基于改进混合算法优化RBF网络的滤波器建模

An improved hybrid algorithm for optimizing RBF neural network filter modeling

为构建精确的微带线滤波器神经网络模型,提出一种结合自适应遗传算法和改进粒子群算法的混合算法。在自适应遗传算法中,构造二次型选择策略以提高优秀个体的复制概率,加快收敛到初始全局最优解;利用粒子群算法良好的局部搜索能力,在标准粒子群算法的位置迭代公式中引入高斯扰动项,以克服收敛速度慢和早熟收敛的缺点,提高搜索全局最优解的可能性。通过对测试函数仿真,验证改进算法的可行性。最后将混合算法用于优化神经网络参数,建立平行耦合微带线滤波器模型。结果表明,滤波器参数S21和S11的均方根误差至少减小18.22%与12.68%,微带滤波器建模精度得到提高,验证了该算法对滤波器建模的有效性和可靠性。

In order to construct an exact neural network model for the microstrip line filter,we propose a hybrid algorithm which combines the adaptive genetic algorithm and the improved particle swarm optimization（PSO）.In the adaptive genetic algorithm,aquadratic form selection strategy is constructed to improve the replication probability of excellent individuals,which can accelerate the process of convergence to the initial global optimal solution.Taking the advantage of the good local search ability of the PSO,a Gaussian perturbation term is introduced into the position iteration of the standard PSO,which can overcome the shortcomings of slow convergence and premature convergence,and improve the possibility of searching for the global optimal solution.Simulations on the testing functions verify the feasibility of the proposed hybrid algorithm.Finally,the hybrid algorithm is used to optimize the parameters of the neural network,and a parallel coupled microstrip line filter model is established.The results show that the root mean square error of the filter parameters S21 and S11 are reduced at least by 18.22%and 12.68%respectively,and the modeling accuracy of the microstrip filter is improved,which verifies the validity and reliability of the proposed algorithm.