Variable neighborhood search for the stochastic and dynamic vehicle routing problem

Abstract

In this paper, the authors consider the vehicle routing problem (VRP) with stochastic demand and/or dynamic requests. The classical VRP consists of determining a set of routes starting and ending at a depot that provide service to a set of customers. Stochastic demands are only revealed when the vehicle arrives at the customer location; dynamic requests mean that new orders from previously unknown customers can be received and scheduled over time. The variable neighborhood search algorithm (VNS) proposed in this study can be extended by sampling for stochastic scenarios and adapted for the dynamic setting. We use standard sets of benchmark instances to evaluate our algorithms. When applying sampling based VNS, on average we were able to improve results obtained by a classical VNS by 4.39 %. Individual instances could be improved by up to 8.12 %. In addition, the proposed VNS framework matches 32 out of 40 best known solutions and provides one new best solution. In the dynamic case, VNS improves on existing results and provides new best solutions for 7 out of 21 instances. Finally, this study offers results for stochastic and dynamic scenarios. Our experiments show that the sampling based dynamic VNS provides better results when the demand deviation is small, and reduces the excess route duration by 45-90 %.

A stochastic time-dependent green capacitated vehicle routing and scheduling problem with time window, resiliency and reliability: a case study

“…Gannouni et al (2017) studied a multi-objective vehicle routing problem and solved it with a probabilistic metaheuristic approach. Sarasola et al (2016) used a (VNS) method to deal with a stochastic and dynamic VRP problem. In stochastic VRP problems usually to different probabilistic objectives are considered in a model.…”

A responsive ant colony optimization for large-scale dynamic vehicle routing problems via pheromone diversity enhancement

“…According to the categorization of approaches to DVRPs in Sect. 2, four state-of-the-art approaches to DVRPs are selected for comparison with the proposed RACO on 12 LSDVRP test instance, namely, MIACO [27] (based on diversity), 2MPSO [31] (based on memory), S-VNS [36] (based on local search), and ES-LC [48] (others).…”

“…The parameter setting in the proposed RACO is listed in Table 2. The parameters in 2MPSO, S-VNS, and ES-LC are set according to the references where 2MPSO and ES-LC were originally suggested [31,36,48], since the parameter settings in their original study have been validated to be effective in solving a set of commonly used DVRP test instances. For two ACO-based algorithms RACO and MIACO, the parameters that they have in common are set to the same values for a fair comparison, whereas their unique parameters are set the same as those suggested in the empirical study in [44].…”

Applications of Soft Computing in Intelligent Transportation Systems

“…More concretely, the authors of this paper introduce three new local search operators, which are then embedded within a VNS. In a more recent publication, Sarasola et al [65] developed a VNS for facing a stochastic and dynamic VRP. This version of the VRP contemplates two different features.…”