Research On Cross Regional Dynamic Pricing Strategy Of Ride-hailing Based On Deep Reinforcement Learning

Online Ride-hailing has become an economical transportation way.In the online ride-hailing system,how to set the price for the riding order is a key issue.The temporal and spatial distribution of supply and demand in different regions is different,and the supply and demand in the same region also changes dynamically over time.Traditional pricing strategies cannot make reasonable decisions to set the riding prices with respect to the dynamic supply and demand in different regions,and they cannot adjust pricing strategy in the face of unbalanced supply and demand.In addition,the ride-hailing platform usually intends to maximize the long-term profit.In this thesis,we intend to address this issue.We firstly design a multi-region dynamic pricing strategy to maximize the long-term profits of the platform by considering different supply and demand in different regions,the passenger’s willingness of accepting the price and the impact of the pricing strategy on the future supply and demand in the case of fixed region partition.Since the differences between regions are also changing dynamically,we further design a region-dynamic clustering algorithm in the scenario where regions can change dynamically,which can dynamically partition regions according to the regional supply and demand,and then propose an adaptive multi-region dynamic pricing strategy to maximize the long-term platform profits.The main contents are as follows:(1)We first introduce the basic settings of the online ride-hailing system,including the process of passenger taking a ride and the work process of platform.The work process of platform is established as a round-based model.Next,we describe the specific settings of regions,orders and vehicles,and finally we give the calculation method of platform profits.(2)In the online ride-hailing system,when the region is fixed,the platform should formulate a reasonable pricing strategy according to the supply and demand between different regions and the characteristics of dynamic changes.Therefore,the problem of dynamic pricing in fixed regions is defined.We design a multi-region dynamic pricing algorithm(MRDP)combined with the Deep Deterministic Policy Gradient(DDPG)algorithm to set the order price for each region,and we run experiments based on the real data set in the central urban area of Chengdu to verify the effectiveness of the algorithm.The MRDP algorithm is evaluated against the FIX algorithm,the SDE algorithm and the GREEDY algorithm.The long-term profits of the platform under the MRDP algorithm is 10.23%,8.25% and 4.29% higher than the other three algorithms respectively.MRDP algorithm can serve more ride requests,bring a higher service rate to the platform,and use fewer vehicles to meet more riding demands.(3)In the online ride-hailing environment,since the dynamic partition of regions can better adapt to changing supply and demand states,we define the problem of dynamic regional partition and regional pricing in the scenario where regions can change dynamically.We use the Deep Q Network(DQN)algorithm to determine the number of regional clusters and then use the K-Means algorithm for clustering.We then propose the region-dynamic clustering algorithm(RDC)and adaptive multi-region dynamic pricing algorithm(AMRDP).Finally,we run experiments based on the real data set in the central urban area of ??Chengdu to verify the effectiveness of the algorithm.The experimental results show that the RDC algorithm prefer to partition the map into larger regions,which can provide passengers with more riding services.Different pricing strategies combined with the RDC algorithm can further improve the platform’s profits,which shows the effectiveness of the RDC algorithm.The combination of AMRDP algorithm and RDC algorithm can create higher profits,serve more orders and have a higher service rate,indicating that the AMRDP algorithm can adapt to the scenario where regions can change dynamically.In summary,we first design a multi-region dynamic pricing algorithm under a fixed region for the ride-hailing platform to maximize platform long-term profits.Secondly,in the scenario where regions can change dynamically,we propose a region-dynamic clustering algorithm and an adaptive multi-region dynamic pricing algorithm.The proposed algorithm can further improve the long-term profits of the platform,the number of served orders and the service rate,which also greatly improves ride-hailing experience for passengers.The work of this thesis can provide some insights for online ride-hailing platforms to design pricing strategies.