A multi-model anomaly detection method

Theory foundation of multivariate statistical anomaly detection is hypothesis test, which require that all data to establish the statistical model must be samples of a unique set of random variables. In most cases, due to the affect of environment and load variation of the system, this assumption can´t be satisfied. In addition, during different operation substage of a system, faults with different frequency may be occurred. This will decrease the anomaly detection efficiency. In this paper, we propose a multi-model anomaly detection method, which can adaptively select history data on different scale to establish the statistical anomaly detection model. Thus we can use the multimodel to well detect every possible fault. Simulation shows its efficiency of this algorithm.