Multi-Class Management With Sub-Class Service for Autonomous Electric Mobility On-Demand Systems

Syrine Belakaria, Mustafa Ammous, Lauren Smith, Sameh Sorour, Ahmed Abdel-RahimIEEE Trans. Vehicular Technology 2019 · DOI

TL;DR

This paper tackles the computational and charging bottlenecks of Autonomous Electric Mobility on-Demand (AEMoD) systems by proposing a fog-based architecture and an optimized multi-class dispatching and charging strategy. The core contribution is a novel queuing model and optimization framework that minimizes maximum response times by intelligently managing vehicle charging and customer-vehicle assignments across different vehicle and customer classes.

What Problem Does This Paper Address?

The widespread adoption of AEMoD services, which promise convenient, on-demand transportation, faces significant hurdles. Two primary challenges are:

  1. Computational and Communication Delays: Relying solely on cloud computing for the vast number of real-time decisions (dispatching, charging) in a large-scale AEMoD system can lead to prohibitive delays.
  2. Charging Infrastructure and Vehicle Availability: The sheer volume of electric vehicles (EVs) and their charging needs can overwhelm charging infrastructure, leading to long customer wait times and system instability. Traditional charging rates may not suffice for the dynamic demands of AEMoD.

Existing research has often simplified these problems by not fully considering the computational architecture, vehicle electrification, or the impact of charging limitations on system stability. This paper aims to bridge this gap by proposing a more holistic approach.

Methods

The authors introduce a novel approach that combines fog computing with a sophisticated multi-class management scheme.

  • Fog Computing Architecture: Instead of a centralized cloud, they propose distributing computational resources to "fog controllers" within each service zone. These controllers can make near-instantaneous decisions regarding dispatching and charging based on local information (customer requests, vehicle SoC, charging availability). This addresses the computational and communication delay bottleneck.
  • Multi-Class Queuing Model: A detailed queuing model is developed to represent the AEMoD system. This model accounts for:
  • Multiple Vehicle Classes: Vehicles are categorized by their State of Charge (SoC).
  • Multiple Customer Classes: Customers are categorized by their trip distance and the required SoC to complete it.
  • Sub-Class Service: Vehicles can serve customers from their own class or any "lower" class (requiring less SoC).
  • Charging Decisions: Vehicles can either serve immediately or undergo partial/full charging at dedicated charging points.
  • Optimization Framework: The core of the solution lies in an optimization problem formulated to minimize the maximum expected response time across all customer classes. This is achieved by optimizing:
  • Charging Decisions (q): The probability of a vehicle with a certain SoC partially or fully charging.
  • Dispatching Decisions (Π): The probability of a vehicle being dispatched to serve a customer of a specific class.
  • The optimization uses KKT conditions and Lagrangian analysis to derive lower-bound analytical solutions, which can then be iteratively tightened to find feasible optimal solutions.

Datasets

Not discussed in the paper. The evaluation is based on simulations.

Key Insights and Ideas

  • Fog Computing is Crucial for Real-time AEMoD: Distributing computation to the edge (fog) is essential for handling the massive, real-time decision-making required by AEMoD systems, significantly reducing delays compared to cloud-only approaches.
  • Intelligent Charging and Dispatching are Intertwined: The paper demonstrates that optimizing charging decisions (when and how much to charge) and dispatching decisions (which vehicle serves which customer) together is key to system stability and minimizing response times.
  • Sub-Class Dispatching Offers Significant Gains: Allowing vehicles to serve customers requiring less SoC than their current charge level (sub-class dispatching) is a critical strategy for improving vehicle utilization and reducing customer wait times, especially under high demand.
  • The Number of Classes Matters: While increasing the number of vehicle/customer classes can improve granularity, there's a point of diminishing returns where too many classes can negatively impact performance and increase response times.
  • Resilience to Charging Infrastructure Failures: The proposed sub-class dispatching model shows improved resilience to disruptions in charging availability, maintaining better customer satisfaction even when charging resources are reduced.

Key Results

  • The proposed optimized multi-class management scheme (optimized charging and sub-class dispatching) significantly outperforms various non-optimized policies and previously proposed schemes.
  • Performance gains (reduction in maximum expected response time) of up to 94.4% were observed compared to other policies.
  • The sub-class dispatching model demonstrated resilience, achieving up to a 65% gain in performance during critical scenarios of reduced charging availability.
  • The study validates the importance of the fog computing approach for managing AEMoD systems effectively.

Limitations and Future Work

  • Analytical Solution Tightening: While the paper derives lower-bound analytical solutions using KKT conditions, it acknowledges that these may not always be the exact optimal solution for non-convex problems. The paper suggests using iterative methods like "Suggest-and-Improve" to tighten these solutions to the feasible set.
  • Simulation-Based Evaluation: The results are primarily based on simulations, and real-world deployment would be necessary to fully validate the proposed system.
  • Future Work: The paper implies that further exploration into the optimal number of classes and more advanced iterative optimization techniques could be beneficial. The resilience aspect highlighted in the simulations also suggests potential for further research into robust AEMoD system design.

Summary updated

BibTeX
@article{belakaria2019multi,
  author = {Syrine Belakaria and Mustafa Ammous and Lauren Smith and Sameh Sorour and Ahmed Abdel-Rahim},
  title = {{Multi-Class Management With Sub-Class Service for Autonomous Electric Mobility On-Demand Systems}},
  journal = {IEEE Trans. Vehicular Technology},
  year = {2019},
  doi = {10.1109/tvt.2019.2920104},
}
Figure