[Paper Review] Proactive rebalancing and speed-up techniques for on-demand high capacity ridesourcing services

Yang Liu, Samitha Samaranayake|arXiv (Cornell University)|Feb 9, 2019

Transportation and Mobility Innovations11 references4 citations

TL;DR

This paper proposes speed-up techniques and a probabilistic proactive rebalancing method to enhance a real-time high-capacity ridesourcing fleet management framework. By pruning the search space and reducing I/O overhead in parallelization, computation time is reduced by up to 97.67%, while proactive rebalancing based on historical demand probabilities increases service rate by 4.8% and reduces waiting time and total delay by 5.0% and 10.7%, respectively.

ABSTRACT

We present a probabilistic proactive rebalancing method and speed-up techniques for improving the performance of a state-of-the-art real-time high-capacity fleet management framework [1]. We improve on both computational efficiency and system performance. The speed-up techniques include search-space pruning and I/O cost reduction for parallelization, reducing the computation time by up to 97.67%, in experiments on taxi trips in New York City. The proactive rebalancing routes idle vehicles to future demands based on probabilistic estimates from historical demand, increasing the service rate by 4.8% on average, and decreasing the waiting time and total delay by 5.0% and 10.7% on average, respectively.

Motivation & Objective

• To improve computational efficiency of a real-time, high-capacity ridesourcing fleet management system.
• To reduce computation time in solving large-scale vehicle routing and assignment problems.
• To enhance system performance by proactively relocating idle vehicles to high-demand areas using historical demand patterns.
• To minimize I/O bottlenecks in parallelized request-vehicle compatibility checks.
• To increase service rate and reduce waiting and delay times through predictive vehicle repositioning.

Proposed method

• Two search-space pruning techniques are applied to reduce the number of generalized pickup-and-delivery problems evaluated during trip feasibility computation.
• An I/O reduction technique is introduced to minimize redundant data transfer in parallel request partitioning by reducing overlapping vehicle data across compute units.
• A probabilistic proactive rebalancing method uses marginal probability estimates of future request assignments to guide idle vehicles to high-utility locations.
• The method leverages historical demand distributions to predict future demand hotspots and aligns vehicle positions with expected future utilization.
• The framework integrates these techniques into the state-of-the-art anytime optimal fleet management system from [1], preserving optimality while accelerating computation.
• Virtual requests are used in the ILP formulation to model future demand, improving rebalancing effectiveness.

Experimental results

Research questions

• RQ1 Can search-space pruning techniques reduce computation time in high-capacity ridesourcing systems without sacrificing solution quality?
• RQ2 How can I/O overhead in parallelized request-vehicle compatibility checks be minimized to improve scalability?
• RQ3 To what extent can probabilistic proactive rebalancing based on historical demand improve system performance metrics like service rate, waiting time, and delay?
• RQ4 How do the proposed speed-up techniques scale with increasing fleet size and demand complexity?
• RQ5 What is the impact of combining speed-up techniques with proactive rebalancing on overall system efficiency and performance?

Key findings

• The speed-up techniques reduced computation time by up to 97.67% in experiments with a 1000-vehicle, 10-capacity fleet under high demand constraints.
• The average computation time reduction across all tested configurations was 87.46%, with no loss in solution optimality.
• Proactive rebalancing increased the service rate by 4.8% on average, corresponding to 4068 additional passengers served in 6 hours.
• Waiting time decreased by 5.0% on average, and total delay was reduced by 10.7% on average across all test cases.
• The largest reduction in waiting time (24.28 seconds) and total delay (68.51 seconds) occurred with a 3000-vehicle, 4-capacity fleet under high demand constraints.
• The integration of speed-up and proactive rebalancing achieved an 81.44% average reduction in computation time compared to the original framework.