关键词: 外卖配送, 遗传算法, 人工调度, 动态配送需求

Abstract: In recent years, the food delivery industry has developed rapidly, which provides not only a convenience to the public but also a guarantee for the integrated development of offline and online catering industries. However, in the process of food delivery, due to the influence of various uncertain factors, it is difficult for the platform to dispatch orders to adapt to various complex scenarios, resulting in unreasonable allocation of distribution resources and untimely delivery services, which in turn affects user satisfaction. For example, if the food delivery is in the peak ordering period or the bad weather, or riders have an accident, or customers cannot be contacted, etc., there are often problems such as inappropriate planning for the secondary delivery of orders and the inability of riders to effectively respond to the real-time allocation needs of orders. In response to the above problems, this paper focuses on the powerful benefits of manual scheduling among riders in adjusting order allocation, planning secondary delivery routes, and managing disturbance factors. A staged optimization method is proposed to further explore the rider's cost-optimized distribution path during the dynamic adjustment process. The research content is closer to the actual food delivery problem and strives to provide a more reasonable reference plan for the planning of the platform delivery route.
For the takeaway delivery path optimization problem, based on the open PDCVRPTW problem solution strategy, this article proposes a dynamic take-out delivery path optimization model, which treats riders as a collaborative delivery set and responds to the manual scheduling demand in the actual delivery process. Furthermore, defining the objective function as the weighted incremental the sum of distance cost and time penalty cost, this article takes into account riders' manual scheduling demand, pickup and delivery cross-delivery, and other characteristics. The constraints are the capacity, the time windows, and so on. The dispatching situation and path optimization problem in the delivery process are discussed in depth. Considering the scenarios of manual scheduling, the scheduling situations are divided into two types of situations: collection point transfer distribution and emergency transfer distribution. Then, aiming at the secondary delivery problems and disturbance management problems in the process of takeaway delivery, the relevant decision-making references are discussed. A phased delivery path optimization algorithm framework is designed to meet the orderliness constraints of order generation, adaptive crossover and variation, and evolutionary learning by exchange sequence. In the verification and comparison of the model algorithm, due to the lack of real-time data sets of takeaway orders in China, a random simulation of a reasonable data set is selected for the experiment.GA, PSO, and IGA-Auto are used to solve the problem, and compared with the exact solution obtained by the Gurobi optimization solver. The optimization gap and the time used are observed, and the better heuristic algorithm and method are selected according to the effect.
The calculation example data in this paper refers to the special data in “Smart Logistics: Rider Behavior Prediction during the New Crown Period” in the Tianchi Big Data Competition. It simulates multiple order data generated in a business district within the same time period, and randomly generates each order data at the same time, for the delivery time window for the order and the location of each rider. With the IGA-Auto algorithm for staged optimization to realize secondary distribution management and disturbance management in the delivery process of takeaway orders, it is found that the total delivery cost in both scheduling situations is reduced, compared with the initial scheduling situation. Secondly, in terms of algorithm performance, while maintaining the same number of iterations and population size, the IGA-Auto algorithm has generally lower deviations from the Gurobi in terms of computing time and optimization gap and can achieve the same solution or even a better solution in some cases. At the same time, in terms of algorithm efficiency, the IGA-Auto algorithm can quickly converge and iterate continuously and can traverse large-scale paths in a short time to find a satisfactory solution. Finally, IGA-Auto is used to solve multiple calculation examples to explore the effectiveness of cost optimization. The study finds that the solution of the model and algorithm can be targeted to consider the secondary delivery management and disturbance management in the takeaway delivery process, generate optimization decisions, and reduce related delivery costs.
The data and results show that the dynamic food delivery route optimization model and algorithm considering manual scheduling can effectively deal with emergencies in the delivery process. It conforms to the rider's actual cooperation awareness and distribution situation, and can effectively solve the secondary distribution planning and emergency transfer distribution problems during the takeaway delivery process. Secondly, the effectiveness of the current model in solving such problems is verified in the comparison process of various heuristic algorithms. At the same time, the IGA-Auto algorithm has excellent performance in computing time and also shows a strong advantage in optimizing the gap. It can find the global near-optimal solution in a relatively short time, has strong overall optimization ability, and can effectively solve the problem of falling into the local optimal solution. However, the research in this paper is still limited, and there are still many factors to be considered, such as human factors related to the rider's decision-making, and the decision-making changes at the customer end caused by scheduling, etc. Future research can consider multiple scheduling scenarios, such as the exploration of collaboration methods among riders in different business districts, the systematic construction of a collaborative delivery model covering each subject, etc., to further study the impact of collaborative delivery on takeaway delivery routes.

Key words: takeaway delivery, genetic algorithms, manual scheduling, dynamic distribution requirement