Fuel Optimal Control Algorithms for Connected and Automated Plug-In Hybrid Vehicles

Abstract

Improving the fuel economy of light-duty vehicles (LDV) is a compelling solution to stabilizing Greenhouse Gas (GHG) emissions and decreasing the reliance on fossil fuels. Over the years, there has been a considerable shift in the market of LDVs toward powertrain electrification, and plug-in hybrid electric vehicles (PHEVs) are the most cost-effective in avoiding GHG emissions. Meanwhile, connected and automated vehicle (CAV) technologies permit energy-efficient driving with access to accurate trip information that integrates traffic and charging infrastructure. This thesis aims at developing optimization-based algorithms for controlling powertrain and vehicle longitudinal dynamics to fully exploit the potential for reducing fuel consumption of individual PHEVs by utilizing CAV technologies.

A predictive equivalent minimization strategy (P-ECMS) is proposed for a human-driven PHEV to adjust the co-state based on the difference between the future battery state-of-charge (SOC) obtained from short-horizon prediction and a future reference SOC from SOC node planning. The SOC node planning, which generates battery SOC reference waypoints, is performed using a simplified speed profile constructed from segmented traffic information, typically available from mobile mapping applications.

The PHEV powertrain, consisting of engine and electric motors, is mathematically modeled as a hybrid system as the state is defined by the values of the continuous variable, SOC, and discrete modes, hybrid vehicle (HV), and electric vehicle (EV) modes with the engine on/off. As a hybrid system, the optimal control of PHEVs necessitates a numerical approach to solving a mixed-integer optimization problem. It is of interest to have a unified numerical algorithm for solving such mixed-integer optimal control problems with many states and control inputs. Based on a discrete maximum principle (DMP), a discrete mixed-integer shooting (DMIS) algorithm is proposed. The DMIS is demonstrated in successfully addressing the cranking fuel optimization in the energy management of a PHEV. It also serves as the foundation of the co-optimization problem considered in the remaining part of the thesis.

This thesis further investigates different control designs with an increased vehicle automation level combining vehicle dynamics and powertrain of PHEVs in within-a-lane traffic flow. This thesis starts with a sequential (or decentralized) optimization and then advances to direct fuel minimization by simultaneously optimizing the two subsystems in a centralized manner. When shifting toward online implementation, the unique challenge lies in the conflict between the long control horizon required for global optimality and the computational power limit. A receding horizon strategy is proposed to resolve the conflict between the horizon length and the computation complexity, with co-states approximating the future cost. In particular, the co-state is updated using a nominal trajectory and the temporal-difference (TD) error based on the co-state dynamics.

The remaining work aims to develop a unified model predictive control (MPC) framework from the powertrain (PT) control of a human-driven to the combined vehicle dynamics (VD) and PT control of an automated PHEV. In the unified framework, the cost-to-go (the fuel consumption as the economic cost) is represented by the co-state associated with the battery SOC dynamics. In its application to automated PHEVs, a control barrier function (CBF) is augmented as an add-on block to modify the vehicle-level control input for guaranteed safety. This unified MPC framework allows for systematically evaluating the fuel economy and drivability performance of different levels and structures of optimization strategies.