基于双层网络的混合PSO算法的RBF建模

Double-layer Network Modeling Based on Mixed Particle Swarm Optimization

针对RBF网络的建模问题,设计了基于双层网络的建模方法。第一层网络采用随机方法确定了隐层单元数,并利用并行PSO算法对网络进行初步训练,第二层网络采用了主从粒子群的方式,借鉴了遗传交叉的思想,对第一个网络的最优解进行了再训练以提高网络的训练精度。从对非线性系统的仿真结果看,该方法最终确定的隐层单元数比较少,与RBF网络相比有着一定的优越性,而且优于单层并行PSO算法的RBF网络。

To sovle the modeling problem of RBF network, double-layer network is designed. First of all, the first layer of the network derermine the number of hidden units by using a random method, and the network is trained initially by using parallel PSO algorithm; and then to improve the accuracy of network training, the second layer of the network using master and secondary particle swarm algorithm and the method of genetic crossover on the optimal solution of the first network is re-trained. From the simulation results of the nonlinear system to see that the method ultimately determine the number of hidden units is relatively small, compared with the RBF it has some advantages, and moreover is better than the parallel PSO based on single RBF network.