[Paper Review] SOUP: Spatial-Temporal Demand Forecasting and Competitive Supply

Bolong Zheng, Qi Hu+5 more|arXiv (Cornell University)|Sep 24, 2020

Transportation and Mobility Innovations53 references4 citations

TL;DR

This paper proposes SOUP, a framework for spatial-temporal demand forecasting and competitive supply in ride-hailing systems. It introduces ST-GCSL, a graph convolutional network model for accurate request prediction across space and time, and DROP, a demand-aware route planning algorithm that minimizes agent idle time by balancing supply and demand, outperforming state-of-the-art methods on real-world data with up to 26.8s reduction in waiting time.

ABSTRACT

We consider a setting with an evolving set of requests for transportation from an origin to a destination before a deadline and a set of agents capable of servicing the requests. In this setting, an assignment authority is to assign agents to requests such that the average idle time of the agents is minimized. An example is the scheduling of taxis (agents) to meet incoming requests for trips while ensuring that the taxis are empty as little as possible. In this paper, we study the problem of spatial-temporal demand forecasting and competitive supply (SOUP). We address the problem in two steps. First, we build a granular model that provides spatial-temporal predictions of requests. Specifically, we propose a Spatial-Temporal Graph Convolutional Sequential Learning (ST-GCSL) algorithm that predicts the service requests across locations and time slots. Second, we provide means of routing agents to request origins while avoiding competition among the agents. In particular, we develop a demand-aware route planning (DROP) algorithm that considers both the spatial-temporal predictions and the supplydemand state. We report on extensive experiments with realworld and synthetic data that offer insight into the performance of the solution and show that it is capable of outperforming the state-of-the-art proposals.

Motivation & Objective

• To minimize the average idle time of agents (e.g., taxis) in dynamic ride-hailing systems.
• To forecast spatial-temporal demand patterns accurately across urban road networks.
• To reduce competition among agents by balancing supply and demand during route assignment.
• To develop a data-driven framework that integrates real-time demand forecasting with intelligent agent routing.
• To outperform existing methods in reducing idle time, waiting time, and request expiration rates.

Proposed method

• Proposes a multi-level partitioning strategy to define spatial regions and temporal time slots for granular request modeling.
• Introduces ST-GCSL, a Spatial-Temporal Graph Convolutional Sequential Learning model that captures spatial-temporal dependencies using graph convolution and sequence modeling.
• Designs DROP, a demand-aware route planning algorithm that uses real-time spatial-temporal predictions and supply-demand balance to assign optimal routes to idle agents.
• Employs an offline ST-GCSL model for long-term demand forecasting and an online DROP module for real-time agent repositioning.
• Uses graph convolution to model non-Euclidean spatial relationships in road networks, improving prediction accuracy over CNN-based methods.
• Combines graph convolution with sequential modeling to capture temporal dynamics, enabling efficient and accurate forecasting of future requests.

Experimental results

Research questions

• RQ1 How can spatial-temporal demand patterns in urban transportation networks be accurately predicted using deep learning?
• RQ2 To what extent can a graph convolutional network model capture complex spatial-temporal dependencies in taxi demand data?
• RQ3 How can route planning for idle agents be optimized to reduce competition and idle time?
• RQ4 Can a framework that integrates real-time demand forecasting with supply-aware routing outperform existing dispatching strategies?
• RQ5 What is the impact of supply-demand balance on reducing agent idle time and request expiration?

Key findings

• The ST-GCSL model outperforms baseline models in spatial-temporal demand forecasting accuracy on real-world datasets.
• The DROP algorithm reduces average idle time by up to 10 seconds compared to the baseline RD method on the New York dataset.
• DROP decreases average waiting time by 26.8 seconds compared to the RD method, significantly improving passenger experience.
• The framework reduces the request expiration percentage by 0.73% compared to the RD method, indicating better request fulfillment.
• The integration of ST-GCSL and DROP leads to a substantial improvement in overall system efficiency and agent utilization.
• The proposed method effectively mitigates the herding effect by balancing agent distribution across high-demand regions.