Algorithm portfolios for logistics optimization considering stochastic demands and mobility allowance
Introduction
There are several NP hard problems in the areas of graph theory, scheduling and coding theory for which a computationally efficient solution has not been found or shown to be non-existent (Stinson, 1987). This uncharacteristic behaviour is due to the stochastic nature of problem, its size and complexity. Further, the nonlinear characteristics of the objective function of NP hard problems also contribute to an increase in computational complexity. Common examples of non-deterministic engineering problems with uncertainties include vehicle routing problems with stochastic demands (VRPSDs) (Moghaddama et al., 2012, Goodson et al., 2012, Yang et al., 2000), inventory routing problems (Shukla et al., 2012, Agghezzaf et al., 2006), travelling salesman problem (TSP) of varying size (De Berg, 2005), and lot-sizing problems with stochastic demands (Raa, 2005). This paper focuses on developing a decision support methodology for resolving varying instances of the VRPSD problem in a computationally efficient way. Next, we briefly discuss the VRPSDs followed by the need for algorithm portfolios and relevant literature.
The Vehicle Routing Problem (VRP) deals with the transportation of goods and services between geographically dispersed cities or customer locations by means of a fleet of vehicles. Solution to such a problem means determining the best set of possible vehicle routes, servicing all customers and optimising related constraints such as vehicles capacity, time windows, driver’s maximum working time, etc. VRPs are of major focus in supply-chain systems today, and they are becoming increasingly complex and challenging. For this reason, it has attracted various researchers to develop routing models that are more dynamic, stochastic, and incorporate all constraints, thereby enhancing the computational complexity associated with the objective function.
Although several VRPSD models exist, they mainly consider the routing of the vehicle to be a flat path that ensures smooth movement of the vehicles. Based on the general performance and design of vehicles, approximate velocity is assigned to these vehicles. The assumption of smooth flow of vehicles may be useful in some but not necessairly all cases. Usually, ground movement of vehicles is hampered by areas of uneven terrain, which diminishes the effective movement of the vehicle (or the distance traversed is increased). Therefore, the distance traveled by vehicles must be approximated in close range with a view to use some distribution function that is able to fully map the variations originated due to uneven paths. This paper models such variations using the concept of mobility allowance that views the land disruptions as an extra distance that the vehicles have to cover. In this paper, levy's distribution function has been used to model various types of land disruptions. The theoretical base of this approach bridges the existing research gap and brings closer a practical solution to VRPSDs by considering and modelling issues of mobility allowance.
It is widely recognized that the VRP is one of the most challenging problems to solve (Lim and Wang, 2005). Conceptually, VRP is an NP hard problem that can be viewed as the combination of the Travelling Salesman Problem (TSP) and the Bin Packing Problem (BPP). It is believed that one may never find a computational technique that will guarantee optimal solutions to larger instances for such problems. Even for small fleet sizes and a moderate number of transportation requests, the planning task is highly complex. The optimal strategies do find their application in resolving academic problems of insignificant dimensions, but the real world problem demands more robust heuristic and metaheuristic approaches to solve such problems in the required time frame. The increasing use of the metaheuristics has dramatically reduced the time taken to resolve these problems without much depreciation in the solution quality.
The main contributions of this study are as follows:
- •
This research introduces the concept of mobility allowance in vehicle routes to mathematically formulate the terrain uncertainties using the Levy distribution function. Thus providing more logical and mathematical grounds for encountering discrepancies in land moves. A mathematical model for estimating the extra distance traversed by vehicles has been also presented to incorporate the complexities.
- •
This research provides the mechanism for employing key metaheuristics running on multiple processors for providing effective and efficient solutions than the individually running algorithms.
- •
The proposed method is a more reliable and efficient decision support methodology in terms of producing quick and reliable solutions for complex and dynamically changing instances of VRPSDs. It takes the advantage of combining metaheuristic search approaches with several processors to arrive at the best solution in minimum computational time. Therefore, this research extends the existing literature in vehicle routing by providing a strong and useful alternative to tradiational VRP solutions.
Further, this paper also provides mechanisms for algorithm portfolio design, mathematical and statistical evaluation, and analysis for the vehicle routhing problems with stochastic demands and mobility allowance.
The rest of the paper is organized as follows: Section 2 discusses literature review and identifies current reserch gaps. Section 3 presents mathematical modeling of Stochastic Vehicle Routing Problem (SVRP). Section 4 gives a detailed modeling and approximation on the mobility allowance for the vehicle routings. Section 5 describes the concept of algorithm portfolios, details the implementation of the four basic metaheuristics and their advanced variants with the new neighborhood generators, and explains the experimental background used in this paper. Section 6 details the wider insights of the study to the comparative performance of the metaheuristics, and analyzes the functioning of the portfolios along with suggestions to optimal soultuions. Finally, Section 7 concludes the paper with some discussion on future research directions.
Section snippets
Literature review
VRP models mainly focusing on the stochastic nature are described in literature as stochastic vehicle routing problems (SVRPs) (Stewart and Golden, 1983, Secomandi, 2000, Bent and Hentenryck, 2004, Kenyon and Morton, 2003, Lim and Wang., 2005, Sungur et al., 2008, Erera et al., 2010). As defined by Stewart and Golden (1983), a VRP is stochastic when the demands at individual customer locations behave as random variables, and the routes must be determined before the values of these random
Mathematical modeling of VRPSD
A VRPSD is defined on a complete graph G=(V,A,C), where V={0,1,…,n} is a set of nodes i.e. (customers) with node 0 denoting the depot, A={(i,j):i,j∈V,i≠j} is the set of arcs joining the nodes, and C={cij:i,j∈V,i≠j} the travel costs (distances) between nodes. The cost matrix C is symmetric and satisfies the triangular inequality. A fleet of TNV homogeneous vehicles, each with capacity Λ has to deliver goods to ncustomers according to their stochastic demands, minimizing the total expected
Mobility allowance in land moves
The mobility allowance is an innovative concept introduced into the vehicle routine problem with stochastic demands. The innovative twist is that the vehicles are allowed to deviate from the given route. Usually, the land moves and distances taken up in most of the research, even though straight, cannot be avoided from having small disruptions. Thus, the interpretation of the routes have to be modified to map the real situations. These variations are attributed to various factors like uneven
Problem characteristics and instances explored
In the absence of any commonly used benchmarks for VRPSD problems we have generated our own test problems. These problem instances are generated by controlling four factors that govern the difficulty of the VRPPSD and include: (a) customer locations, (b) capacity over demand ratio, (c) variance of stochastic demand, and (d) number of customers. The position of customers are randomly assigned in a uniformly discretized space.
In this research, three different problems are designed by varying four
Portfolio computing: results, discussions and statistical insights
To test the performance of the proposed methodology on the VRPSD, three problems of different dimensionality and complexity are simulated as discussed in Section 5.1.
Conclusion and future research
This paper presents a critical decision making combinatorial optimization problem with focus towards generating feasible and optimized solutions for vehicles moving in a stochastic environment. The paper proposes an algorithm portfolio to solve the vehicle routing problem with stochastic demand (VRPSD). The AHP based approach is utilized in this research to determine the best portfolio to get the result for a given set of processors.
The VRPSD considered in this research generalizes the
References (47)
- et al.
A genetic algorithm for the vehicle routing problem
Computers & Operations Research
(2003) - et al.
TABU search algorithm for the multi-trip vehicle routing and scheduling problem
European Journal of Operational Research
(1997) - et al.
TSP with neighbourhoods of varying size
Journal of Algorithms
(2005) - et al.
Invited review: stochastic vehicle routing
European Journal of Operational Research
(1996) - et al.
A TABU search heuristic for the vehicle routing problem with stochastic demands and customers
Operations Research
(1996) - et al.
Algorithm portfolios
Artificial Intelligence
(2001) - et al.
A two-stage algorithm with valid inequalities for the split delivery vehicle routing problem
International Journal of Production Economics
(2007) - et al.
A memetic algorithm for the multi-compartment vehicle routing problem with stochastic demands
Computers &Operations Research
(2010) A hybrid genetic algorithm for the vehicle scheduling problem with due times and time deadlines
International Journal of Production Economics
(2001)The algorithm selection problem
Advances in Computers
(1976)
Stochastic vehicle routing: comprehensive approach
European Journal of Operational Research
Modelling inventory routing problems in supply chains of high consumption products
European Journal of Operational Research
Scenario-based planning for partially dynamic vehicle routing with stochastic customers
Operations Research
Improvement heuristics for the vehicle routing problem based on simulated annealing
European Journal of Operational Research
Scheduling of vehicles from a central depot to a number of delivery points
Operations Research
Learning and Optimization Using the Clonal Selection Principle
IEEE Transactions on Evolutionary Computation, Special Issue on Artificial Immune Systems
The Vehicle Routing Problem with Stochastic Demand and Duration Constraints
Transportation Science
Genetic Algorithm and Engineering Design
An exact algorithm for the vehicle routing problem with stochastic demands and customers
Transportation Sciences
Cited by (30)
Redesigning product line for integrated manufacturer-supplier ecosystem in a centralized supply chain: Case of an industrial consumer product
2024, International Journal of Production EconomicsHybrid evolutionary algorithm for large-scale project scheduling problems
2020, Computers and Industrial EngineeringCitation Excerpt :Although MMRCPSP is more practical than its single-mode counterpart, it is more complex and challenging. According to Chakrabortty, Abbasi, and Ryan (2019), determining a feasible solution for an MMRCPSP with two types of non-renewable resources is NP-complete, and solving such a problem using a computationally efficient approach is very challenging (Shukla, Choudhary, Prakash, Fernandes, & Tiwari, 2013). Nevertheless, the majority of real and hypothetical MMRCPSPs have up to nine non-renewable resources (Peteghem & Vanhoucke, 2014).
Sustainable agro-food supply chain design using two-stage hybrid multi-objective decision-making approach
2018, Computers and Operations ResearchCitation Excerpt :explicitly applied this method to solve their models. Other works existing in the literature deal with operational and tactical planning of supply chains [51–53]. [2] and [20] conducted an exhaustive review of the application of planning models in the agro-food supply chain and mainly focused on the models that have been successfully implemented.
Designing bijective S-boxes using Algorithm Portfolios with limited time budgets
2017, Applied Soft Computing JournalCitation Excerpt :Alternatively, the algorithms of the AP can communicate by trading elite solutions at the cost of execution time, as in the models proposed in [19,30]. Non-interactive APs have also been applied on problems that originate from Operations Research [28,29] and Combinatorial Optimization [31]. It has been shown that the lack of information exchange can be beneficial in some applications where the search can rapidly become biased in narrow neighborhoods of the search space.
Resolving forward-reverse logistics multi-period model using evolutionary algorithms
2017, International Journal of Production EconomicsCitation Excerpt :García-Rodríguez et al. (2013) showed that application of reverse logistics can be beneficial in acquiring raw materials in developing countries as it can reduce the problem of acquisition of production inputs and mitigate environmental damage caused by the production of raw materials. A number of researchers have also investigated vehicle routing problem in reverse logistics operations (Dethloff, 2001; El-Sayed et al., 2010; Shukla et al., 2013; Tiwari and Chang, 2015; Soysal et al., 2015; Kim and Lee, 2015). Since vehicle routing is an essential element of reverse logistics operations, it is important that manufacturing organizations manage this efficiently.
An optimization model for a monopolistic firm serving an environmentally conscious market: Use of chemical reaction optimization algorithm
2015, International Journal of Production EconomicsCitation Excerpt :The extant literature has witnessed developments in this domain, for example comparison of the outputs of different RSOTs, namely Genetic Algorithm, Simulated Annealing and Tabu search, for solving large combinatorial optimization problems (Tyagi et al., 2012). Recently, an algorithm portfolio methodology, based on various evolutionary algorithms, has evolved (Shukla et al., 2013a, 2013b) which incorporates existing meta-heuristics: Genetic Algorithm, Simulated Annealing, Artificial Immune System, Tabu search. Through various experiments on different sizes of problems, it validates the different properties of the algorithm portfolio.