A Practice of Sensory Cooperation for Insulation Diagnosis of Electrical Apparatus: Probabilistic Identification and Risk Assessment

In this study, a set of algorithms is tailored to achieve insulation defect probability identification and risk assessment based on multiphysical detections of partial discharges (PDs). Initially, three types of PD detection units are integrated into the flange to achieve a relatively uniform detection distance. Also, the synchronous PD detection platform is constructed based on intelligent sensor units, achieving extraction of sequential PD impulse information. Then, the multiphysics energy release power (ERP) is extrapolated based on the amplitude of the measurement signal amplitude. Also, three ERP values are normalized to analyze the proportional relationship of energy releases at different stages of discharge, which is further mapped in a ternary power pattern. In addition, we develop a probability recognition model based on the ternary error-correcting output coding (ECOC) united with a support vector machine (SVM). Meanwhile, by introducing discharge data of the same type with different defect scales, it can be verified that the proposed model possesses significant robustness. Besides, Gaussian process regression (GPR) is applied to fit the relationship between the ternary power pattern and apparent discharge energy (ADE). Also, the risk index and risk level are defined based on the temporal information of ADE during the entire discharge process, which is incorporated with the pulse repetition rate to achieve the visualization of discharge risk assessment.