A modified squirrel search algorithm based on improved best fit heuristic and operator strategy for bin packing problem

Abstract

Bin Packing Problem (BPP) is one of the main optimization problems in which a set of items of known sizes should be packed into a minimum number of bins. In which the total size of items in each bin is not exceeding the bin capacity. Several approximation algorithms have been suggested that provide an approximate solution to the BPP. Nevertheless, they are not applicable to solve large-scale instances within a reasonable time or obtain an optimal number of bins. In this paper, we propose a modified version of the most recent optimization algorithm squirrel search algorithm (SSA) for solving the one-dimensional bin packing problem (MSBPP). The proposed algorithm is based on an improved best fit heuristic for generating a feasible initial packing of items into bins. During improving the solution, the operator’s strategy can take place to obtain an optimized solution for the BPP. To the best of our knowledge, this paper is the first work of applying squirrel search algorithm in a bin packing problem as an example of the optimization problem. The modified algorithm has been tested on various benchmark datasets ranged from easy to hard class dataset. As well as, we compare it against other popular algorithms such as Particle swarm optimization (PSO), African buffalo optimization (ABO) and Crow search algorithm (CSA) to demonstrate its efficiency and accuracy. According to the results of an easy class, the proposed algorithm has achieved an improvement ranged from 16% to 32% for N2C1W1_A instance when compared with other algorithms to reach an optimal result with the least number of iterations. In addition, the average execution time of it beats other compared algorithms with improvement reached up to 91% faster than PSO, 98% faster than ABO and 94% faster than CSA. For medium class, the bin efficiency of the proposed algorithm has been improved by 26% during iterations for N1W1B2R9 instance with minimum number of iterations and least execution time while for hard class, it has obtained an average efficiency 0.175 which is lower than other algorithms and average execution time 99.8 as a result of HARD3 instance. The experimental results validate the superiority of the proposed MSBPP in terms of not only the solution accuracy but also both the execution time and the number of iterations.