摘要
提出了基于改进PSO算法的RBF神经网络混合优化(MPSO-RBF)方法,并将其应用到非线性系统的辨识中。该方法将改进PSO算法的全局搜索能力和RBF神经网络局部优化的高效性相融合,克服了普通PSO算法收敛的不稳定性和RBF网络易陷入局部极小值的缺点。经典型非线性系统仿真试验,并与GA-RBF和RBF辨识效果进行了对比,结果表明基于MPSO-RBF的混合优化方法较GA-RBF和RBF优化速度快、逼近性能好,可以达到更优的辨识精度。最后,通过对火电厂的过热汽温动态特性的辨识实例,同样证明了MPSO-RBF方法具有更好的性能指标。
A hybrid optimization algorithm (MPSO-RBF) for radial basis function (RBF) neural networks based on modified particle swarm optimization (MPSO) was proposed, and it was applied to nonlinear system identification. This method may take full advantage of the global searching performance of MPSO and the local optimized effectiveness of RBF neural networks, and it overcomes general PSO algorithm convergent instability and the disadvantage of RBF networks with falling into local minimum. At the same time, the typical nonlinear system simulation experiments were made, and also this method is a contrast to RBF neural networks based-on genetic algorithm (GA-RBF) and RBF identification effects. The simulation results show this MPSO-RBF method has quicker optimization speed and better approximation performance, and it obtains higher identification precision. Finally, an identification example for the dynamic characteristic of power plant superheated steam temperature was made, and the results proved that MPSO-RBF method has better performance index.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2007年第6期1382-1385,1389,共5页
Journal of System Simulation
基金
安徽省电力科学研究院的资助
关键词
改进PSO算法
RBF神经网络
非线性系统辨识
混合优化策略
过热汽温模型
Modified Particle Swarm Optimization (MPSO)
RBF Neural Networks
nonlinear system identification
hybrid optimized strategy
, superheated steam temperature model