Study On Logistics Path Optimization And Delivery Time Prediction

With the rapid development of e-commerce,problems arising in logistics distribution system have attracted increasingly attentions.A good logistics system can not only reduce delivery costs,but also improve customer experience.This thesis focuses on two problems related to logistics system.One problem is the path optimization problem of delivery system and the other is the prediction of package delivery time.When the direction of path is not taken into account,a path optimization problem can be divided into two categories: the traveling salesman problem(TSP)and the traveling salesman problem with time windows(TSPTW).At present,some algorithms,such as Tabu Search Algorithm and Ant Colony Algorithm,are very well suit for the problem of TSPs without time window.However,models and optimization methods for TSPTW problem are relatively few.By improving Freud Algorithm and Tabu Search Algorithm,this thesis constructs a model and develops corresponding algorithms to solve TSPTW problems arising in logistics system.After verified with small data set(7 points),it is evidently that the two improved algorithms can effectively solve the targeted problems.In addition,after been verified in data(66 points)from real world logistics system,the Freud Algorithm exhibits better performance under require time constrains,meanwhile the Tabu Search Algorithm performances is better for short distance requirement.Based on this finding,this study proposes a new algorithm that integrates the merits of the two aforementioned methods.The final results obtained in this thesis can be visualized with Java code.The visualized results also illustrate the feasibility of the proposed algorithm.The other problem studied in this thesis is the prediction of delivery time.An accurate delivery time forecasting model is of great significance for improving customer experience and delivery efficiency.This thesis investigates three time prediction models: a decision tree based machine learning model,a multiple regression analysis model and an autoregressive time series model.Our experiment results show that accuracy of all three models is better than the benchmark model(averaging the time at the same site).After auto turning hyper-parameter,the decision tree model performs well,and the average absolute error obtained is about 10 minutes.Prediction errors of the other two methods are also less than 10 minutes.Three models are all better than the 15-minute error from the benchmark model.In data processing part,this study uses improved K-means clustering to segment order areas.By optimizing the initial clustering center and the number of clusters,clustering results are more compliant with the actual requirements.After comprehensively analyzing performance of the three models,and taking into account of the feasibility of experiments in a real world,we suggest that the decision tree method is the best one for batch forecasting in real world applications.This thesis focuses on provides a better solution to these two problems related to logistics system.One problem is the path optimization problem of delivery system and the other is the prediction of package delivery time.A better solution can not only reduce delivery costs,but also improve customer experience.A further research on path optimization recommends taking the vehicle allocation problem into account,and turning into a vehicle routing problems with time windows(VRPTW).A further research on the prediction of package delivery time can also take the path in to account.In order to adjust optimization plans more effectively,Combining the prediction of package delivery time with the path optimization problem of delivery system and doing simulation experiments through offline simulation platforms(simulating weather conditions,traffic conditions,etc.)...