Credibility reconstruction and anomaly consistency identification algorithm for power generation operational data based on graph structure inference

The susceptibility of the power generation systems to reliable anomaly detection is of the essence in ensuring the safety of the operation, allowing the maintenance costs to be minimised, as well as energy efficiency. Nevertheless, traditional data-driven methods can hardly account for intricate interdependences between operational sensors and tend to be extremely vulnerable to noise and irregular faults. This research will resolve these difficulties by offering a Graph Structure Inference -based Credibility Reconstruction and Anomaly Consistency Identification framework of wind turbine operational data. The suggested approach simultaneously builds a latent graph structure to capture latent correlations among turbine sensors and presents a credibility-sensitive reconstruction scheme that changes the reliability of sensors in a dynamic fashion through training. Further, an anomaly identification strategy is based on consistency to differentiate between enduring fault patterns and anomalies that are short lived or noise induced on temporal and structural levels.The given framework is tested with a chronological train validation test split on an actual wind turbine SCADA dataset. Experiments prove that the proposed algorithm has superior performance with over state-of-the-art baseline models with a mean absolute error (MAE) of 0.041, root mean square error (RMSE) of 0.058 and an overall detection rate of 96.3%. Moreover, the approach has a precision of 95.1, recall of 94.6, F1-score of 94.8 and AUC of 0.982, which implies that it has a high discriminative power. The analysis of robustness under injected noise and missing data conditions demonstrates that the analysis suffers degradation with consistently low variance in the results of several runs. These findings confirm that the suggested method is effective in improving the reliability and robustness of anomaly detection, which outplays its use in intelligent monitoring and predictive maintenance of the contemporary power generation systems.