基于粒子群优化与SVR-ADLA算法的MIMO系统识别研究

Identification of MIMO Systems based on particle swarm optimization,support vector regression and annealing dynamic learning

针对现有基于径向基函数(RBF)网络对多输入多输出(MIMO)系统识别中存在收敛速度慢、系统识别稳定性不高的问题,提出了一种新的MIMO系统识别方法:采用支持向量回归(SVR)算法建立RBF网络初始化结构,确定初始化网络参数;采用退火动力学习(ADL)算法对系统识别网络进行训练,在训练过程中采用粒子群优化(PSO)迭代算法选出最佳学习率组合,使识别网络实现对MIMO系统的识别。对一个两输入输出系统进行了识别仿真,仿真结果表明,用该识别方法重建的识别系统性能优于目前RBF网络参数优化过程中常用的最小平方算法或梯度下降法算法。RBF网络识别系统易于实现,在MIMO系统识别中具有广泛的应用前景。

The identification of a multiple-input multiple-out-put （MIMO） system based on a radial basis function （RBF） network was studied, and a new identification method was put forward to solve the problems of slow conver- gence rate and low identification stability in current identification. The new method uses the support vector regres- sion （SVR） algorithm to establish the initialized structure of a RBF network and set the parameters of the initialized network, and then adopts the annealing dynamic learning （ADL） algorithm to train the system＇ s identification net- work, and in the training, uses the iteration of particle swarm optimization （PSO） to select the best learning rate combination to make the MIMO system recognized by the identification network. The identification simulation for a two-input and two-output, system was conducted, and the simulation results showed that the system identification performance of the proposed method was better than the least square algorithm and the gradient descent algorithm frequently-used in the current process of optimizing RBF network parameters. The identification system based on a RBF network is easy to implement and it has a wide application prospect in MIMO system identification.