Stability of neural-network based train cruise advisory control with aperiodical measurements

Abstract

Abstract In this paper, a neural-network based driver advisory train cruise control system is considered. The controller assists the train driver with advisory signals by considering train and driver’s actual state information (attention and fatigue) measured by dedicated sensors. Considering delays in sensor measurements, this paper aims to assess closed-loop stability of driver-in-the-loop advisory train cruise control. For this purpose, the driver model is considered as a time-varying system, the train model includes rolling and aerodynamic resistance forces and the advisory control is considered to be a sampled-data based three layer multi-layer perceptron. Further, the aperiodic measurement problem is approached as stability analysis of time-varying delayed system. Based on recent developments on the design of augemented Lyapunov Krasovskii Functional (LKF) using Bessel-Legendre inequality for time-varying delays, sufficiency conditions for the existence of L2 stability of the driver-train system in terms of solvable Linear Matrix Inequalities are provided. Further a case study is presented to illustrate the effectiveness of the proposed method.