Route Optimized Energy Management of Plug-in Hybrid Electric Vehicles

Plug-in hybrid electric vehicles have the potential to significantly reducethe oil dependence within the transportation sector. However, there willalways be some trips that exceed the electric driving range, meaning thatboth electric energy and fuel must be used. For such trips the fuel economyis intimately connected with the energy management system and its abilityto schedule the use of the battery. The fundamental problem is that theoptimal fuel economy can be reached only if the future trip is known a priori.It is therefore desirable to have a system that can perform three principaltasks: i) acquire a prediction of the future trip, ii) given the predictionprecompute feedforward information for the real-time level, and iii) at thereal-time level identify the optimal operating points in the powertrain.This thesis investigates all three of the mentioned tasks. It is shown thatfrequently travelled routes can be identified from logged driving data usinghierarchical clustering. Based on the historical driving conditions along theroute, it is then possible to precompute an optimal strategy that can be usedas feedforward information for the real-time level. Two different methods forsuch a precomputation are investigated, convex optimization and DynamicProgramming. Particular attention is given to the implementation of acomputationally efficient Dynamic Programming algorithm.A real-time control strategy that is based on a closed-form minimizationof the Hamiltonian is also presented. The strategy is derived for a power-train with two degrees of freedom, and is implemented in a dynamic vehiclemodel that is used by a vehicle manufacturer. Simulations with a linearlydecreasing battery state of charge reference indicate that the fuel economycan be improved with up to 10%, compared to a depleting-sustaining strat-egy. Real-time compatible controller code is also generated and tested in aproduction vehicle. The vehicle behaviour during a test drive is similar tosimulated behaviou