Volume 65

Day-Ahead Electricity Price Forecasting with Weather Consistency via Meta-Path Attention in GNN-Transformer-BiLSTM Author Zaiqing Ren1, Author Hongtao Wei2, Author Baorui Zhang3, Author Chongyi Tian 4*

Abstract

Accurate electricity price forecasting is crucial for optimizing power market operations and trading strategies. In this study, we propose a hybrid model that integrates Graph Convolutional Networks (GCN), meta-path attention, and a Transformer-BiLSTM architecture to capture both spatial correlations among heterogeneous power system features and temporal dependencies in electricity price sequences. The GCN module models feature interactions across multiple sources, the meta-path attention mechanism adaptively emphasizes important relational paths, and the Transformer-BiLSTM combines global sequence encoding with bidirectional temporal modeling. Experimental results demonstrate that the proposed method achieves an accuracy of 85.5%, corresponding to a 14.3% improvement over the second-best baseline, highlighting its effectiveness and superiority in electricity market scenario forecasting.

Keywords Electricity price forecasting, Deep learning, Graph Convolutional Network, Meta-path Attention