Michigan Technological University
MACHINE LEARNING AND MULTI-SCALE OPTIMIZATION FOR CONTROL AND ENERGY MANAGEMENT IN CONNECTED AND AUTOMATED VEHICLE PROPULSION SYSTEMS
Abstract
dc:description.abstractThis dissertation presents a multi-scale optimization framework leveraging machine learning (ML) to enhance energy efficiency in connected and automated vehicle (CAV) propulsion systems. As transportation transitions toward hybridization and automation, the integration of vehicle-to-everything (V2X) connectivity and advanced control algorithms offers unprecedented opportunities for energy reduction. This research addresses three critical scales of vehicle energy management: multiple vehicle-level coordination, component-level powertrain dynamics, and real-time vehicle parameter estimation. First, the research investigates the energy consumption characteristics of heterogeneous propulsion systems—ranging from internal combustion engines to battery electric vehicles across light- and heavy-duty sectors—on arterial roadways. Utilizing Particle Swarm Optimization (PSO) and machine learning (ML) based classifiers, a cooperative driving strategy was developed to optimize velocity trajectories and lane utilization. Analysis of a 140,000+ run design of experiments, validated against experimental data, demonstrates energy reductions between 10% and 50%, contingent upon cohort composition and infrastructure constraints. Second, the study optimizes mode shift dynamics for dedicated hybrid transmissions, focusing on the Chevrolet Volt II. By coupling a reduced-order transmission model with PSO and connectivity data (e.g., road grade), the research identifies optimal shift trajectories for the engine and electric motors to meet a desired axle torque profile. Results indicate that optimizing individual shift events can reduce energy consumption by up to 80% per shift, translating to a 3–4% improvement across real-world drive cycles. Finally, the dissertation proposes a real-time ML-based method for in-vehicle mass and road load characterization. By employing PSO to converge on estimated coefficients more representative of those determined by standard EPA coast-down procedures, the system enables more accurate predictive energy management. This method achieves axle torque predictions that correlate directionally with changes in physical vehicle parameters, providing a robust foundation for look-ahead control strategies. Collectively, these contributions demonstrate that the synergy of machine learning, connectivity, and multi-scale optimization can significantly reduce the energy footprint of modern vehicle fleets, providing a scalable pathway for the next generation of intelligent propulsion system control.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy in Mechanical Engineering-Engineering Mechanics (PhD)
- Level thesis:degree_level
- Open Access Dissertation
- Discipline thesis:degree_discipline
- Department of Mechanical and Aerospace Engineering
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Orlando, Joshua D.
- Committee members dc:contributor.committeemember
-
- Jeffrey D. Naber
- Jason R. Blough
- Anthony J. Pinar
- Contributors dc:contributor
-
- Darrell L. Robinette
Subjects
dc:subject × 12Identifiers
dc:identifier.*- Identifier
- https://digitalcommons.mtu.edu/etdr/2131