A comparison study of data-driven projection to latent structures modeling and monitoring methods on high-speed train operation

Abstract

High-speed train operates in a multi-level style with nonlinearity which makes it difficult to build the overall operation model. Traditional model-based methods are thus difficult to model and monitor the high-speed train operation. In this paper, data driven modeling and monitoring of high-speed train operation are discussed. First, multi-level high-speed train operation and fault are described; Secondly, three popular latent structure modeling methods, i.e., concurrent projection to latent structures (CPLS), kernel partial least squares (KPLS), and concurrent kernel projection to latent structures (CKPLS), are briefly reviewed and applied to model the high-speed train operation and the corresponding monitoring methods are provided. Thirdly, a comparison study of the above three methods is achieved by the simulation using practical operation data. The results demonstrate that i) the nonlinear modeling and monitoring methods, i.e., KPLS and CKPLS, can reduce false alarms compared to linear monitoring method, i.e., CPLS; ii) the CKPLS based monitoring can enhance fault explanation compared to KPLS based monitoring.