A Turnout Anomaly Detection Method Based on Kernel-Aligned Mixed Kernel Function SVDD

Undertaking anomaly detection on turnout with missing fault data shows an important role in ensuring railway safety. In this paper, the kernel alignment technique is used to calculate the weights of each base kernel function, to construct a hybrid kernel function to replace the single kernel function in the traditional SVDD, a new KASVDD model is obtained to avoid the dependence on the model performance on the kernel parameters selection. In addition, we discuss the adaptive power curve segmentation algorithm and feature selection optimization algorithm based on mRMR to reduce the unnecessary features in the extracted time domain features and avoid information redundancy. For validation, the method was applied to the power data collected from the Guangzhou Metro Test Platform for S700 K-type switch machines, and the validity and feasibility of the framework are verified.