Improved Bat Algorithm for Multiple Knapsack Problems

Abstract

Bat algorithm has been paid more attention because of its excellent conversion ability between global search to local search and its high robustness. To solve the problem of 0–1 single knapsack problem, scholars introduced binary encoding on the basis of bat algorithm and put forward the binary bat algorithm. However, when solving the multiple knapsack problem (MKP), the binary encoding will lead to the emergence of illegal solutions, so it is necessary to use the multi-value encoding to re-model the MKP, thereby applying the bat algorithm to MKP. To improve the entire search ability of the algorithm, we optimized the effective solution in the algorithm using the greedy algorithm, and then proposed a greedy algorithm-based bat algorithm, namely MKBA-GA, for solving the MKP. To further improve the solution ability of the MKBA-GA algorithm, we used Single Running Technique (SRT) to optimize the effective solution, and then proposed an efficient SRT-based bat algorithm called MKBA-SRT. In order to verify the performance of the proposed MKBA-GA algorithm and MKBA-SRT algorithm, we compare them with BBA, IRT and SRT algorithms on twelve datasets, and the experimental results show that the solution ability of MKBA-GA algorithm is stronger than that of BBA algorithm, and the ability of MKBA-SRT algorithm is superior to that of other four compared algorithms on eleven datasets.