Research On Vehicle Routing Optimization For Urban Logistics Delivery Under Multiple Constraints

Under the new economic background and logistics demand,urban logistics distribution has become an important link connecting producers and consumers.Customer satisfaction and logistics distribution costs are directly affected by distribution efficiency,and logistics distribution is playing an increasingly important role.Optimizing urban logistics distribution has important practical significance in reducing logistics operation costs and improving customer service satisfaction.However,the urban logistics vehicle routing problem faces multiple practical constraints,making its optimization more challenging.On the other hand,in the face of complex and multi constrained large-scale urban logistics distribution environments,existing research is difficult to find the optimal solution to this problem.Based on the above issues,this thesis focuses on the vehicle routing problem of urban logistics distribution,considers multiple constraint constraints,and aims to improve distribution efficiency.Combining with the characteristics of urban logistics distribution,a three-echelon logistics multi constraint vehicle routing optimization model is constructed,and a series of optimization algorithms are proposed for solving this problem.The contents and innovations of this thesis are as follows:(1)A multi-constraint three-echelon logistics distribution model(3-Echelon Rich Vehicle Routing Problem,3E-RVRP)is establishsed.This model comprehensively considers the complex road network constraints,vehicle load constraints,time window constraints,and separable constraints in the actual urban logistics distribution process,analyzes the possible logistics distribution facilities in the urban logistics distribution system,and builds a four-layer three-echelon logistics distribution model based on the logistics park,distribution center,terminal distribution network,and customers,which theoretically enriches the research of urban logistics distribution problems.(2)Aiming at the low efficiency of urban logistics distribution under multiple constraints,a modified initialization and mutation strategy based differential evolution(IIMBDE)is proposed.This algorithm has the characteristics of solving multi-objective and multi-modal problems.As the most important part of the three-echelon logistics multi-constraint vehicle routing problem,vehicle routing optimization is a key step to improve the efficiency of logistics distribution.However,it is difficult to find the optimal solution of all objectives under multiple constraints,and in some cases,the decision-makers can accept the sub-optimal solution of each objective,thus forming a solution set of multiple equivalent optimal solutions.Based on this,this thesis regards the Rich Vehicle Routing Problem(RVRP)as a multi-modal multi-objective optimization problem,and based on this,in order to obtain multiple equivalent optimal solutions,a differential evolution algorithm is proposed to expand the population size and maintain the strategy of individual multiplicity.Finally,through theoretical analysis and experimental comparison on Solomon benchmark data-set,Homberger benchmark data-set,extended VRPTW data-set and actual case datasets of different sizes,the effectiveness of the proposed algorithm is verified.(3)A two-tier intelligent algorithm called Distance Entropy based Affinity Propagation(DEBAP)-IMBDE is proposed to address the challenges of large-scale urban logistics distribution.In the upper-tier algorithm,using logistics and distribution facilities at all layers as reference points,and the average distance between each subordinate facility or customer node and the facility point as distance entropy,the entropy weight is determined,and the AP clustering algorithm is improved for the division of the overall urban logistics and distribution system.In the lower-tier algorithm,IIMBDE is used for the planning of vehicle routings at all echelons,the upper and lower tier solutions are interrelated and mutually constrained.The decision-making of the lower level algorithm must be based on the decision-making results of the upper tier algorithm,and the solving results of the upper tier algorithm will affect the path optimization results of the lower tier.At the same time,the decision results of the lower tier algorithm are also the evaluation indicators of the decision results of the upper tier algorithm,that is,the IIMBDE path optimization results will in turn affect the DEBAP partition results.Finally,through theoretical analysis and experimental comparison on Solomon benchmark data-set,Homberger benchmark data-set,extended VRPTW data-set and actual case datasets of different sizes,the effectiveness of the proposed algorithm is verified.(4)In order to further improve the efficiency of solving 3E-RVRP,an iterative heuristic Residual Edge Graph Attention Capsule Networks(Res-E-G-AT-Caps Net)deep reinforcement learning framework is improved.This framework takes the optimization results of heuristic algorithms as the initial input.Firstly,based on the idea of decomposition,the 3E-RVRP MOP is transformed into a set of scalar sub objective optimizations,and all sub objectives are synergistically optimized based on the idea of parameter transfer.Secondly,G-Caps Net is used to encode and extract features from node and edge information in the graph during sub object optimization,capturing the local position global structure correlation of graph nodes,and considering residual connections to reduce feature loss caused by information filtering.Then,a loss function suitable for multi-objective problems was designed,using the baseline reinforcement algorithm with rollout for policy network training to accelerate and stabilize model training.Different local search strategies were used to improve the model and further improve the quality of solutions.Finally,the efficiency of the proposed model was evaluated on a randomly generated instance data-set,and the generalization ability of the proposed framework was verified on standard test cases.