Scalable Network-Constrained Electric Vehicle Charging in Multiphase Distribution Grids

Liang Zhang, Vassilis Kekatos, Georgios B. GiannakisarXiv 2015 · arXiv

Abstract

Although electric vehicles are considered a viable solution to reduce greenhouse gas emissions, studies have shown that their uncoordinated charging will have a noticeably adverse influence on power network operation. To guarantee the secure and economic operation of distribution grids, a network-constrained electric vehicle scheduling problem is formulated here. Leveraging a recently proposed linearized distribution grid model, the goal is to coordinate electric vehicle charging while minimizing the power supply cost, and respecting critical voltage regulation, power flow, and unbalance constraints. Optimizing variables across grid buses is accomplished using the alternating direction method of multipliers. Provisioning large-scale fleets, the task of scheduling co-located vehicles is handled by the recently revived Frank-Wolfe method. The derived charging scheme features affordable computational requirements from the vehicle controller and minimal information exchanges. Numerical tests demonstrate the efficiency of the novel schemes, and confirm the necessity of orchestrating vehicle charging across phases.

Introduction

As electric vehicles (EVs) become more common, "uncoordinated" charging—where users plug in as soon as they arrive home—poses a significant threat to power grid stability. This behavior can lead to massive voltage fluctuations, transformer failures, and "unbalance" in the three-phase systems typical of residential distribution. Existing solutions often struggle with scalability or oversimplify the grid as a "balanced" single-phase system, which ignores the physical realities of multiphase distribution and can lead to equipment damage.

This paper introduces a scalable, decentralized framework for managing EV charging across complex, multiphase power grids. By combining a linearized grid model with the Alternating Direction Method of Multipliers (ADMM) and the Frank-Wolfe algorithm, the authors propose a system that respects real-world constraints while converging significantly faster than previous state-of-the-art methods.

Methodology

The proposed framework employs a three-layered mathematical approach to ensure computational efficiency and privacy:

  • Linearized Multiphase Model: A linear approximation of power flow equations is used to account for the complexities of three-phase grids (Phases A, B, and C) without the heavy computational burden of non-linear AC power flow equations.
  • ADMM (Spatial Decomposition): The Alternating Direction Method of Multipliers allows for decentralized optimization. Each grid node (bus) solves a local problem and shares minimal data with neighbors to maintain global grid limits.
  • Frank-Wolfe Algorithm (Fleet Scaling): To manage large EV fleets at a single location, the Frank-Wolfe algorithm is utilized. Unlike standard methods requiring expensive "projections," Frank-Wolfe uses a simple sorting-based update, making it exceptionally fast for large-scale scheduling.
The two-step ADMM process enables decentralized spatial decomposition across grid buses.

Experimental Setup

The framework was validated using standard benchmarks and real-world data:

  • IEEE 13-bus feeder: A standard, highly unbalanced test case for distribution networks.
  • Southern California Edison (2014): Residential load data for realistic base demand.
  • Federal Highway Administration: Statistical travel data to model EV plug-in times, travel distances, and required state of charge.

Key Insights and Results

A critical insight of this work is the necessity of orchestrating EV charging across phases. Explicitly managing phase unbalance prevents neutral current buildup that wastes energy and damages infrastructure. Furthermore, the authors found that for EV charging, the optimal direction in optimization steps can be determined through simple sorting of time slots by marginal cost, rather than complex projections.

The results demonstrate significant performance gains:

  • Speedup: The Frank-Wolfe-based update was 100 times faster than the "Optimal Decentralized Charging" (ODC) benchmark, with average update times dropping from 1.8 seconds to 3.5 microseconds.
  • Convergence: The decentralized ADMM algorithm reached global optimality within approximately 2,000 iterations.
  • Grid Health: The algorithm successfully "filled the valley" of the load profile, charging vehicles during low-demand periods while maintaining voltage and phase balance within safety limits.
Comparison of relative cost value error shows the proposed algorithm converges faster than the ODC benchmark.

Conclusion and Future Work

While the linearized model is highly efficient, its accuracy may decrease under extremely stressed grid conditions. Future research will focus on moving toward asynchronous updates to better support real-time deployment where different grid components may have varying data availability. Additionally, exploring the system's resilience to communication delays or packet loss will be vital for practical implementation.

Summary updated

Abstract

Although electric vehicles are considered a viable solution to reduce greenhouse gas emissions, studies have shown that their uncoordinated charging will have a noticeably adverse influence on power network operation. To guarantee the secure and economic operation of distribution grids, a network-constrained electric vehicle scheduling problem is formulated here. Leveraging a recently proposed linearized distribution grid model, the goal is to coordinate electric vehicle charging while minimizing the power supply cost, and respecting critical voltage regulation, power flow, and unbalance constraints. Optimizing variables across grid buses is accomplished using the alternating direction method of multipliers. Provisioning large-scale fleets, the task of scheduling co-located vehicles is handled by the recently revived Frank-Wolfe method. The derived charging scheme features affordable computational requirements from the vehicle controller and minimal information exchanges. Numerical tests demonstrate the efficiency of the novel schemes, and confirm the necessity of orchestrating vehicle charging across phases.
BibTeX
@misc{zhang2015scalable,
  author = {Liang Zhang and Vassilis Kekatos and Georgios B. Giannakis},
  title = {{Scalable Network-Constrained Electric Vehicle Charging in Multiphase Distribution Grids}},
  year = {2015},
  eprint = {1510.00403},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1510.00403},
}
Figure