A novel hybrid spatiotemporal prediction algorithm of solar radiation based on topology and deep learning models

Abstract

Listen

Translate article

As a renewable energy source, solar energy has become essential to alleviate traditional energy scarcity because of its environmentally friendly characteristics. The intermittence and instability of solar energy bring challenges to energy utilization. Accurate solar radiation prediction models are necessary for the economical and reliable operation of building energy systems. Most existing models predict solar radiation based solely on time-dependent features and neglect the influence of spatial information, resulting in inferior prediction results. A novel hybrid spatiotemporal prediction model based on the graph attention network and bi-directional gated recurrent unit (GAT-BiGRU) is proposed in this paper. The GAT-BiGRU algorithm innovatively adopts the GAT to explore the spatial dependence of solar radiation from the graph topology, while the BiGRU is being applied to capture the temporal dynamic features of solar radiation. Detailed prediction experiments were conducted for the solar radiation of 16 districts in Tianjin. Compared with the state-of-the-art algorithms, the results indicate that the proposed model has a more robust generalization ability and minor prediction errors. Thus, it is expected to provide an effective reference for the actual operation strategy of solar building energy systems.