Online Anomaly Detection in Switching-Mode Power Module Using Statistical Property Features Comparison and Gaussian Process Regression

Switching-mode power module (SMPM) plays an important role in the electrical system. The reliability of SMPM has the decisive effect on the working state of back-end components even the entire electrical system. To identify the anomaly state effectively, the paper proposes an online anomaly detection method using statistical property features comparison and Gaussian Process Regression (GPR), especially for the SMPM with the unknown circuit structure. Firstly, for describing the uncertainty in evaluation and prediction, Gaussian Process Regression (GPR) is adopted to perform the prediction normal output range with the mean and variance values as the uncertainty representation of the output signal. Then six statistical property features are used to analyze the output signal, they can identify the anomaly in different ways, and reduce the sampling frequency and hardware cost. When one of six statistical property features of the online output deviates from the prediction normal output range, which demonstrates that SMPM has existed the anomaly. The simulation experimental results validate two conclusions: The prediction accuracy using the combination covariance function based on GPR is higher than the single covariance function; The anomaly will be online detected remarkably using statistical property features comparison between the online actual output voltage and the prediction normal output range.