Research On Dynamic Heterogeneous Green Vehicle Routing Optimization Based On Deep Reinforcement Learning

Logistics transportation is the main component of modern service industry,which is very important to the development of national economy.China’s logistics industry has developed rapidly in recent years.At the same time,as a big country with energy consumption and carbon emissions,energy shortage and environmental pollution are becoming increasingly serious.It is extremely urgent to develop green and low-carbon logistics transportation mode.Research on vehicle routing problem is one of the effective ways to realize energy saving and emission reduction and green transportation.Existing studies on vehicle routing problems based on deep reinforcement learning mostly focus on reducing transport costs and transport time,with little consideration of carbon emissions.Moreover,they are static homogeneous fleet problems in nature,which cannot solve problems such as dynamic change of customer demand and sudden congestion in the process of actual logistics distribution.This paper studies the dynamic heterogeneous green vehicle routing problem,in order to achieve energy conservation and emission reduction and rational vehicle scheduling,improve transportation efficiency and reduce environmental pollution caused in the transportation process.The specific research contents are as follows:Based on the basic vehicle routing problem,the dynamic heterogeneous green vehicle routing problem is analyzed in detail and its mathematical modeling is carried out.In order to solve the problem that model factors are not considered in the carbon emission model,a heterogeneous version of the comprehensive calculation model of fuel consumption and carbon emissions is introduced.The carbon dioxide emissions in the process of vehicle transportation are converted into green costs for path optimization,and mathematical tools such as optimization theory are used to find environmentally friendly green paths.Aiming at the problem that the dynamic heterogeneous green vehicle routing problem has NP(non deterministic polynomial)attribute and the traditional heuristic algorithm is difficult to plan the high-quality distribution scheme,DHGRL(reinforcement learning for dynamic heterogeneous green vehicle routing problem)algorithm is designed to solve it.The algorithm is based on the improved encoder decoder architecture and senses the dynamic changes of customer and traffic flow information through the attention mechanism.For the selection of different vehicles and customer nodes in the solution process,the vehicle selection decoder and node selection decoder are designed.The vehicle decoder obtains the probability that a specific vehicle is selected.The node decoder calculates the probability distribution of all unserved nodes by giving the node embedded from the encoder and the vehicle selected from the vehicle decoder.Aiming at the problem of slow convergence of algorithm model training,the reinforce algorithm with baseline is used for training,which improves the efficiency of model training.The simulation results show that,compared with the current classical deep reinforcement learning method,the method proposed in this paper reduces the carbon emission cost by12.65% and the total distribution cost by 12.41%.Compared with traditional heuristic algorithm,carbon emission cost and total distribution cost can be reduced by 20.68% and18.36% respectively.It effectively reduces the carbon emission cost and total distribution cost in the process of vehicle transportation,reduces environmental pollution and improves the efficiency of logistics distribution.