具有灰层的非线性神经网络辨识

Identification of Nonlinear System with Gray-Layer Neural Network

提出了一种改进非线性神经网络辨识性能的新方法。通过神经网络输出灰层活动函数的选择，使辨识网络结构的选择更加合理和有效，收敛速度加快。

Feedforward artificial neural network (ANN ), commonly used for identifying nonlinear system, requires too much calculation when numbers of layers and neuaronns in network become large and high precision is required. We present now gray-layer neural networkmethod that can obtain high precisioh without requiring too much computation.Gray-layer neural network structure is show in Fig. 1. Activation fUnctions d, γ, sin (xu), … of output gray layer affect network convergence. Taylor and Fourier series are incorporated into the activation functions. Because the two series contain some information about unknown plant, gray-layer neural network can approach nonlinear mapping of unknown system faster.Finally, a certain static (Fig. 2) and a certain dynamic (Fig. 3) nonlinear system are taken as numerical examples. In Fig. 2, results obtained with ANN are show as full-line curve, three gray-layer networks (s1, s2 and s3) show higher precision than ANN, especially s3 network, which utilize both Taylor and Fourier series. In Fig. 3, results obtained with ANN are again shown as full-line curve. Two gray-layer networks (d1 and d2) both show much higher precision than ANN.