Model predictive eco-driving control for heavy-duty trucks using Branch and Bound optimization

TL;DR

This paper presents a real-time branch and bound model predictive control algorithm for eco-driving in heavy-duty trucks, achieving significant fuel savings through optimized velocity and gear control in dynamic environments.

Contribution

It introduces a novel BnB MPC approach for eco-driving that incorporates a finite-mode velocity model and real-time optimization techniques.

Findings

Achieved 25.8% fuel savings compared to human drivers.

Achieved 12.9% fuel savings compared to Pontryagin's Minimum Principle solutions.

Validated effectiveness through simulations on real-world routes.

Abstract

Eco-driving (ED) can be used for fuel savings in existing vehicles, requiring only a few hardware modifications. For this technology to be successful in a dynamic environment, ED requires an online real-time implementable policy. In this work, a dedicated Branch and Bound (BnB) model predictive control (MPC) algorithm is proposed to solve the optimization part of an ED optimal control problem. The developed MPC solution for ED is based on the following ingredients. As a prediction model, the velocity dynamics as a function of distance is modeled by a finite number of driving modes and gear positions. Then we formulate an optimization problem that minimizes a cost function with two terms: one penalizing the fuel consumption and one penalizing the trip duration. We exploit contextual elements and use a warm-started solution to make the BnB solver run in real-time. The results are evaluated in numerical simulations on two routes in Israel and France and the long haul cycle of the Vehicle Energy consumption Calculation Tool (VECTO). In comparison with a human driver and a Pontryagin's Minimum Principle (PMP) solution, 25.8% and 12.9% fuel savings, respectively, are achieved on average.