Multi-Objective Self Regulating Particle Swarm Optimization Algorithm for BMOBench Platform M. R. Tanweer 1,2 , A. Al-Dujaili 2 , and S. Suresh 2 1 Hamdard Institute of Engineering and Technology, FEST, Hamdard University, Karachi, Pakistan. Email: rizwan.tanweer@hamdard.edu.pk, muhammad170@e.ntu.edu.sg 2 School of Computer Science and Engineering, Nanyang Technological University, Singapore, 639798. Email: aldujail001@e.ntu.edu.sg, ssundaram@ntu.edu.sg Abstract—These days, most of the real-world problems have become multi-criteria in nature and the demand for an effective multi-objective optimization algorithm has been significantly increased. This paper presents a new Multi-Objective Self- Regulating Particle Swarm Optimization (MOSRPSO) algorithm whereby the SRPSO algorithm originally developed for single objective problems has been modified to tackle with Multi- objective Optimization Problems (MOPs). The classical approach of Pareto dominance has been applied in the SRPSO frame- work together with the roulette wheel selection scheme for leader identification. The proposed MOSRPSO algorithm has been evaluated on all the hundred problems from Black-Box Multi- Objective Optimization Benchmarking (BMOBench) platform and the results are presented. The performance clearly indicate that MOSRPSO is a potential candidate for solving MOPs. Index Terms—MOO: Multi-Objective Optimization, PSO: Par- ticle Swarm Optimization, MOSRPSO: Multi-Objective Self Reg- ulating Particle Swarm Optimization, BMOBench: Black-Box Multi-Objective Optimization Benchmarking I. INTRODUCTION Over the past decades, the complexity of the global opti- mization problems have significantly increased and has been extended to the multi-criteria optimization problems. In multi- criteria, the nature of optimization problems are extremely complex whereby several conflicting objectives are required to be handled simultaneously. Further, there is also a need of such an algorithm that can effectively solve the (Multi-Objective Optimization Problems (MOPs) with minimum computational budget. This has outstretched the demand of real parameter optimization schemes for solving real problems. Among the numerous algorithms introduced, the population-based opti- mization algorithms are effectively providing promising solu- tions with lesser computational requirements. There are several MOO algorithms developed in the literature starting from initially developed Pareto Archived Evolutionary Strategy (PAES) [1], Strength Pareto Evolutionary Algorithm (SPEA) [2], Non-dominated Sorting Genetic Algorithm (NSGA-II) [3], recently developed hybrid framework for MOPs [4], dividing rectangles [5] and Gap Optimized Multi-objective Optimization (GOMORS) [6]. One of the effective population based nature inspired optimization algorithm is the Particle Swarm Optimization (PSO) algorithm initially introduced in 1995 by Eberhart and Kennedy [7]. PSO is derived from the collective behavior of swarms in search of food. Members of the swarm are represented as particles in the algorithm and the food searched by birds is presented as potential solution. The particles fly in the search space updating the search patterns using their self and social experiences. Information sharing mechanism has been implemented in the algorithm to ensure proper movement towards the global optimum solution. Lesser computational efforts of PSO attracted researchers towards development of efficient MOO algorithm. Multi-Objective PSO (MOPSO) was introduced in [8], [9] whereby PSO was modified for solving MOPs. A recent area in the development of PSO algorithms is derived from human thought processes and learning principles. Human beings are known to be intelligent and have good social cognizance [10]–[12]. Inspired from this, researchers developed several variants of PSO by incorporating different human learning principles. A Human cognition inspired PSO was introduced in [13]. Further different areas from human learning principles have been explored and incorporated in PSO framework for better convergence [14]–[19]. All the human learning principles inspired PSO variants have exhib- ited significant performance improvement in the convergence characteristics of the PSO algorithm. For a comprehensive empirical analysis on the performance of PSO variants, one can refer to [20]–[22]. In this paper, recently proposed variant of PSO, the Self Regulating Particle Swarm Optimization (SRPSO) algorithm has been modified to handle MOPs. The SRPSO algorithm is inspired from human self-learning principles where self- regulated inertia weight and self-perception based selection of direction from global best position strategies were incorpo- rated. Utilizing these strategies, SRPSO has exhibited much better results and has provided faster convergence closer to the optimum solution. It has been stated in [16], [17] that SRPSO has exhibited competitive performances as compared to other PSO variants and other meta-heuristics. Therefore, the same algorithm has been modified for MOPs by incorporating the classical Pareto dominance scheme in the algorithms’ framework. Further, a roulette wheel selection scheme has been incorporated for selection of leader particle. With the help of these schemes, the algorithm has become capable of solving multi-objective problems. The algorithm is referred to as 978-1-4799-7492-4/15/$31.00 c 2016 IEEE