
NSF Org: |
CMMI Division of Civil, Mechanical, and Manufacturing Innovation |
Recipient: |
|
Initial Amendment Date: | July 30, 2019 |
Latest Amendment Date: | July 30, 2019 |
Award Number: | 1932250 |
Award Instrument: | Standard Grant |
Program Manager: |
Jordan Berg
jberg@nsf.gov (703)292-5365 CMMI Division of Civil, Mechanical, and Manufacturing Innovation ENG Directorate for Engineering |
Start Date: | September 1, 2019 |
End Date: | August 31, 2023 (Estimated) |
Total Intended Award Amount: | $198,698.00 |
Total Awarded Amount to Date: | $198,698.00 |
Funds Obligated to Date: |
|
History of Investigator: |
|
Recipient Sponsored Research Office: |
438 WHITNEY RD EXTENSION UNIT 1133 STORRS CT US 06269-9018 (860)486-3622 |
Sponsor Congressional District: |
|
Primary Place of Performance: |
371 Fairfield Way, Unit 4155 Storrs CT US 06269-4155 |
Primary Place of
Performance Congressional District: |
|
Unique Entity Identifier (UEI): |
|
Parent UEI: |
|
NSF Program(s): | CPS-Cyber-Physical Systems |
Primary Program Source: |
|
Program Reference Code(s): |
|
Program Element Code(s): |
|
Award Agency Code: | 4900 |
Fund Agency Code: | 4900 |
Assistance Listing Number(s): | 47.041 |
ABSTRACT
As electric vehicle technologies become mature, they have been rapidly adopted in modern transportation systems, such as electric taxis, electric buses, electric trucks, and shared-personal rental electric vehicles, due to their environment-friendly nature. Since electric vehicles require frequent yet time-consuming recharges, their dispatching and charging activities have to be managed efficiently considering the high charging demand of large-scale electric vehicles and the limited charging infrastructure. Therefore, an efficient electric vehicle management framework has the potential to (i) reduce the traveling distance to a charging station, (ii) reduce the wait time for electric vehicles to charge, and (iii) balance the demand and supply for charging infrastructure. However, current management strategies for electric vehicles are mainly based on homogeneous electric vehicles, ignoring challenges and opportunities introduced by heterogeneous electric vehicles, for example, electric personal vehicles, electric taxis, and electric buses. In this project, the research team will design and implement a set of management strategies for heterogeneous electric vehicle fleets, which utilize real-time data from various electric vehicles to improve the overall performance of heterogeneous electric vehicle fleets. If successful, the research team will develop a clear understanding of how to manage large-scale heterogeneous electric vehicles to improve urban mobility efficiency from a fleet-oriented perspective, with potential applications to future autonomous electric vehicles. Such an understanding of heterogeneous electric vehicles will improve the quality of every-day life such as more efficient commutes and lower energy usage, which will benefit the environment.
To date, researchers have accumulated abundant knowledge on how to manage individual electric vehicles, even homogeneous electric vehicle fleets, based on precise mathematical models. Nevertheless, such models are over-simplified as they do not consider the cyber-physical hybrid state space or model uncertainties for heterogeneous electric vehicle fleets. Heterogeneous electric vehicle fleets are mobile cyber-physical systems with heterogeneous properties, for example, mobility patterns, energy consumption, and incentives to accept control decisions. However, the research community has a limited understanding of how to make network control decisions for heterogeneous mobile cyber-physical systems at large scale in a real-world setting. In this project, the research team aims to investigate the fundamental theories and applications to manage heterogeneous mobile cyber-physical systems by utilizing electric vehicles as an example platform. Specifically, the research team will (i) develop a set of data-driven cyber and physical models to predict the essential status of heterogeneous electric vehicle fleets, for example, mobility patterns and energy consumption rates and (ii) establish a hierarchical control framework to achieve performance guarantees for heterogeneous electric vehicle dispatching and charging management by developing data-driven distributionally robust optimization methods for hybrid systems.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
PUBLICATIONS PRODUCED AS A RESULT OF THIS RESEARCH
Note:
When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external
site maintained by the publisher. Some full text articles may not yet be available without a
charge during the embargo (administrative interval).
Some links on this page may take you to non-federal websites. Their policies may differ from
this site.
PROJECT OUTCOMES REPORT
Disclaimer
This Project Outcomes Report for the General Public is displayed verbatim as submitted by the Principal Investigator (PI) for this award. Any opinions, findings, and conclusions or recommendations expressed in this Report are those of the PI and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed its content.
We have achieved several goals in this project: (1) design efficiently managing electric vehicle (EV) fleets as a mobile CPS with heterogeneous mobility and charging patterns from a data-driven perspective by enabling data-driven prediction of both EV fleets and charging network status; (2) design decision making strategies by a set of novel techniques based on Optimization, Control Theory, and Machine Learning in the NSF vision of Harnessing Data Revolution; (3) potentially generalize these techniques to a broader setting of scenarios with a system of heterogeneous mobile CPS where each system has its own operating and mobility patterns but has the potential to be managed collectively for better performance.
The findings and results are published and accepted in top tier IEEE/ACM conferences, e.g., KDD, ICDE, UbiComp, IROS, ICRA, ICLR workshops. We also have some results accepted to journals with high impacts, such as Transactions on Machine Learning Research and, IEEE Transactions on Knowledge and Data Engineering, ACM Transactions on Intelligent Systems and Technology, ACM Transactions on Cyber-Physical Systems. Multiple invited talks have been delivered to top conferences, workshop and universities. The main results contribute to the mobile CPS and EV research society, by developing an integrated learning, optimization, and control algorithm that utilizes social and contextual information hidden behind heterogeneous EV fleets. The practical challenges in the complicated dynamic systems of EVs also motivate theoretical research topics such as robust multi-agent reinforcement learning, and the research progress will benefit both the CPS research field and general AI research field. We design a data-driven fairness-aware deep reinforcement learning algorithm for EV charging and repositioning scheduling, while improves the overall profit efficiency and fairness with different practical factors consideration. We have released some sample EV data to benefit the research community.
The findings and results will also bridge the gap between learning, optimization, control, and social science communities to develop better integrated methodologies for data-driven socially informed decision-making systems. We consider our EV charging problem based on machine learning in a broader setting of Fairness in AI and Computational Social Science. The data-driven models we developed consider different social factors and practical real-world constraints have the potential to serve as a concrete use case of AI technology considering fairness and equity. The practical challenges in considering model uncertainties for learning-based decision-making of CPS also motivate theoretical research topics in Embodied AI, such as robust multi-agent reinforcement learning with state uncertainties. The research progress will benefit Embodied AI and learning-based decision-making for CPS research field.
In particular, for the work published in IEEE transactions on Intelligent Transportation systems in 2023, we provide some new findings which have not been uncovered by existing literature. (1) Our proposed mathematical system-level vehicle balancing framework is the first to consider both future mobility demand uncertainties and EV supply uncertainties for EV AMoD systems. While model predictive control algorithms have been designed considering AMoD system demand uncertainties in the literature, the supply side uncertainties for EV AMoD are not well studied yet. (2) We design a distributionally robust optimization approach to balance EVs across a city to provide fair passenger mobility and EV charging service while reducing the total balancing cost. (3) We design efficient algorithms to construct uncertainty sets of probability distributions given different prediction models for the demand and supply. (4) Based on EV dataset, we show that our method reduces the average total balancing cost by 14.49%, the average mobility unfairness and charging unfairness by 15.78% and 34.51%, respectively, compared to non-robust solutions.
For the work accepted by IROS 2023, we design a multi-agent reinforcement learning (MARL)-based framework for EAVs balancing in E-
AMoD systems, with adversarial agents to model both the EAVs supply and mobility demand uncertainties that may undermine the vehicle balancing solutions. We then propose a robust E-AMoD Balancing MARL (REBAMA) algorithm to train a robust EAVs balancing policy to balance both the supply-
demand ratio and charging utilization rate across the whole city. Experiments show that our proposed robust method performs better compared with non-robust MARL method without considering system model uncertainties; it increases the reward, charging utilization fairness, and supply-demand fairness by 19.28%, 28.18% and 3.97% respectively.
For the work published at Transactions on Machine Learning Research, we study the problem of MARL with state uncertainty in this work. We provide the first attempt to the theoretical and empirical analysis of this challenging problem. We first model the problem as a Markov Game with state perturbation adversaries (MG-SPA) by introducing a set of state perturbation adversaries into a Markov Game. We then introduce robust equilibrium (RE) as the solution concept of an MG-SPA. We conduct a fundamental analysis regarding MG-SPA such as giving conditions under which such a robust equilibrium exists. Then we propose a robust multi-agent Q-learning (RMAQ) algorithm to find such an equilibrium, with convergence guarantees.
Last Modified: 12/31/2023
Modified by: Fei Miao
Please report errors in award information by writing to: awardsearch@nsf.gov.