Overview
The research of our group focuses on building the pathway to a reliable integrated-decentralized power system. In particular, we aim at developing new control, optimization, and machine learning theories and bridging them to the building blocks including smart buildings, vehicle-grid integration (VGI), microgrids, cyber-physical security, and grid-edge resources (GERs) integration. We seek to reform the power system to improve public health and reduce greenhouse gas (GHG) emissions, make electric services more resilient and secure, reduce long-term energy costs, and put our economy on a dynamic and sustainable footing.
1. On Scalability of Large-Scale Multi-Agent Cooperative Optimization in Networked Environment
Large-scale multi-agent cooperative optimization problems broadly exist in industrial applications, e.g., electric vehicle charging control and distributed energy resource control for grid services. In such problems, edge agents collaborate to reach the globally optimal solution with both the local and network constraints considered. The industrial usefulness of such setups can only be justified by involving a vast number of agents, creating unprecedented scalability challenges. Usually, the fomulation of such problems are strongly coupled in the sense of strongly couple, non-separable objective fuctions and network constraints. Though existing decentralized/distributed optimization algorithms can achieve scalability w.r.t. agent population size, they fall short in concurrently achieving optimality and convergence, converging in a limited number of iterations, and achieving scalability w.r.t. network dimension. Along this line, we are working towards two-facet scalable decentralized/distribured algorithms for large-scale multi-agent cooperative optimization in networked environment.
[P1] X. Huo* and M. Liu, “Two-facet scalable cooperative optimization of multi-agent systems in the networked environment”, arXiv preprint arXiv: 2010.06157 [math.OC], pp. 1-17, 2020.
[J14] M. Liu, P. K. Phanivong, Y. Shi, and D. S. Callaway, “Decentralized charging control of electric vehicles in residential distribution network,” IEEE Transactions on Control Systems Technology, vol. 27, no. 1, pp. 266-281, 2019.
[C7] M.Liu, “Chance-constrained SPDS-based decentralized control of distributed energy resources,” in Proceedings of the IEEE Conference on Decision and Control, Nice, France, Dec. 11-13, 2019.
[C6] M.Liu and M. Sahraei-Ardakani, “Chance-constrained shrunken-primal-dual subgradient (CC-SPDS) approach for decentralized electric vehicle charging control,” in Proceedings of the IEEE PES Innovative Smart Grid Technologies Asia, Chengdu, China, May. 21-24, 2019.
[C5] M.Liu, P. K. Phanivong, and D. S. Callaway, “Customer- and network-aware decentralized EV charging control,” in Proceedings of the Power System Computation Conference, Dublin, Ireland, Jun. 11-15, 2018.
[C4] M. Liu, P. K. Phanivong, and D. S. Callaway “Electric vehicle charging control in residential distribution network: A decentralized realization,” in Proceedings of the IEEE Conference on Decision and Control, Melbourne, Australia, Dec. 12-15, 2017.
2. On Privacy Preservation of Decentralized/Distributed Optimizations
Controlling a vast number of grid-edge resources (GER) has a huge potential in provisioning grid-level services. However, one of the main barriers to large-scale implementation of GER coordination and control lies in privacy and data confidentiality, which are also the main pillars for successful GER integration and control. In a centralized control framework (suppose it can achieve scalability), end users’ private information, e.g., device nameplate information and electric vehicle battery information, must be collected by the central operator beforehand, leading to serious privacy concerns; In a decentralized/distributed control framework, end users either need to communicate with other or with a central operator to exchange intermediate decisions, e.g., proposed energy use patterns, leading to possible privacy breaches. To promote the concept of scalable GER control for more reliable distribution network operation, we are working towards privacy-preserving decentralized/distributed optimization algorithms to protect end users’ privacy against honest-but-curious agents, external eavesdroppers, and even the central operator: 1. We target establishing integrated cryptography-based decentralized/distributed optimization algorithms; 2. We target developing non-cryptography-based decentralized/distributed optimization algorithms based on primal splitting and dual splitting.
[P2] X. Huo* and M. Liu, “Distributed privacy-preserving electric vehicle charging control based on securet sharing,” arXiv preprint arXiv: 2110.02280 [eess.SY], pp. 1-7, 2021.
[U2] X. Huo* and M. Liu, “Privacy-preserving multi-agent optimization in cyber-physical systems – a synthesis of cryptography and decentralized optimization”, submitted to IEEE Transactions on Industrial Informatics, pp. 1-8, 2021.
[J17] X. Huo* and M. Liu, “Privacy-preserving decentralized multi-agent cooperative optimization – paradigm design and privacy analysis,” IEEE Control Systems Letters, vol. 6, pp. 824-829, 2021.
[C12] X. Huo* and M. Liu, “A novel cryptography-based privacy-preserving decentralized optimization paradigm,” in Proceedings of the IEEE International Conference on Industrial Cyber-Physical Systems, Victoria, BC, Canada, May 10-12, 2021. Best Student Paper Award
[C10] X. Huo* and M. Liu, “Privacy-preserving decentralized multi-agent cooperative optimization – paradigm design and privacy analysis,” accepted, IEEE Conference on Decision and Control, Austin, TX, USA, Dec. 13-17, 2021.
[C9] X. Huo* and M. Liu, “Privacy-preserving decentralized optimization using homomorphic encryption,” in Proceedings of the IFAC Workshop on Cyber-Physical & Human Systems, Beijing, China, Dec. 3-5, 2020.
3. On For-Purpose Cyber-Security of GER control
As the power grid moves toward a distributed fashion, more control rights are shifted to GERs and end users. The deep digitalization and paradigm shift surrender the power system to vulnerabilities associated with cyber-attacks. Cyber-attacks targeting the power grid are growing in number and sophistication and may lead to a cascade of failures, e.g., massive blackout, infrastructure destruction, market impacts, and socioeconomic impacts. Without strong cyber-security means, GERs are the most vulnerable point in the deeply digitalized power grid. Beyond addressing scalability and privacy, it is crucial that decentralized/distributed GER control algorithms be designed to be able to identify, localize, and mitigate the cyberattacks. Yet, cyber-security research is far outpaced by the development of optimization algorithms, and research on the intersection of the two fields, i.e., the secure decentralized optimization, is still nascent. Most widely investigated power system cyber-attacks, such as DoS, repeated transmission, and FDI, target general purposes such as paralyzing the overall control or estimation framework. Due to the simple objective, these attacks normally adopt brute force mechanisms and have obvious attacking characteristics, thus can be easily detected and mitigated and their long-term impacts are limited. In contrast, an emerging type of FDI attacks that have specific purposes rely on delicate mathematical models so that their attack behaviors are stealthy and imperceptible. These for-purpose threats can stem from either external attackers or internal participating agents, aiming at system stability jeopardy or self-interest; they are chronic threats to the system – if not detected and mitigated in a timely manner, long-term impacts will be made. GER control based on decentralized optimization is especially prone to for-purpose attacks as the algorithm iterations offer a hotbed. To significantly improve the technical baseline of for-purpose cyber-attack research, we are investigating novel approaches to identify various attack vectors that can be launched in decentralized optimization algorithms, and developing corresponding attack detection and mitigation strategies.