论文部分内容阅读
结合支持向量机和神经网络各自的优点,提出了一种新颖的自适应支持向量回归神经网络(SVR-NN).首先,利用支持向量回归方法确定SVR-NN的初始结构和初始化权值,基于支持向量自适应地构造SVR-NN神经网络的隐层节点;然后,使用退火过程的鲁棒学习算法更新网络节点参数和权值.为了验证所提出方法的有效性,给出了自适应SVR-NN应用于非线性动态系统辨识的实例.仿真结果表明,与以前的神经网络方法相比,基于SVR-NN网络的辨识方案能获得相当好的性能,它具有很快的收敛速度.因此,自适应的SVR-NN为非线性系统辨识提供了极有吸引力的新途径.
Based on the advantages of SVM and neural network, a novel SVR-NN is proposed.Firstly, the initial structure and initial weights of SVR-NN are determined by using support vector regression Support vector adaptively construct hidden nodes of SVR-NN neural network, and then use robust learning algorithm of annealing process to update network node parameters and weights.In order to verify the effectiveness of the proposed method, an adaptive SVR- NN is applied to the nonlinear dynamic system identification. The simulation results show that compared with the previous neural network method, the identification scheme based on SVR-NN network can get quite good performance, which has a fast convergence rate. Adaptive SVR-NN provides a very attractive new way for nonlinear system identification.