Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Real-time Operation of Electric Autonomous Mobility-on-Demand System Considering Power System Regulation

Lyuzhu Pan, Hongcai Zhang

TL;DR

This work addresses the risk that real-time, centralized charging of electric autonomous mobility-on-demand fleets can destabilize urban power networks. It introduces a coupled real-time framework based on a Markov decision process for EAMoD dispatch and a second-order cone program for power-system regulation, incorporating depot charger configurations with fast and slow charging. To overcome computational challenges, it develops a piecewise-linear value function ADP with dual-variable–driven slope updates and couples it with MPC to improve charging decisions, achieving near-optimal performance with much faster computation than long-horizon MPC. Case studies on Manhattan and a 14-node distribution network show that the proposed method substantially improves profit and service quality while preventing undervoltage, thanks to coordinated charging strategies that flexibly switch between fast and slow charging. The results highlight the practical potential of integrated real-time coordination between mobility-on-demand systems and power grids, enabling scalable, grid-aware deployment of EAMoD services.

Abstract

Electric autonomous mobility-on-demand (EAMoD) systems are emerging all over the world. However, their potential swarm charging in depots may deteriorate operation of the power system, further in turn affecting EAMoD system's optimal operation. To prevent this latent risk, we develop a real-time coordination framework for the EAMoD system and the power system. First, the temporal-spatial characteristics of EAMoD fleets are fully described based on a Markov decision process model, including serving trips, repositioning, and charging. Second, charger accessibility of EAMoD depot charging is well modeled as real-world configuration, wherein fast and slow charge piles are both included. Third, the power system regulation model provides real-time charging regulation constraints for EAMoD systems to prevent potential overload and undervoltage. To address the poor solution quality attributed to the complex decision space of the EAMoD system, this paper proposes a piecewise linear-based approximate dynamic programming algorithm combined with model predictive control. Numerical experiments in the Manhattan and a 14-node power distribution network validate the effectiveness of the proposed algorithm and underscore the necessity of system coordination.

Real-time Operation of Electric Autonomous Mobility-on-Demand System Considering Power System Regulation

TL;DR

This work addresses the risk that real-time, centralized charging of electric autonomous mobility-on-demand fleets can destabilize urban power networks. It introduces a coupled real-time framework based on a Markov decision process for EAMoD dispatch and a second-order cone program for power-system regulation, incorporating depot charger configurations with fast and slow charging. To overcome computational challenges, it develops a piecewise-linear value function ADP with dual-variable–driven slope updates and couples it with MPC to improve charging decisions, achieving near-optimal performance with much faster computation than long-horizon MPC. Case studies on Manhattan and a 14-node distribution network show that the proposed method substantially improves profit and service quality while preventing undervoltage, thanks to coordinated charging strategies that flexibly switch between fast and slow charging. The results highlight the practical potential of integrated real-time coordination between mobility-on-demand systems and power grids, enabling scalable, grid-aware deployment of EAMoD services.

Abstract

Electric autonomous mobility-on-demand (EAMoD) systems are emerging all over the world. However, their potential swarm charging in depots may deteriorate operation of the power system, further in turn affecting EAMoD system's optimal operation. To prevent this latent risk, we develop a real-time coordination framework for the EAMoD system and the power system. First, the temporal-spatial characteristics of EAMoD fleets are fully described based on a Markov decision process model, including serving trips, repositioning, and charging. Second, charger accessibility of EAMoD depot charging is well modeled as real-world configuration, wherein fast and slow charge piles are both included. Third, the power system regulation model provides real-time charging regulation constraints for EAMoD systems to prevent potential overload and undervoltage. To address the poor solution quality attributed to the complex decision space of the EAMoD system, this paper proposes a piecewise linear-based approximate dynamic programming algorithm combined with model predictive control. Numerical experiments in the Manhattan and a 14-node power distribution network validate the effectiveness of the proposed algorithm and underscore the necessity of system coordination.

Paper Structure

This paper contains 16 sections, 15 equations, 10 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Electric Autonomous Mobility-on-Demand Dispatch Model
  4. State
  5. Decision
  6. Transition Function
  7. Objective Function
  8. Power System Regulation Model
  9. Solution Algorithm
  10. Piecewise-Linear Value Function Approximation
  11. Approximate Dynamic Programming with Model Predictive Control
  12. Experiments
  13. Case Study Configuration
  14. Algorithm Performance Analysis
  15. Impact Analysis of Power Network Constraints
  16. ...and 1 more sections

Figures (10)

  • Figure 1: Interdependency between the EAMoD system and the power system.
  • Figure 2: Illustration of EAMoD dispatch model. The simple model consists of three zones with three SoC levels.
  • Figure 3: Piecewise-linear value function approximation.
  • Figure 4: Manhattan with 61 taxi zonestlc.
  • Figure 5: A 14-node distribution networkzhang2018pev.
  • ...and 5 more figures