On GPU Implementation of the Island Model Genetic Algorithm for Solving the Unequal Area Facility Layout Problem

Abstract

Facility layout problem (FLP) is one of the hottest research areas in industrial engineering. A good facility layout can achieve efficient production management, improve production efficiency, and create high economic values. Because FLP is an NP-hard problem, meaning it is impossible to find the optimal solution when problem becomes sufficiently large, various evolutionary algorithms (EAs) have been proposed to find a sub-optimal solution within a reasonable time interval. Recently, a genetic algorithm (GA) was proposed for unequal area FLP (UA-FLP), where the areas of facilities are not identical. More precisely, the GA is an island model based, which is called IMGA. Since EAs are still very time consuming, many efforts have been devoted to how to parallelize various EAs including IMGA. In recent work, Steffen and Dietmar proposed how to parallelize island models of EAs. However, their parallelization approaches are preliminary because they focused mainly on comparing the performances between different parallel architectures. In addition, they used one mathematical function to model the problem. To further investigate on how to parallelize the IMGA by GPU, in this paper we propose multiple parallel algorithms, for each individual step in the IMGA when solving the industrial engineering problem, UA-FLP, and conduct experiments to compare their performances. After integrating better algorithms for all steps into the IMGA, our GPU implementation outperforms the CPU counterpart and the best speedup can be as high as 84.

Quality-Oriented Study on Mapping Island Model Genetic Algorithm onto CUDA GPU

“…In the latest literature, we first proposed how to use GPUs to parallelize the IMGA for solving UA-FLP [32]. The structure of our parallel IMGA on a GPU is illustrated in Figure 5.…”

“…In IMGA, selection is one of the most time-consuming and quality-influencing operations. The parallelisms of the two most common selection methods on a GPU have been discussed in our previous work [32]. In the following sections, we firstly proposed an effective method of generating random seeds for the tournament method on a GPU, which can ensure us how to find a better quality of solution, within a shorter execution time.…”

“…The IMGA can take full advantage of the computing abilities by exploiting coarse and fine grained parallelisms of GPU. However, only a very limited number of studies have focused on how to reduce the execution time of IMGA on GPU [32][33][34]. In our previous work [32], we have proposed two different parallel algorithms for each individual step of IMGA when solving UA-FLP on GPU.…”

“…However, only a very limited number of studies have focused on how to reduce the execution time of IMGA on GPU [32][33][34]. In our previous work [32], we have proposed two different parallel algorithms for each individual step of IMGA when solving UA-FLP on GPU. Which parallel algorithm has a better performance for each step has also been reported.…”

Power Transformer’s Electrostatic Ring Optimization Based on ANSYS Parametric Design Language and Response Surface Methodology

“…When the analytical expression for the relationship between the optimization variables and optimization objectives cannot be found, the other type of methods, the global optimization methods, or the response surface methodology(RSM)combined with FEM can be used for the optimization design. Global optimization methods such as the genetic algorithm [12][13][14][15][16] and the particle swarm optimization algorithm [8,[17][18][19] need to call a large number of modeling and optimization programs repeatedly, which significantly affect the efficiency of optimization. Most importantly, without appropriate fitness functions the local search ability may become worse and the search efficiency may be reduced.…”