Swarm intelligence is an area of computational intelligence that has been growing and excelling as practical applications in complex problems. His inspiration comes from nature and the most common are based on: Birds (PSO), ant (ACO), bees (ABC) fireflies (GSO) and fish (FSS). They each have their peculiarities, such as how to represent success in a process of optimization and also the operators used to generate convergence or maintain diversity. Based on this, each technique also has weaknesses due to its unique way to represent success. The purpose of this research is to develop hybrid algorithms of swarm intelligence to decrease these weaknesses, without forgetting the need that you have to adapt the operators of convergence and diversity, maintaining the coherence of hybridization.

The algorithms based on swarm intelligence are inherently parallel, since the agents inside the swarm perform their operations individually. Hence, it is possible to take advantage of this behaviour and implement them in a parallel platform. This way, it is expected to reduce the execution time of their implementations. In the recent years, the use of Graphic Processing Units (GPUs) have been proposed for some general purpose computing applications. GPUs have shown great advantages on applications requiring intensive parallel computing. Therefore, we focus on increasing the perfomance of such algorithms by using the GPU capacities. Moreover, we study crucial aspects that may harm the perfomance of the GPU-based implementations. We perform simulations using a parallel platform developed by NVIDIA called CUDA.

A new approach based in Particle Swarm Optimization (PSO) for multiple objective optimization problems. PSO is modified by using the mechanisms crowding distance and roulette wheel, specifically on particles social leader selection and in the deletion method of solutions of the external archive. Another proposed change is the mechanism to update the particles cognitive leader. The performance of this approach is evaluated from test functions and metrics of the traditional literature. The results show that the proposed approach is highly competitive with traditional approaches.