A load forecasting method based on heterogeneous network for integrated electricity and heat systems

Abstract

Accurate load forecasting in Integrated Electricity and Heat Systems (IEHS) is crucial for optimizing energy utilization and ensuring system stability. However, existing models struggle to capture the complex interdependencies between electricity and heat networks, leading to suboptimal predictions, particularly at coupling nodes. This study proposes a Heterogeneous Graph Attention Network-Gated Recurrent Unit (HAN-GRU) model within a Heterogeneous Network (HetNet) framework to enhance forecasting accuracy. By leveraging node-level and semantic-level attention mechanisms, HAN-GRU effectively integrates heterogeneous spatial dependencies and temporal correlations, prioritizing critical coupling nodes and optimizing information aggregation. The proposed model is validated on the Bali integrated energy system, as well as a 23-node radial heating network coupled with IEEE 33-bus and IEEE 118-bus power distribution networks. Simulation results demonstrate that HAN-GRU significantly outperforms baseline models in both accuracy and efficiency, highlighting its potential as a scalable and interpretable solution for multi-energy system forecasting.