科研之友
微信
新浪微博
Facebook
分享链接
摘要
State prediction of wind turbines is a key aspect of wind power’s digitalization as well as its intelligent operation and maintenance (O&M). Deep learning has been gradually applied to the state prediction due to its powerful potential in mining the hidden relationships of complex and high-dimensional data; however, it also has some practical limitations such as poor inference and interpretation. Therefore, effectively combining domain knowledge with intelligent algorithms gives an important direction to the intelligent O&M. In this paper, we propose a knowledge-embedded graph neural network (K-GNN) model based on the general framework of multivariate time series graph neural network (MT-GNN), so as to predict the multivariate time series state data of wind turbines. We combine the knowledge embedding module and the auto-graph-learning module to better describe the relationships among state variables by embedding three knowledge matrices, i.e. correlation, causality and expert experience. The results demonstrated that among the three types of K-GNN models, the one with the embedded expert experience matrix performs best in prediction, indicating that domain knowledge can effectively improve the generalizability and interpretability of GNN models. It is also believed that the demonstrated work is valuable to the R&D and promotion of wind power predictive maintenance techniques.
- 单位
