非平稳环境下提高神经网络辨识能力的新方法

New method to improve identification ability of neural networks in non-stationary environment

利用非平稳环境下系统时变参数变化规律的先验信息,构造参数转移矩阵来刻画系统的时变动态特征,并基于此推导了非平稳环境下神经网络训练的改进的卡尔曼滤波算法.仿真结果表明:该方法显著地提高了神经网络在非平稳环境下的辨识能力.

A prior information on time varying parameters in nonstationary environment is used to construct a parameter transfer function matrix (k) to deal with time variant dynamic system. A modified Kalman filter algorithm is derived to train neural networks.Simulation results show that the method presented in this paper remarkably improves identification ability of neural networks in nonstationary environment.