On sequential Monte Carlo sampling methods for Bayesian filtering
Statistics and Computing
SSP '07 Proceedings of the 2007 IEEE/SP 14th Workshop on Statistical Signal Processing
| Hi-index | 0.00 |
We propose a novel generalized algorithmic framework to utilize particle filter for optimization incorporated with the swarm move method in particle swarm optimization (PSO). In this way, the PSO update equation is treated as the system dynamic in the state space model, while the objective function in optimization problem is designed as the observation/measurement in the state space model. Particle filter method is then applied to track the dynamic movement of the particle swarm and therefore results in a novel stochastic optimization tool, where the ability of PSO in searching the optimal position can be embedded into the particle filter optimization method. Finally, simulation results show that the proposed novel approach has significant improvement in both convergence speed and final fitness in comparison with the PSO algorithm over a set of standard benchmark problems.