Stochastic Dynamic Programming for Hybrid

Vehicle Control

Dr. Ed D. Tate

General Motors

Abstract

In the design of hybrid electric vehicles (HEVs), one of the challenges is creating control laws that optimize performance. Like conventional vehicles, this performance includes fuel economy, tailpipe emissions and the driver's perceptions of the powertrain's operation. One way to develop a causal controller for this problem is to use stochastic dynamic programming (SDP). To apply SDP, the driver is modeled as a stochastic process. This stochastic process is coupled to a deterministic model of the vehicle, powertrain, and driver's perceptions. This combined model is transformed into an equivalent discrete SDP. Using a variety of methods, the SDP is solved to find an optimal controller. The methods used include linear programming, barycentric interpolation and constraint generation, This combination of techniques reduces the solution time from several thousand hours to three hours on a desktop PC. This controller synthesis process is applied to a 2-Mode EVT with a thermally transient catalyst. Families of controllers are synthesized to evaluate trade-offs among the competing objectives. In the example considered, for a few percent increase in fuel consumption, the tailpipe emissions are decreased by almost fifty percent. An interesting aspect of these controllers is that they discover techniques and strategies that were previously developed using heuristic processes.

Friday, March 23. 2007

3:30 – 4:30 p.m.

Rm. 1500 EECS