Distributed optimization for the optimal control of electric vehicle fleets

Doctoral thesis, 2023

Owing to their absence of tailpipe emissions and their independence from fossil fuels, Electric Vehicles (EVs) are currently experiencing a rapid deployment in an attempt to curb global greenhouse gas emissions. EV operation represents a technical challenge, however, as new control algorithms need to be developed to address their limited driving range and their longer charging times. Optimization-based control techniques offer a promising way to plan EV operation over a prediction horizon while including key operational constraints, but they can be prohibitively slow for real-time applications as they rely on solving computationally hard optimization problems. One way to address the computational complexity of these approaches is by deploying adapted decomposition methods with which the computational load of solving these optimization problems can be distributed across the vehicles involved, where most computations can then be carried out in parallel.

This thesis presents decomposition-based solution procedures for optimal control problems involving groups of EVs. In particular, the problems covered in this work are (i) the cooperative eco-driving control of a platoon of electric trucks, (ii) the eco-driving and operational control of an electric bus line, and (iii) the operational control and charging scheduling of an electric bus network. Even though their particular objective functions and constraints may differ, the coupling structures of these problems, i.e. how each vehicle's influence on the others is organized, share some similarities.

The platoon control problem is formulated as a Nonlinear Program (NLP) and solved with second-order optimization methods. The Riccati recursion is used as part of a decomposition scheme that exploits the chain-like coupling structure of a truck platoon and makes it possible to fully distribute all computations. Similarly, the bus line problem is formulated as an NLP. A primal decomposition scheme where the NLP is split into a master problem and independent bus subproblems is presented. The hierarchical control architecture obtained makes it possible to distribute most of the computations. Finally, the bus network problem is formulated as a Mixed-integer Linear Program (MILP). A dual decomposition scheme based on Lagrangian relaxation is deployed to relax the coupling constraints between the different bus lines.