which suffer from delayed updates, limited dimensionality, and low transparency, making them inadequate for modern credit evaluation demands for timeliness, diversity, and comprehensiveness. To address this, we propose GridTransformer, a novel multimodal credit scoring model that integrates high-continuity, wide-coverage, and reliable electricity data. Built on a tabular deep learning framework and enhanced with Transformer architecture, the model unifies static enterprise attributes and electricity time series via feature-level embeddings, and employs residual connections and attention mechanisms to capture complex feature interactions. Experimental results show that GridTransformer outperforms baseline models across multiple metrics, providing an efficient and scalable solution for multimodal data fusion in financial risk assessment.