Water motor rotating wheel state correlation analysis method based on multi-index graph attention network

The invention discloses a water motor rotating wheel state correlation analysis method based on a multi-index graph attention network. The method comprises the following steps: 1) utilizing a graph structure to carry out explicit modeling on a multivariate time sequence generated by operation of a water motor rotating wheel; 2) reconstructing a connection relation between nodes by using graph diffusion convolution to realize transmission and fusion of features; 3) combining the weighted graphs and learning attention weights among the nodes by using a graph attention network; and 4) introducing fastDTW pre-training and combining with double contrast loss learning, updating model parameters, and optimizing a water motor runner state relevance measurement result. The method is used for discovering the incidence relation between the states of the water motor rotating wheel, provides a related basis for fault diagnosis of the water motor rotating wheel according to the incidence relation, can help to formulate eff