Research On Online Scheduling Model Based On Machine Learning Of Vehicle Outbound Logistics

With the continuous growth of automobile production and sales in China,the national demand for domestic cars is on the rise.This poses higher requirements for the operation and service level of automotive vehicle outbound logistics.The core problem in vehicle logistics is the vehicle dispatching problem.In order to cope with the ever-increasing transportation demand and guarantee service satisfaction,in recent years,automobile logistics enterprises have accelerated the construction of informatization,and they have replaced the traditional manual scheduling mode with intelligent dispatching mode based on mathematical models and exact algorithms to improve efficiency and reduce transportation costs.At present,intelligent dispatching is still in its infancy.At present,only static and offline dispatching problem is solved,and it is easy to fall into the predicament of local optimization.In order to overcome these shortcomings and provide a reference for the next phase of intelligent dispatching,this paper studies online dispatching problem of vehicle logistics based on machine learning methods.By means of forecasting useful information in the future on the data-driven techniques and designing mathematics model and algorithms,we can further optimize process loading plan from the angle of global optimization in order to shorten the mileage,reduce carpool situation and transportation costs.First of all,this paper analyzes the operation process,basic information,constraints and optimization objectives of the automobile vehicle logistics system.On the basis of good understanding for the whole system,the problem is divided into two parts: the prediction problem and the online dispatching problem.In the part of prediction problem,an efficient combination forecasting model based on several machine learning algorithms is proposed to predict the order of cars on the next day.The prediction results are used as input for subsequent dispatching problems.In the part of online dispatching problems,relevant constraints are refined according to actual business rules.The pre-dispatching constraints and optimization objectives based on the prediction results were further summarized.A mixed-integer linear programming model was established,and the practical problems were converted into combinatorial optimization mathematics problems that can be solved by algorithms.Secondly,we design efficient and reliable algorithms to solve the two problems respectively.In the prediction problem,we choose three kinds of machine learning algorithms: deep neural network,support vector regression and random forest to establish forecasting model respectively.Then,based on the idea of combination forecasts,a combination forecasting model is proposed to forecast the second day's orders.Furthermore,bootstrap method is used to calculate the confidence intervals and finally output the predicted results of the next day's orders.In the online dispatching problem,a heuristic algorithm and a branch-and-bound algorithm for solving the problem are designed.For the heuristic algorithm,the steps of the algorithm and the relevant processes are discussed.For the branch and bound algorithm,according to the characteristics of the problem,the branch strategies,dominance rules,search patterns,upper and lower bound are discussed to ensure high efficiency of the algorithm.Finally,numerical experiments and case study are applied to verify the efficiency and reliability of the model and algorithm.In the part of prediction problem,a cross-validation experiment was designed to compare the performance of the individual prediction model and the combination forecasting model in order to select the optimal model.Then,the accuracy of the prediction is evaluated to evaluate the efficiency of the prediction results.In the part of online dispatching problem,large-scale experiments are applied to analyze the algorithm's accuracy and speed of solution,furthermore,the algorithms designed are compared with the CPLEX optimization tools.In case study,we verified the validity of the models and algorithms in the actual scenario through examples based on real data.Experiments and case study results prove that the mathematical models and algorithms studied in this paper are effective and reliable.