Neuroevolution and Machine Learning Research Applied to Connected Automated Vehicle and Powertrain Control
Date of Award
Open Access Dissertation
Doctor of Philosophy in Mechanical Engineering-Engineering Mechanics (PhD)
Administrative Home Department
Department of Mechanical Engineering-Engineering Mechanics
Jung Yun Bae
Committee Member 1
Committee Member 2
This dissertation focuses on advancing Predictive Energy Management (PrEM) functions applied to modern connected and automated vehicles (CAV) cohorts. PrEM aims to utilize connectivity and ADAS functions to adaptively minimize vehicle energy consumption in a wide array of operations, extending the original control designed around a reduced set of test cycle procedures to adapt to real-world stochastic operating conditions. This research document is built upon three journal publications covering two PrEM schemes; the global cohort and local vehicle optimization paths. Both optimal control solutions are generated using various Neuroevolution centric processes.
Chapter 1 discusses the methods and reasoning behind the need to increase the development speed of readily implementable optimal control functions for both complex and system-of-systems (SoS) applications. Neuroevolution allows for fast development time, optimal design space exploration, high-fidelity modeling usage, and seamless integration with data science processes. It additionally enables real-time implementation without modification and requires a low compute footprint. This provides a new paradigm for future automotive product development where conventional adaptive and optimal techniques deployment is still lagging due to their complexity and shortcomings.
At the global level, vehicle energy consumption is minimized by optimally controlling vehicle speed in diverse environments. Chapters 2 and 3 relate to connected traffic lights and uncontrolled intersection operations respectively. In the first study, the CAV cohort optimizes its velocity based on connected traffic light information. Thanks to the Traffic Technology Services (TTS) network, this information is shared via cellular communication. Energy consumption reduction of up to 22\% is reported using simulation and during closed-loop track testing. In the second study, no such timing information exists, and the cohorts must collaborate to enable safe operation at uncontrolled intersections. Here, the cohorts share states' information to minimize deceleration and acceleration events for comfort and energy savings, primarily focusing on safety. Simulation demonstrates that effective collaboration can be achieved with cohorts' lengths of up to 100 meters in congested environments.
At the local PrEM level, additional energy savings can be achieved for each specific cohort's vehicle based on its powertrain architecture. One of the more complex and relevant architectures to apply localized PrEM to are hybrid electric vehicles (HEV), where two sources of energy can be blended optimally based on a vehicle's predicted speed profile, which is directly controlled by the global PrEM optimization function. In Chapter 4, Neuroevolution and vehicle speed profile classification is applied to a P3 HEV in demonstrating significant additional energy consumption improvements.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.
Jacquelin, Frédéric F., "Neuroevolution and Machine Learning Research Applied to Connected Automated Vehicle and Powertrain Control", Open Access Dissertation, Michigan Technological University, 2023.
Available for download on Monday, July 31, 2023