Design of a neuro-classifier/detector for Amtrak rail-road track operations

Abstract

This paper proposes a design for a neural network that can be used to detect and classify generic railroad operating conditions as abnormal, not reverse, normal and reverse. The proposed neural net would be of a learning vector quantization type, and would be trained online to capture the nonlinear mapping that transforms a specific location, time of the day, and direction of travel into a quantitative statement of whether or not an abnormal operating condition is possible at these inputs. Chosen as the test bed for this work is the Centralized Electrification and Traffic Control (CETC) system operated by Amtrak on the northeast corridor. Specifically, it is planned to develop and install a neural net based system and allow it to detect "normal" traffic patterns, switch settings and security conditions. To the best of the authors' knowledge no similar work is outstanding, planned or anticipated at this time.