Aggregative charging control of electric vehicle populations
Lead Research Organisation:
University of Oxford
Department Name: Engineering Science
Abstract
The ambitious targets in the United Kingdom for increasing the share of renewable energy sources integrated to the network, and the need for providing affordable, resilient and clean energy, call for a paradigm shift in energy systems operations. Electric vehicles offer the means to address these challenges and achieve uninterrupted operation by deferring their demand in time and acting as dynamic storage devices. As a result, their number is expected to increase rapidly over the next years, leading to a "green car revolution". This constitutes an opportunity for modernizing energy systems operation, but will unavoidably give rise to coordination and scheduling issues at a population level so that cost savings are achieved and reliability is ensured. The latter is of significant importance to prevent from undesirable disruptions of service.
This project will address this problem using tools at the intersection of control theory, optimization and machine learning, allowing for a decentralized computation of the electric vehicle charging strategies, while preventing vehicles from sharing information about their local utility functions and consumption patterns that is considered to be private. We will develop algorithms capable of dealing both with cooperative and non-cooperative vehicle behaviours in large fleets of vehicles, and immunize the resulting strategies against uncertainty due to unpredictability in the vehicles' driving behaviour and due to the presence of renewable energy sources. The presence of an algorithmic tool with these features will allow for scalable charging solutions amenable to problems of practical relevance, will provide insight on the mechanism driving the response of large populations of electric vehicles, and embed robustness in the resulting charging schedules. As such, the proposed project will offer the means for reliable system operation and facilitate the integration of higher shares of renewable energy sources.
This project will address this problem using tools at the intersection of control theory, optimization and machine learning, allowing for a decentralized computation of the electric vehicle charging strategies, while preventing vehicles from sharing information about their local utility functions and consumption patterns that is considered to be private. We will develop algorithms capable of dealing both with cooperative and non-cooperative vehicle behaviours in large fleets of vehicles, and immunize the resulting strategies against uncertainty due to unpredictability in the vehicles' driving behaviour and due to the presence of renewable energy sources. The presence of an algorithmic tool with these features will allow for scalable charging solutions amenable to problems of practical relevance, will provide insight on the mechanism driving the response of large populations of electric vehicles, and embed robustness in the resulting charging schedules. As such, the proposed project will offer the means for reliable system operation and facilitate the integration of higher shares of renewable energy sources.
Planned Impact
The project will develop an algorithmic toolkit for analysis and synthesis of charging control strategies in large populations of electric vehicles, addressing two key issues that are faced by the industry: dealing with shared resource constraints, as it is most often the case in situations of practical relevance; and handling uncertainty in an efficient, albeit rigorous, manner. These contributions will facilitate future application and experimental validation of the developed electric vehicle charging control strategies by the energy and transportation industrial sector, and accommodate vehicle populations of growing size.
The following impact criteria are addressed:
1. Economic: Performance improvement in electric vehicle charging control methods, leading to charging cost savings, intelligent resource allocation and facilitating the integration of higher shares of renewable energy sources. These features will make the developed algorithms appealing to electric vehicle charging network operators, and to leading industrial partners in energy efficient management technologies and smart-grid solutions.
2. Societal: Robustness in charging schedules enhancing reliability. This will increase the confidence of electric vehicle users and provide cost efficient solutions, while respecting their privacy concerns regarding their consumption patterns.
3. Knowledge & Skills: Developing novel solutions for control of multi-agent systems, merging tools from different disciplines, namely, control theory, optimization and machine learning. This will set the ground for breakthrough solutions to the problem of control and performance optimization in uncertain environments with multiple agents.
4. Users: Direct engagement with industrial representatives from the energy and transportation sector, exploiting support by Siemens and Honeywell. Development of an open access algorithmic repository, allowing for benchmarking and result reproducibility; support by MathWorks will be exploited.
The following impact criteria are addressed:
1. Economic: Performance improvement in electric vehicle charging control methods, leading to charging cost savings, intelligent resource allocation and facilitating the integration of higher shares of renewable energy sources. These features will make the developed algorithms appealing to electric vehicle charging network operators, and to leading industrial partners in energy efficient management technologies and smart-grid solutions.
2. Societal: Robustness in charging schedules enhancing reliability. This will increase the confidence of electric vehicle users and provide cost efficient solutions, while respecting their privacy concerns regarding their consumption patterns.
3. Knowledge & Skills: Developing novel solutions for control of multi-agent systems, merging tools from different disciplines, namely, control theory, optimization and machine learning. This will set the ground for breakthrough solutions to the problem of control and performance optimization in uncertain environments with multiple agents.
4. Users: Direct engagement with industrial representatives from the energy and transportation sector, exploiting support by Siemens and Honeywell. Development of an open access algorithmic repository, allowing for benchmarking and result reproducibility; support by MathWorks will be exploited.
Publications
Belluschi F
(2020)
Distributed optimization for structured programs and its application to energy management in a building district
in Journal of Process Control
Deori L
(2018)
Price of anarchy in electric vehicle charging control games: When Nash equilibria achieve social welfare
in Automatica
Deori L
(2019)
Regularized Jacobi Iteration for Decentralized Convex Quadratic Optimization With Separable Constraints
in IEEE Transactions on Control Systems Technology
Edwards A
(2018)
Distributed Optimisation for Energy Management in Building Networks
Fabiani F
(2020)
On the robustness of equilibria in generalized aggregative games
Falsone A
(2023)
On the Sensitivity of Linear Resource Sharing Problems to the Arrival of New Agents
in IEEE Transactions on Automatic Control
Falsone A
(2018)
A Distributed Iterative Algorithm for Multi-Agent MILPs: Finite-Time Feasibility and Performance Characterization
in IEEE Control Systems Letters
Falsone A
(2020)
A scenario-based approach to multi-agent optimization with distributed information
in IFAC-PapersOnLine
Falsone A
(2019)
A decentralized approach to multi-agent MILPs: Finite-time feasibility and performance guarantees
in Automatica
| Description | The following key findings were achieved, providing the means to overcome existing challenges and go beyond state-of-the-art methodologies. 1) Developed intelligent scheduling algorithms for electric vehicle charging (Work Packages 1, 2.1 and 2.2 of the EPSRC award proposal) To avoid vehicles charging at time instances of high electricity price (as a result of an uncoordinated scheme), intelligent automated solutions are needed. To achieve a reliable and at the same time sustainable performance, novel algorithmic schemes based on solid mathematical foundations have been developed. Within the EPSRC award we developed and analyzed algorithms for two settings of contemporary interest: i) cooperative setting: electric vehicles belong to the same company and/or are managed by the same aggregating authority, thus cooperate in view of coordinating their charging schedule so that they collectively minimize the energy cost for the entire fleet. ii) non-cooperative setting: electric vehicles belong to different companies/aggregators, competing for shared resources (energy). We treated this case from a game theoretic perspective, providing an algorithmic solution which is guaranteed to return a Nash equilibrium solution, or in other words a solution of the least regret for all vehicles. For both cases, we adopted tools from decentralized optimization, allowing us to trade different, often conflicting objectives for each vehicle, and respect constraints pertaining the individual vehicles. Moreover, the developed skills are by construction scalable, thus amenable to parallel computation. At the same time, an important feature of the proposed algorithms is that they only based on local information, thus preventing vehicles from sharing information considered as private. 2) Quantified the price of (non-)cooperation (Work Package 2.3 of the EPSRC award proposal) Another key finding of the project is to provide a conclusive answer to a theoretical question with significant practical implications. In particular, we showed that as the number of vehicles in a fleet increases, even if vehicles act in a non-cooperative manner, the resulting strategy ends up being the same with its cooperative counterpart. This finding is based on transferring tools from the computer science literature based on the celebrated "price of anarchy" to an optimization context tailored to the electric vehicle management problem. Such a result opens the road for the development of appropriate pricing mechanisms to achieve a social welfare maximizing behaviour from self-interested users/vehicles. 3) Embedded robustness against uncertainty (Work Package 3 of the EPSRC award proposal) To the best of our knowledge, we constructed the first (probabilistically) robust charging schedule with respect to price volatility, a major challenge, as charging strategies computed based on a nominal price signals may fail to remain at an equilibrium when a different price is realized. We determined an set of equilibrium charging strategies that are robust with respect to a specific collection of price signals available through historical data, and accompanied the computed tuple of strategies with an a priori probabilistic certificate that determines the confidence with which such a solution remains an equilibrium when it comes to new price/uncertainty realizations other than those included in the data. 4) Trading robustness with performance Following the award's completion, our team has further expanded its research towards analyzing the robustness properties of equilibria in multi-agent games. The algorithms that emanated out of this award, have more general feature that facilitate their deployment to a more general class of games. In a series of publications/submissions we shifted focus towards a posteriori robustness certificates (extending key finding 3 above); the latter offers the means to trade robustness with performance, this offering cost-efficient, less conservative solutions. Moreover, our activities attempted to overcome the need for a central authority/coordinator, focusing on the development of algorithms that require communication only with neighbouring agents/vehicles. This feature embeds resilience to communication failures and accounts for time varying connectivity architectures. |
| Exploitation Route | Academic value: The developed algorithms set the foundations for further research in the field of multi-agent games arising in electric vehicle charging control problems. In particular the work related to the development of probabilistically robust charging schedules sets the first theoretical framework to account for price volatility, hence immunising the computed charging strategies with respect to uncertainty and partial information. The proposed approach is of immediate interest to researchers in game theory, machine learning and optimization, as the developed theoretical approach and proof line captures a wide class of multi-agent games over networks, not necessarily limited to electric vehicle problems. Non-academic value: Industrial representative from the energy and transportation sector like electric vehicle aggregators constitute the direct non-academic beneficiaries of the project outcomes. The developed algorithms offer a toolkit for coordinated vehicle charging, accommodating the so called peak shaving and valley filling concerns. The scalability properties of the proposed solutions and the fact that they are only based on local information, thus preventing users from sharing information considered as private, makes them amenable to parallelized computation. As such the could be integrated into high level controllers for providing consumption set points in electric vehicle fleets of realistic size. Educational value: Results originating in this project set the ground for the development of advanced control and optimization courses on distributed and robust optimization. This will have a direct educational impact when training the next generation of control engineers on topics related to control of large scale multi-agent systems. |
| Sectors | Education Energy Transport |
| URL | https://ev-charge.github.io/ |
| Description | 1) Economic: The performance improvement, as this is quantified by the developed electric vehicle charging control methods, has the potential to lead to substantial charging cost savings, and facilitate the integration of higher shares of renewable energy sources. The theoretical and algorithmic developments of the project led to a follow-up project funded by MathWorks on real-time, distributed control and game theory for multi-agent networks. Moreover, a seed-fund was acquired to foster collaboration with TU Berlin and Siemens. 2)Societal: Robustness in charging schedules (a key finding of the project), has the potential to increase the confidence of electric vehicle users while respecting their privacy concerns. 3) Education: Our developed algorithms are currently documented and serve as an open access algorithmic repository (including codes and documentation), allowing for benchmarking, result reproducibility, and contributing to interoperability activities. It should be noted, that our algorithms have direct implication on the current operational paradigm shift witnessed in the energy and transportation sector. The strong interest in the methods and tools developed by our team is reflected by the several invitation to present our findings either in dedicated seminars, or as part of courses with an international audience. Beyond the academic impact associated with these activities, participation in such events contributes to the education of the next generation of researchers, enhancing not only their technical skills but also their awareness on the challenges associated with energy and transportation problems contemporary interest, related to the penetration of electric vehicles. Besides the application of the developed methods to charging control of electric vehicles, algorithms that emanated from the research carried out within this project motivated subsequent theoretical and algorithmic developments. In particular, current research activities o our team in the area of distributed optimization and data driven algorithms for control under uncertainty have their roots to the project findings. This is also witnessed by the volume of publication submitted upon completion of the project, albeit in directly relevant topics: 5 journal submissions, 9 conference manuscripts (accepted, scheduled to appear at conference eproceedings). Further generalizing these algorithms and analyzing their relative merits is topic of current investigation. Since January 2018, our team has been invited to contributed to five courses at international Universities; one highly attended and selective summer school (SIDRA); one multi-disciplinary educational event as part of an elite program between the University of Oxford and China; two graduate courses organized at Politecnico di Milano, Italy, and one at the University of Seville, Spain. Moreover, our team has contributed to three invited sessions at international conferences. Most recently, capitalizing on some of the algorithms developed within this award, a new international graduate course within the European Embedded Control Institute was developed (scheduled to run in 2024). One tutorial and two invited session were organized as part of international conference just before the project started, introducing the attendees to the challenges and the problems to be tackled within the award. Moreover, these developments reinforced the position of the PI's team within the control and learning communities, and the PI will be acting as general co-chair of the International Conference on Learning for Dynamics & Control (L4DC) 2024. |
| First Year Of Impact | 2020 |
| Sector | Education,Energy,Transport |
| Impact Types | Societal Economic |
| Description | Distributed Control of Aerial Vehicles |
| Amount | £91,323 (GBP) |
| Organisation | MathWorks |
| Sector | Private |
| Country | United States |
| Start | 09/2022 |
| End | 12/2024 |
| Description | Trilateral University of Oxford, Tu Berlin, Siemens seed-fund |
| Organisation | Technical University Berlin |
| Country | Germany |
| Sector | Academic/University |
| PI Contribution | Successful seed-fund proposal with the PI being one of the main beneficiaries. The team of the PI contributed to theoretical/algorithmic developments/expertise on multi-agent and networked control systems, capitalizing on the outcomes of the current project. |
| Collaborator Contribution | This was a trilateral agreement with all parties (University of Oxford, TU Berlin and Siemens) contributing equally. |
| Impact | Establishment of a network of contacts and current synergy on the preparation of an EU grant proposal. |
| Start Year | 2021 |
| Description | Communication, Dissemination and Technology Transfer: Invited conference sessions; Invited lectures |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | The algorithms developed within the EPSRC award have direct implication to the current operational paradigm shift witnessed in the energy and transportation sector. The strong interest in the methods and tools developed by our team is reflected by the several invitation to present our findings either in dedicated seminars, or as part of courses with an international audience. Beyond the academic impact associated with these activities, participation in such events contributes to the education of the next generation of researchers, enhancing not only their technical skills but also their awareness on the challenges associated with energy and transportation problems contemporary interest, related to the penetration of electric vehicles. Since January 2018, our team has been invited to contributed to two courses at international Universities; one highly attended and selective summer school (SIDRA); one multi-disciplinary educational event as part of an elite program between the University of Oxford and China; and one contribution in invited sessions at international conferences (in all cases attendees were between 50-100). See for detailed description in the awards section. Moreover, the PI co-organized one tutorial and one invited session as part of an international conference just before the project started, at the IFAC World Congress 2017, introducing the attendees to the problems to be tackled within the award and the associated challenges . There are two invitations to special thematic sessions organized in upcoming international conferences, which we have accepted and manuscripts of our team have been accepted for publication. Moreover we have organized the following two invited sessions, presenting the main findings achieved during the project and immediate subsequent developments. This served as an opportunity to bring together outstanding researchers worldwide with complementary research interests to further communicate our results and maximize the academic impact of our findings. Details of these sessions can be found below: -- Co-organizer of two invited sessions on "Multi-agent and networked systems" (jointly with Dr. Filippo Fabiani, Dr. Filiberto Fele (supported by the award) and Prof. Paul Goulart), IEEE Conference on Decision and Control, Austin, Texas, United States. -- Co-organizer of an invited session on "Risk Assessment in Learning-Based Control and Decision Making" (jointly with Dr. Filippo Fabiani, Dr. Filiberto Fele (supported by the award) and Prof.Simone Garatti), IEEE Conference on Decision and Control, Cancun, Mexico. Moreover, the PI has been invited to deliver a workshop presentation and act as panel member at a workshop organized at upcoming European Control Conference, Stockholm, Sweden, 2024. The topic of the presentation will be based on joint work with Georgios Pantazis and Dr. Filiberto Fele (supported by the award). Such activities are inline with Task 4 of the award proposal. |
| Year(s) Of Engagement Activity | 2018,2019,2021,2022 |
| URL | https://2021.ieeecdc.org/ |
| Description | General co-chair International Conference on Learning for Dynamics & Control (L4DC 2024) |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Other audiences |
| Results and Impact | General conference co-chair. Record number of submissions; topic directly related to the obtained award. |
| Year(s) Of Engagement Activity | 2024 |
| URL | https://l4dc.web.ox.ac.uk/home |
| Description | Intelligent sustainable networked City (i-City) competition |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | Local |
| Primary Audience | Postgraduate students |
| Results and Impact | i-City (Intelligent sustainable networked City) is an exciting postgraduate student innovation competition running January-February 2023, sponsored by MathWorks and jointly run by MPLS division, University of Oxford and Reuben College. It involves training of post-graduate students from diverse disciplines to work in teams and find new ways to deploy AI and Machine Learning tools to solve challenges in the following three broad pillars of focus: -- Connected Mobility: How can we use AI tools to regulate traffic? Can we use aerial vehicles to monitor ground ones and guide them in real time? -- Smart Electric Vehicle Charging: Can we achieve vehicle-to-grid and/or vehicle-to-home charging? How to optimally use electric vehicles as dynamic storage devices and direct the stored power to cover grid or home needs? -- Energy Efficiency and Sustainability: How can we enhance integration of renewable energy sources? How to control buildings efficiently? How can we avoid wasting energy? |
| Year(s) Of Engagement Activity | 2023 |
| URL | https://www.mpls.ox.ac.uk/training/mpls-training/our-courses/mpls-enterprise-courses/i-city |
| Description | Invited Session Organization |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | Two invited sessions are currently being organized for the international peer-reviewed Conference on Decision and Control on "Learning with Guarantees in Control and Decision-Making". The sessions are in collaboration with Filippo Fabiani, Filiberto Fele (researcher under the current award) and Paul Goulart. |
| Year(s) Of Engagement Activity | 2021 |
| Description | New graduate course on control and optimization for multi-agent systems |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | A new graduate course, capitalizing on the developments related to the current award, was developed. Upon successful application it was selected to take place as part of the International Graduate School on Control of the European Embedded Control Institute (scheduled to run in 2024). This is the mot prestigious and well attended international graduate course series run in Europe. |
| Year(s) Of Engagement Activity | 2024 |
| URL | http://www.eeci-igsc.eu/doc/users/975/bib/2024/eeciigsc2024finalsummaries.pdf |