M2GSNet: Multi-Modal Multi-Task Graph Spatiotemporal Network for Ultra-Short-Term Wind Farm Cluster Power Prediction

• DOI: 10.3390/app10217915
• Publication Date: 2020-11-09T00:03:37Z
• Authors: Hang Fan; Xuemin Zhang; Shengwei Mei; Kunjin Chen; Xinyang Chen
• Keywords: Technology; QH301-705.5; T; Physics; QC1-999; graph neural network; multi-task learning; 02 engineering and technology; Engineering (General). Civil engineering (General); 7. Clean energy; wind power prediction; Chemistry; wind farm cluster; 0202 electrical engineering, electronic engineering, information engineering; NWP; multi-modal learning; TA1-2040; Biology (General); QD1-999

Ultra-short-term wind power prediction is of great importance for the integration renewable energy. It foundation probabilistic and even a slight increase in accuracy can exert significant improvement safe economic operation systems. However, due to complex spatiotemporal relationship intrinsic characteristic nonlinear, randomness intermittence, regional farm clusters each farm’s still challenge. In this paper, framework based on graph neural network numerical weather (NWP) proposed ultra-short-term prediction. First, adjacent matrix farms, which are regarded as vertexes graph, defined geographical distance. Second, two networks designed extract feature historical NWP information separately. Then, these features fused multi-modal learning. Third, enhance efficiency method, multi-task learning method adopted common cluster it output at same time. The cases located Northeast China verified that improved, time consumption increases slowly when number farms grows. results indicate has potential be used large-scale clusters.