Research on anomaly detection and positioning of marine nuclear power steam turbine unit based on isolated forest

The steam turbine unit is an important power output and energy conversion device of the marine nuclear power system. The existing research on anomaly detection and positioning of marine nuclear power steam turbine units has the problems of unreasonable threshold setting, training model relying on abnormal data, and inability to locate anomalies. This study only uses normal data to train unsupervised clustering isolated forest algorithm, which avoids dependence on abnormal data and makes decision threshold more reasonable. At the same time, the concepts of time window and anomaly score are introduced to nalyse time factors and anomaly influence propagation factors, so as to realize anomaly detection and positioning at the system level, equipment level and parameter level of marine nuclear power steam turbine units. The recall rate of this method for each abnormal state exceeds 84%, the precision rate exceeds 94%, and the recall rate and precision rate for some abnormal states exceed 99%. At the same time, this method can accurately locate abnormal equipment and parameters.