Decentralized optimal management of a large-scale EV fleet: optimality and computational complexity comparison between an Adaptive MAS and MILP

Increasing the penetration of variable and uncertain renewables and electric vehicles in power systems may give rise to problems (such as network congestion and commitment mismatches) if not controlled strategically. This demands control solutions in the form of energy management strategies for active distribution networks which would control the connected distributed energy resources and storage units in real-time to address the mentioned challenges. Centralized strategies may fail to serve this purpose for large-scale distribution networks due to their inherent shortcomings like vulnerability to single point of failures and large computing times. Unlike centralized approaches, decentralized control strategies show more potential. This paper presents one such solution, based on an adaptive multi-agent system, to control a large-scale distribution network in real-time. Its performance is compared with the results obtained with the corresponding centralized optimization problem, modeled as a mixed integer linear programming problem. Both the centralized version and the decentralized multi-agent version of the problem under consideration are presented and a case study is designed for the comparison. The comparison shows that the designed multi-agent system produces a near-optimal solution in real-time while the centralized optimization strategy struggles in terms of computational complexities for larger distribution networks.