Anomaly Detection of Electric Gate Valve Based on Multi-Kernel Support Vector Machine

Abstract

Valve is an indispensable fluid control component in nuclear power system. Nuclear power station has a large number of gate valve equipment, which works under high temperature, high pressure, high radioactivity and other harsh conditions. In nuclear power plant accidents and economic losses, a considerable part of them are caused by valve failure. Aiming at the fault of electric gate valve, this paper proposes an anomaly detection method based on multi-kernel support vector machine. Firstly, the acoustic emission instrument is used to measure the fault state data and extract the fault features. Secondly, on the basis of classical support vector machine, multiple kernel function combinations are selected to decompose the model into convex optimization problems to realize the abnormal state detection of internal leakage fault of electric gate valve in nuclear power plant. The results show that, compared with the classical support vector machine method, the constructed support vector machine method based on multikernel learning has better effect and higher accuracy in anomaly detection of electric gate valve.