Optimal Control of Heavy-Haul Freight Trains to Save Fuel

Dr. Paul K. Houpt, Principal Scientist, Controls, GE Global Research, Automation and Controls Laboratory

Fuel burned by North American Class 1 railroads in diesel-electric freight service in 2007 exceeded 4.1 Billion gallons, resulting at 2007 fuel prices, in 13% of overall operations expense1. Trends in fuel prices, ignoring the present hiatus due to the economy, are only upward. As more freight shifts to rail directly and via inter-modal truck/train haulage to leverage incredible efficiency of rail transit compared to other modes, total fuel usage will accelerate. This presentation describes a software control system applicable to diesel-electric locomotive hauled freight that can achieve double-digit fuel savings. Energy savings derive from managing train momentum, with anticipation of its effects, to reduce the net energy outlay by the train as it completes a trip. GE's system has two major components: the first is a planning system that derives an optimal way to drive the train (throttle together with a corresponding speed trajectory versus distance) subject to speed restrictions along the route and locomotive operating constraints; the second component is a dynamic control system that executes the plan closed-loop, correcting for modeling errors from various sources and assuring safe train handling. We'll show how this reduces to a familiar two-point boundary value problem in optimal control. Solving this optimal control problem requires a simplified model of the physics of motion and a highly efficient optimization algorithm. Key features of the algorithm are ability to optimize for minimum fuel subject to a desired arrival time (pacing) or to minimize travel time, and in both cases to assure safe train handling. A closed-loop dynamic control regulator system takes the optimal solution and corrects for various disturbances and model errors. Location along the track route is derived from GPS blended with on-board measurements in a Kalman filter. The resulting software was rapid prototyped 100% in a Matlab xPC platform, installed in locomotives on various railroads in live revenue service. This enabled quick proof of projected benefits and was a major factor in the decision to commercialize the system, which GE calls,  ``Trip Optimizer''.  The talk will present field test results using the system prototype in revenue service on several North American railroads, where savings exceeding 10% were successfully demonstrated on trains hauling between 3000 and 19000 tons. Finally, some lessons learned about taking complicated controls ideas from concepts to commercial products are summarized.