With the advent of 4G/5G mobile networks and vehicle-to-everything (V2X) technology, ride-sharing platforms have brought unparalleled convenience to personal mobility. However, the imbalance between traffic supply (vehicles) and traffic demand (orders) has negatively affected the efficiency of these platforms. Vehicle dispatching emerges as a potential solution to alleviate the supply-demand imbalance challenge. In this paper, we propose the Road-aware Grid-based Multi-Agent Reinforcement Learning (RGMARL) approach to effectively dispatch idle vehicles. Diverging from traditional grid-based methods, the RGMARL approach leverages road information to augment vehicle dispatching decisions. Firstly, we design an Adjacent Grid Travel Time Estimation algorithm to convert road information into grid-based travel time. Subsequently, we propose a Weight Mechanism to convert travel time into weights. And then we assign these weights to action values calculated by a Grid-based MARL algorithm. Based on the weighted action values, idle vehicles will select the dispatching action tailored to real-time road conditions and supply-demand distribution. Finally, we designed a simulator capable of replicating various traffic environments. Experimental results reveal that our proposed algorithm has an average increase of 5.75% in order response rate compared to state-of-the-art algorithms.
RGMARL: Vehicle Dispatching Based on Road Information and Supply-Demand Distribution
2023-09-24
480586 byte
Conference paper
Electronic Resource
English
VEHICLE DISPATCHING SYSTEM, VEHICLE DISPATCHING METHOD, AND VEHICLE DISPATCHING DEVICE
European Patent Office | 2023
|Vehicle dispatching system, vehicle dispatching method, and vehicle dispatching device
European Patent Office | 2023
|VEHICLE DISPATCHING DEVICE, VEHICLE DISPATCHING SYSTEM AND VEHICLE DISPATCHING PROGRAM
European Patent Office | 2023
|