A novel sparrow search algorithm based co-correlation graph construction strategy for wind turbine group anomaly identification via graph attention networks

Abstract

Anomaly identification by using Supervisory Control and Data Acquisition (SCADA) data is an important means to improve the reliability of wind turbine group (WTG) operation. However, due to the low reliability of SCADA systems, anomalies in the data itself may occur as a result of sensor failures or data transmission errors. The anomalies in the data itself will reduce the accuracy and reliability of WTG anomaly identification. In this paper, a sparrow search algorithm (SSA) based co-correlation graph (CG) construction strategy using graph attention networks (SSACG-GAT) is proposed for WTG anomaly identification. First, the adjacency matrix representing the correlation of sample parameters is constructed by taking the monitoring parameters as nodes and the correlation between parameters as edges. Second, the proposed SSAGC strategy is used to obtain a co-correlation graph by fusing the adjacency matrices calculated by different correlation analysis models. In the proposed SSAGC strategy, the SSA is used to obtain the optimal fusion weights of the different adjacency matrices. Finally, the obtained optimal co-correlation graph is input into the GAT network for WTG anomaly identification. Nine models are selected as benchmarks to validate the effectiveness and superiority of the proposed SSACG-GAT. The experimental results show that the proposed SSACG-GAT has the best identification performance compared with nine benchmark methods. In addition, the ablation experiment results also demonstrate that the proposed SSACG strategy can effectively improve the accuracy and reliability of WTG anomaly identification.