基于改进QPSO算法的小波神经网络参数优化

暋针对传统的小波神经网络在参数优化过程中所采用的梯度下降法容易产生局部最优,提出了一种改 进的量子行为PSO 算法。新算法通过在最优平均值的全局搜索点中加入权重系数,用于改善粒子群的全局、局部搜 索能力和收敛速度,当粒子进化到后期,满足早熟条件时,粒子群在该维上发生变异,重新初始化后的位置均匀分布 在可行区域上,用于提高搜索精度。仿真实验结果表明,改进QPSO 算法比常规网络训练方法在寻优能力方面更加 有效。

Parameter Optimization of Wavelet Network based on the Improved QPSO Algorithm

In the process of traditional wavelet network for parameter optimization, the gradient descent method is easily to produce the local optimum. To solve this problem, an improved quantum behavior of QPSO algorithm was proposed. In the proposed method, a weighted coefficient was added to improve the global and local search and convergence speed of PSO. When the evolution became premature, particle swarm began to mutate in this dimension. The reinitialized position of the particles in the dimension re uniformly was distributed in the feasible region for improving search accuracy. The simulation results show that the improved QPSO algorithm outperformed in the searching ability than conventional network training method.