Enhancing genetic algorithms based solutions for multi source flexible multistage logistics network models

Abstract

A multistage logistics network problem deals with determining the optimal routes for product delivery to customers through a network of multiple facilities namely plants, distribution centers and retailers. The optimal routes should maximize revenues or minimize costs to a business or logistics provider. The flexible multistage logistics network (fMLN) problem is an extension of the traditional multistage logistics network whereby a customer can procure goods directly from plants or distribution centers needless of retailers. It is well known that fMLN problem is NP-hard, thus, it requires, for a large size problem, a non-polynomial time to solve analytically. In addition, an fMLN problem usually involves optimization that has a large number of constraints and decision variables. Previous researchers have attempted to use soft computing approaches namely Genetic Algorithms (GA) to address the fMLN problem. In terms of modeling, previous research considered fMLN problem with single source assumption, whereby each customer would be served by only one facility. In reality, a customer may be served by a number of facilities or by multi source and can order a number of different products. Besides that, business or logistics provider is required not only to minimize the total logistics costs but also other criteria such as the total delivery time simultaneously. Under these circumstances, the fMLN problem becomes more complex, and the standard GA could not perform reasonably well due to a decreasing the quality of solution. In this research a single source fMLN problem is extended to cater for multi source, multi product and multi objective fMLN cases. It is proven that the standard GA and the previous chromosome representation could not be used to solve the extended fMLN problems. Here, two new chromosome representations were proposed and implemented on GA with penalty method. In addition, heuristic rules were developed and embedded into GA to cope with the constraints in the fMLN problems. The experimental results showed that the proposed chromosome representations and the heuristic rules have substantially improved the GA performance in terms of running time and solution quality for the extended fMLN problems.