Intelligent Decision Technologies 10 (2016) 419–430 419 DOI 10.3233/IDT-160268 IOS Press Particle Swarm Optimisation algorithm for multi-agent system with dynamic ray (PSORM) Yasmina Hernane a,b,∗ , Soumeya Leila Hernane a,b and Mohamed Benyettou a,b a Department of Computer Science, University of Sciences and Technology of Oran USTO, Oran, Algeria b Laboratory of Modeling and Simulation of Industrial Systems, Algeria Abstract. This paper deals with the issue of reaching a target within an unknown search space. The target could be a human being, a bomb or any other object and the research space is composed of virtual robots, commonly known as agents that are continuously evolving together with the same purpose. To that aim, we use the bio-inspired Particle Swarm Optimization (PSO) method that we extend to the Particle Swarm Optimiza- tion with dynamic ray (PSORM) algorithm. The neighborhood is dynamically adjusted during the research process according to the value of fitness, which helps significantly to escape from local optimums. Despite the same variations we have made in some initial values such as population size and ray intensity, PSORM converges earlier. Our approach has the advantage to be completely distributed and of low-resources consumption. Keywords: Distributed systems, bioinspired, optimization, swarm intelligence, heuristic, parallel computing, virtual agents, dy- namic neighborhood, collective robotics, PSOM, PSORM 1. Introduction and overview The localization of targets in an unknown space is a well-suited task for mobile robotics [14,18]. Collective robotics are used particularly when the explored envi- ronment is dangerous or inaccessible to human beings; a target could become a victim in ravaged zones, or if there is the presence of bomb radiation. Mobile robots are usually equipped with sensors which have been installed with distributed programs, thus helping an exchange of data. Their resources are often limited because of their small size. The proces- sor on each robot has limited memory, and a very lim- ited power source (a battery). They are designed to fa- cilitate displacement and increase mobility. Therefore, ∗ Corresponding author: Yasmina Hernane, Department of Com- puter Science, University of Sciences and Technology of Oran USTO, Oran, Algeria. E-mail: yasmina.hernane@univ-usto.dz. the main challenge in this area of research is to de- velop a distributed program that does not require many resources, while at the same time it maintains a low cost. Computationally, it should be simple to set up, and scalable to any number of robots in a space. Before setting up such a model, it is necessary to simulate both the modelling and the dynamic of the space. Models can be considered differently according to the abstraction level that is used. This level should be defined according to the purpose of the simulation. In the following paper, we consider the robots as agents because we focus our research on the algorithmic as- pect of the system. The multi-agent systems area is closely linked to the one of collective robotics because the simulation is a preliminary step before the actual setting up of the environment [8]. Metaheuristics optimization algorithms are widely used in real world practical applications that include a wide range of areas, and aim to help communities of ISSN 1872-4981/16/$35.00 c  2016 – IOS Press and the authors. All rights reserved