Computer Simulation And Optimization Of O2O Order Distribution Platform

The popularity of the Internet provides the foundation for the development of online to offline(O2O)business.Due to the rapid increase in the number of users,the pressure and challenges faced by the platform are also increasing rapidly.The efficiency of order processing directly determines the comprehensive competitiveness of the platform.A complete meal delivery system includes two parts: order dispatching and take out delivery.The purpose of this thesis is to make the order dispatching and path planning more scientific and efficient,and then improve the efficiency of order processing.In terms of order dispatching,the current dispatching mode of each platform is the combination of courier's will and platform dispatching.The dispatching method is nearby dispatching.When the order quantity is large,this allocation method will cause a large time cost.This system fully considers the relationship between the courier's route and the order to be allocated.In the aspect of route planning,the current delivery route of riders is completely decided according to the wishes of couriers.the proposed dispatching method plans the courier's path and recommends the result to the courier.Through the above improvements,the unnecessary time and distance cost caused by the subjective judgment of the courier can be reduced.In this thesis,from the perspective of complex adaptive system,the open source simulation software SUMO is used to establish the computer simulation model of the delivery platform based on agent,and the visualization processing is carried out.For agent modeling,a complete meal delivery system includes customer module,courier module,merchant module and order dispatching system.This thesis mainly studies the order dispatching and courier module.For the courier agent,its main function is to deliver the takeaway to customers on time.According to the change of external environment information,the agent changes some of its own parameters,such as travel speed,travel distance and so on.The goal is to minimize the cost of time and distance.For order dispatching system,our goal is to make order dispatching more scientific and efficient,and then improve the efficiency of order completion.After the model is established,for order dispatching and single courier's path planning,this thesis uses simulated genetic algorithm to solve the traveling salesman problem.Firstly,the traditional genetic algorithm is improved to accelerate its convergence speed and avoid falling into local optimum as much as possible.Moreover,multi-thread / process and GPU related parallel processing are used to further improve the running speed of the algorithm and reduce the running time.By improving the genetic algorithm and combining with multi-thread / process and GPU related parallel processing,the PGA algorithm is proposed.By comparing the convergence speed of PGA algorithm and traditional genetic algorithm,the superiority of the improved algorithm is determined.In order to deal with a large number of order dispatching and multi-courier path planning,at the same time,in order to avoid the waiting caused by centralized data processing and the data delay caused by the need to transmit data to the central processor when the data volume is too large,This thesis proposes E-DQN algorithm by combining edge computing,deep learning and reinforcement learning.The advantages of the improved algorithm are verified by comparing E-DQN algorithm with ordinary reinforcement learning algorithm.The main contributions of this thesis are as follows:1.A PGA algorithm for order dispatching and single courier path planning is proposed.The algorithm improves the traditional genetic algorithm,combines multi-process / thread parallel and GPU,and effectively improves the speed and accuracy of order dispatching and path planning.2.An E-DQN algorithm is proposed for large order dispatching and multi-courier path planning.By combining reinforcement learning with deep learning and introducing potential function to improve the reward function,the algorithm makes the state space continuous in path planning.Combined with edge computing equipment,the algorithm effectively improves the speed and accuracy of processing a large number of order dispatching and multiple couriers' path planning.