Multiagent Reinforcement Learning-Based Taxi Predispatching Model to Balance Taxi Supply and Demand

Abstract

With the improvement of people’s living standards, people’s demand of traveling by taxi is increasing, but the taxi service system is not perfect yet; taxi drivers usually rely on their operational experience or cruise randomly to find passengers. Without macroguidance, the role of the taxi system cannot be fully utilized. Many scholars have studied taxi behaviors to find better operational strategies for drivers, but their researches rely on local optimization methods to improve the profit of drivers, which will lead to imbalance between supply and demand in the city. To solve this problem, we propose a Multiagent Reinforcement Learning- (MARL-) based taxi predispatching model through analyzing the running data of 13,000 taxis. Different from other methods of scheduling taxis based on the real-time location of orders, our model first predicts the demand for taxis in different regions in the next period and then dispatches taxis in advance to meet the future requirement; thus, the number of taxis needed and available in different regions can be balanced. Besides, in order to reduce computational complexity, we propose several methods to reduce the state space and action space of reinforcement learning. Finally, we compare our method with another taxi dispatching method, and the results show that the proposed method has a significant improvement in vehicle utilization rate and passenger demand satisfaction rate.

Taxi Demand Prediction Based on a Combination Forecasting Model in Hotspots

“…Previous studies on taxi demand prediction are generally based on historical taxi trajectory data. Previous studies have shown the feasibility of obtaining predictions from historical taxi trajectory data [1,[5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23]. Methods of traffic demand prediction can be classified into three types: linear system theory (such as the autoregressive moving average model [24], Kalman filtering model, and time series model), nonlinear system theory (such as the neural network model, gray prediction model, and random forest model (RFM)), and combination forecasting model (CFM).…”

Dispatching Model in Online Taxi System by Using Stable Marriage and First Price Sealed Bid Auction Models

“…The first agent-based dispatch model can be called as limited-movement-grid (LMG) model [9]. In it, the map is divided into square-shaped grids with specified size.…”

“…This dispatching model is then implemented into online taxi dispatch system simulation. In this simulation, the proposed model is compared with the stable marriage [5], limited-movement-grid [9], freemovement-grid [21], circle-shaped-maximumrevenue [17], and circle-shaped-nearest [22] models. These second to fifth models are the agent-based models explained in the related work section.…”

“…Unfortunately, many of these studies in the dispatching system used only distance or time between the pickup and the current vehicle locations in matching the request and the vehicle. It occurs in studies that adopted either FCFS [4], stable marriage [5], or agent-based model [9]. The reason is the only aspect that is concerned by the passenger is waiting time and the only aspect that is concerned by the driver is the pickup cost [9].…”

“…It occurs in studies that adopted either FCFS [4], stable marriage [5], or agent-based model [9]. The reason is the only aspect that is concerned by the passenger is waiting time and the only aspect that is concerned by the driver is the pickup cost [9].…”