Research On Steel Cargo Allocation And Recommendation Based On Driver Preferences

The continuous development and progress of logistics technology has put forward higher requirements for the distribution efficiency and cost control of bulk commodity logistics represented by steel logistics enterprises.Cargo allocation is responsible for as-signing cargo to vehicles according to the loading requirements of each type of cargo,a process that combines and encapsulates cargo downward into a loading list and assigns transportation tasks upward to drivers,which is the core link in the field of steel logistics.The traditional cargo allocation scheme is manually assigned to maximize the loadable weight of each truck,ignoring driver preferences such as transportation variety preference,transportation location preference and loading queue waiting time,but these preference in-formation is closely related to the cost and profit of logistics enterprises.Therefore,this thesis designs cargo allocation and cargo recommendation schemes from two perspectives of prioritizing allocation rules and prioritizing driver preferences,respectively,to achieve cost reduction and efficiency for the platform.This thesis addresses the challenges of low utilization of transportation resources and low driver satisfaction in steel logistics enterprises,and provides a suitable cargo dispatch-ing solution for steel logistics platforms.Firstly,we extract and summarize the relevant rules in the cargo distribution process by combining the actual logistics scenarios.Sec-ondly,a multi-objective optimal cargo distribution method is designed for the distribution scenario that integrates platform and driver preferences.The method uses an optimized genetic algorithm to split,combine and then encapsulate cargo to realize the cargo distribu-tion process.Finally,two cargo sorting recommendation models are proposed to prioritize driver preferences,among which the cargo sorting recommendation model based on ma-trix decomposition and deep neural network mines driver preference features based on historical dispatch information and designs a deep neural network model to recommend cargoes that meet drivers’ preferences? the Bayesian personalized sorting model incorpo-rating user feedback increases user feedback information and introduces driver similarity calculation The process of pairwise ranking model is used to build a ranking algorithm to provide drivers with more accurate preference ranking results,thus improving the long-term economic benefits of the platform.The main work of this thesis includes the following aspects:（1）Multi-objective cargo allocation based on genetic algorithm: This thesis pro-poses a multi-objective optimization method for cargo allocation,which is designed with three optimization objectives: maximizing the total order cargo weight,mini-mizing the actual number of transportation trips,and minimizing the number of truck cargo loading and unloading places.Based on genetic algorithm,Pareto hierarchical structure,dual population evolution strategy and adaptive evolution operator are in-troduced to construct the multi-objective optimized cargo allocation scheme,and real cargo data are also used for experimental comparison.（2）Ranking recommendation based driver preferences:Cargo ranking recommenda-tion based on matrix factorization and deep neural network.In this thesis,we propose a cargo recommendation algorithm that matches driver preferences,which is divided into two phases: learning and ranking.The learning phase uses matrix factorization to mine the implied feature vectors of drivers and goods as the input of the deep neu-ral network model? the ranking phase predicts the preference degree of drivers for goods to be ranked,and generates the final set of recommended goods according to the preference degree value.（3）Bayesian personalized ranking recommendation incorporating user feedback information: In this thesis,we propose a pairwise ranking algorithm incorporat-ing user feedback information,which incorporates user feedback data and also intro-duces driver similarity,and uses a Bayesian personalized ranking algorithm to more accurately represent the corresponding ranking recommendation results in the case of current user preferences by considering the biased order relationship between items and items.In this thesis,we design cargo allocation and cargo recommendation solutions for the scenario of considering driver preferences,and verify the performance of the algorithms through extensive experiments and compare them with other algorithms.The results show that the proposed allocation and recommendation solutions can improve the utilization of transportation resources and increase driver satisfaction at the same time.