Research On Logistics Transit Time Forecasting Model Based On Divide And Conquer Strategy

Logistics is one of the important foundations for a country 's industry.In logistics management,making relevant workflows coordinate efficiently and reducing the inventory backlog of goods rely on accurate management of transit time in logistics.Therefore,efficient logistics management requires accurate transit time prediction.At present,there are few researchers in logistics transit time prediction.Although,most of the existing emphasis and application scenarios of travel prediction models differ from the focus and application scenarios of transit time prediction in logistics.But there are still many similarities between the forecast of travel time and the forecast of transit time in logistics.Based on these similarities,some of the theory and method of travel time forecast can be used to predict transit time in logistics.In this paper,we studied the solution to solve the problem of transit time prediction in logistics on the basis of existing travel time prediction theories,algorithm and models.The specific research contents are as follows:(1)The research scenarios of travel time prediction and transit time prediction in logistics were compared and analyzed and the current research on the theory and models of travel time prediction at home and abroad were studied.(2)The driving status of delivery vehicles in the transport line were tracked and recorded by the vehicle positioning equipment and related internet technology.In order to improve the quality of the original data,some data cleaning methods were used to deal with data redundancy,data loss and abnormal data and other issues of the original data.(3)We presented a vehicle driving state matrix model VDSM.The model effectively makes use of the structural characteristics of the matrix and the space-time correlation of the driving status of the transportation vehicle.Therefore,the driving state matrix can show the driving state of the vehicle intuitively from the point of view temporal and spatial.Based on the matrix,a travel time prediction model was proposed.The algorithm model was compared with several existing travel time prediction algorithms in the experimental part.The experimental results shows the performance of this matrix model is comparable to other models and in some of the transport line the performance of matrix model is significantly superior to other models.Therefore,the model can effectively forecast the transit time.(4)We proposed a combined model of travel time forecast of logistics based on the matrix model and long short-term memory networks in this paper.Compared the combined model with the general recurrent neural network model,the exponential smoothing model and ARIMA model,the experimental results shows the prediction error of the combined model is less than these single models and the stability of the matrix prediction model is better than these single model.(5)By analyzing and comparing the application scenarios of travel time prediction and logistics transit time prediction,we found that the route in most travel time prediction relatively simple.In enterprise logistics,the route is much more complex.Therefore,some of the original single models may be difficult to accurately model logistics transport lines containing a variety of complex scenes at the same time.Aiming at this feature,we used a divide and conquer strategy to predict the logistics time under complex scene.