Deep Reinforcement Learning For Dynamic Pricing

With the development of the mobile Internet,many industries have shifted their sales and trading business online,forming O2 O business model.Some large-scale O2 O platforms in China,as the third parties,have accumulated a large amount of data concerning transactions,consumers,sellers.Therefore,online trading in the era of big data has formed the basis for the application environment of real-time dynamic pricing,which has caused automated dynamic pricing of e-commerce platforms more and more popular.When faced with increasing complexity of environment of big data application scenarios on e-commerce platforms,on the one hand,the probability distribution of data becomes complicated and changeable,on the other hand,the features of data are sampled from highdimensional space,so it is difficult to deal with a dynamic pricing problem with traditional methods.In recent years,it is effective to apply deep reinforcement learning based on big data in many fields.This paper applies deep reinforcement learning theory to study dynamic pricing problem.Firstly,this paper analyzes the O2 O business model for a cold drink application scenario on Group-buying websites,and establishes a model of real-time dynamic pricing problems under limited inventory.Secondly,the dynamic pricing problem is formulated as a Markov decision process,and the components of the Markov decision process for the dynamic pricing problem are constructed.Thirdly,a simulation environment for dynamic pricing problems is designed and a dynamic pricing algorithm for application scenarios is developed based on deep reinforcement learning.Finally,this paper carries out the simulation experiment and makes an analysis of the experimental results.The experimental results show that the proposed dynamic pricing algorithm based on deep reinforcement learning performs well in the simulation environment and has certain robustness against changes in online demand distribution and random risks.Therefore,modeling dynamic pricing problems based on deep reinforcement learning is feasible and has good application value for dynamic pricing problems in the big data environment.