Modified ant colony optimization algorithms for deterministic and stochastic inventory routing problems / Lily Wong

• Main Author: Lily , Wong
• Format: Thesis
• Published: 2018
• Subjects: HD28 Management. Industrial Management; Q Science (General)
• Online Access: http://studentsrepo.um.edu.my/8446/1/All.pdf; http://studentsrepo.um.edu.my/8446/6/lily.pdf; http://studentsrepo.um.edu.my/8446/

Inventory routing problem (IRP) integrates two important components of supply chain management: routing and inventory management. In this study, a one-to-many IRP network comprises of a single depot (warehouse) and geographically dispersed customers in a finite planning horizon is presented. Multi products are transported from the warehouse by using a fleet of a homogeneous vehicle which located at the ware house to meet customer’s demand on time. The customers are allowed to be visited more than once in a given period and the demand for each product is deterministic and time varying. The problem is formulated as a mixed integer programming problem and is solved using CPLEX to obtain the lower and upper bound (the best integer solution) for each instance considered. The classical Ant Colony Optimization (ACO) is modified by including the inventory cost in the global pheromones updating is proposed in this study. The sensitivity analysis on important parameters that influence decision policy in ACO in order to choose the appropriate parameter settings is carried out. Among the two proposed algorithms, that is, ACO and ACO2, ACO2 outperform than ACO. Both ACO and ACO2 perform better on large instances compared to the upper bound and perform equally well for small and medium instances. In order to improve the proposed algorithms, population based ACO where the ants are subdivided into subpopulations and each subpopulation represents one inventory level is proposed. In addition, a new formulation for customer’s inventory pheromones is proposed and the selection of inventory updating mechanism is based on these pheromone values. The computational results show that the algorithms which implement this new formulation are able to produce better solutions. The computational results also show that the algorithms of population based ACO performs better than the algorithms of non-population based ACO. The deterministic IRP model is then extended to solve the Stochastic Inventory Routing Problem (SIRP). The demands in SIRP are modeled by some probability functions and due to the stochastic nature of customer demands, the service levels constraint where it limits the stock out probability at each customer and the probability of overfilling the stock of each customer is introduced in this study. A two phase algorithm named SIRPACO1 is proposed to solve the SIRP. Phase I solved the inventory sub problem to determine the quantity to be delivered to each customer as well as inventory level at each customer while Phase II employs the population based ACO to determine the routes for each period. The algorithm was further enhanced by incorporating the inventory updating mechanism into Phase II with the aim of obtaining a set of inventory level which will give minimum overall cost and named as SIRPACO2. The computational experiments are tested on different combinations of two important parameters that are standard deviation and service level. The computational results showed that the enhanced SIRPACO2 gave better performance compared to SIRPACO1.