Skip to Main Content
This paper demonstrates the feasibility of employing artificially intelligent automation for the task of monitoring the Amtrak railroad track system in a real-time transportation environment. The neural-net-based device (automation) processes several quantities that portray the localized existence of the Amtrak system. These quantities may be one or more of the following: location of the switch on the railroad track; time of observation; and the direction of travel. Given these quantities, it is desired that the state of the system, (which can only belong to one of several distinct categories) be predicted as outputs of the automation. Possible outputs are conditions classified as NORMAL, NOT NORMAL , REVERSE, and NOT REVERSE . Implicit in the choice of a configuration of inputs and outputs is the hypothesis of the existence of a multivariable mapping connecting these inputs and outputs-a mapping that hopefully coincides with the real-world dynamics of the railroad track. The neural-net-based device is tested on a specific, already-in-place transportation control system-the Centralized Electrification & Traffic Control (CETC) system operated by Amtrak on the northeast corridor. The CETC is chosen because of the clear value which such an operational safety and security monitor would bring to it. The test results obtained in this paper confirm that artificial neural networks can be effectively used to solve the pattern recognition problem posed by the Amtrak system. To the best of the author's knowledge, no similar work is outstanding, planned, or anticipated at this time.