基于改进粒子群算法的SVR参数优化选择

he study performance and generalization performance of support vector machine depend on a proper setting of its parameters. Parameters selection for support vector machine is very complex in nature and quite hard to solve by conventional optimization techniques. So, a new kind parameter optimization algorithm of SVR is proposed based on modified particle swarm optimization algorithm. Particle swarm optimization is an overall situation reconnaissance method, does not need to consider the model complexity and the variable dimension when select SVR parameters. This modified particle swarm optimization algorithm is modified by added certain new particles at each iterative to broaden search ability, which makes particles free of local optimization. Simulation results show that the modified particle swarm algorithm is valid method for parameters selection of SVR and the PSO-SVR model has superior learning accuracy and generalization performance.