Malfunction Diagnosis of Microgrid Devices Based on Optimized Graph Neural Network

Malfunction diagnosis based on deep learning ensures the stable operation of microgrid. However, the microgrid system with new energy access appears to be increasingly complex and changeable, and its frequently changing topology brings challenges to the training of traditional neural network models. A method for malfunction diagnosis of microgrid devic-es based on optimized Graph Neural Network is proposed in the paper. Firstly, data reconstruction is completed through the mapping of nodes, edges and graphs in order to standardize changeable data. Secondly, abnormal data is processed to improve the quality of database. In that way, device misoperation caused by extreme data loss can be avoided, thus im-proving the training efficiency. Then, aiming at fully capturing the characteristics of time and space dimensions of data at different latitudes, the graph-convolution network under Multi-head Attention mechanism is adopted in the process. Finally, a microgrid simulation model is built to prove that the proposed scheme has the advantages of good robustness, strong adaptability, high reliability and accurate diagnosis.