Evaluation And Analysis Of Swarm Intelligence Algorithms

Swarm Intelligence (SI) has been widely used in multiple robots systems and wireless sensor networks (WSN) recently. It has become one of the hottest research areas in both home and abroad. Efficient evaluation and optimization methods could show the overall performance of an SI system and improve the algorithm, which has a great realistic significance to the analysis of SI system.SI algorithms focus on balancing two fundamental conflict performance indexes of WSN: sensing ability and energy cost. Each node in SI system is considered simple, asynchronous and homogeneous. Simple means that the ability of each robot is limited, asynchronous means the behavior of each robot are parallel. and homogeneous means the structure and function of each robot are almost the same.Although each object is simple, if work together, the whole swarm is able to perform complex coorperation task. Much work has been done in the research of SI algorithm, yet most of them focused on how to design a good SI algorithm, seldom tried to figure out efficient method to evaluate these algorithms or to improve the performance.The evaluation of SI algorithms is quite different from traditions evaluations. As a result, the method and focusing point are not in the same way. Compared with traditional ones, the difficulties of the SI algorithms lie in that: the data collection of multiple objects, the modeling of multiple objects, the various descriptions of all kinds of algorithms, and the balance between the energy cost and the sensing ability.In this paper, we use a modified multiple target tracking method to capture a swarm consisting of foraging robots, and then evaluate the SI algorithm that drives these robots. Finally, we use Markov chain and master equation to advise algorithm improvements.Experiment results indicate efficiency and correctness of our method. Our method is an effient, universal and automatical analysis system for multiple robot systems.