[Paper Review] Communication-Censored-ADMM for Electric Vehicle Charging in Unbalanced Distribution Grids

Abhishek Bhardwaj, Wilhiam de Carvalho+3 more|arXiv (Cornell University)|Jul 26, 2022

Electric Vehicles and Infrastructure4 citations

TL;DR

This paper proposes Communication-Censored ADMM (CC-ADMM) for decentralized electric vehicle (EV) charging in unbalanced distribution grids, reducing peer-to-peer communications by up to 80% while maintaining voltage regulation and minimizing time-of-use costs. The method integrates reactive power control and communication censoring to preserve convergence and solution accuracy under sparse communication patterns.

ABSTRACT

We propose an alternating direction method of multipliers (ADMM)-based algorithm for coordinating the charge and discharge of electric vehicles (EVs) to manage grid voltages while minimizing EV time-of-use energy costs. We prove that by including a Communication-Censored strategy, the algorithm maintains its solution integrity, while reducing peer-to-peer communications. By means of a case study on a representative unbalanced two node circuit, we demonstrate that our proposed Communication-Censored-ADMM (CC-ADMM) EV charging strategy reduces peer-to-peer communications by up to 80%, compared to a benchmark ADMM approach.

Motivation & Objective

• To address the growing communication overhead in ADMM-based distributed EV charging coordination, especially in large-scale networks.
• To maintain voltage regulation within safe limits (±5%) in unbalanced distribution grids during EV charging and discharging.
• To minimize time-of-use energy costs for EV users through optimal charging scheduling.
• To reduce peer-to-peer communication costs without sacrificing solution accuracy or convergence.
• To prove theoretical convergence of the communication-censored ADMM algorithm to the optimal solution.

Proposed method

• Extends the ADMM-based EV coordination framework from [17] by incorporating inverter-based reactive power control for improved voltage regulation.
• Introduces a communication censoring mechanism that allows EVs to skip broadcasting when changes in dual variables are below a threshold, reducing redundant transmissions.
• Applies the communication-censored ADMM framework from [21] to the EV charging problem, ensuring convergence to the optimal solution under censoring.
• Reformulates the centralized EV (dis)charging problem into an ADMM-compatible form with consensus constraints and dual decomposition.
• Uses a peer-to-peer communication model where each EV communicates only with neighbors, and implements a dynamic censoring rule based on dual variable changes.
• Employs a convergence proof based on Lyapunov stability and subgradient methods, showing that CC-ADMM converges to the optimal solution despite reduced communication.

Experimental results

Research questions

• RQ1 Can communication censoring be applied to ADMM-based EV charging coordination without degrading solution quality or convergence?
• RQ2 To what extent can communication overhead be reduced in decentralized EV charging while maintaining voltage regulation and cost minimization?
• RQ3 How does the censoring strategy affect the voltage profile and system stability in unbalanced distribution networks?
• RQ4 Does the CC-ADMM algorithm preserve convergence guarantees under reduced communication frequency?
• RQ5 How do different communication topologies (fully connected vs. sparse) impact the performance and communication savings of the CC-ADMM approach?

Key findings

• CC-ADMM reduced peer-to-peer communications by 79% (to 21%) compared to the benchmark ADMM algorithm in a fully connected network.
• In a sparse communication network (70 neighbors per EV), CC-ADMM reduced communications to 17% of the benchmark’s level.
• Both the benchmark and CC-ADMM solutions maintained voltages within ±5% of nominal voltage, ensuring safe operation.
• The voltage profile under CC-ADMM was flatter and closer to 1 p.u. than the benchmark, indicating improved voltage regulation.
• Censoring was most active in early iterations, with communication increasing only when significant changes in dual variables occurred.
• Theoretical analysis confirms that CC-ADMM converges to the optimal solution of the EV (dis)charging problem despite reduced communication.