基于函数连接神经网络的传感器Hammerstein模型辨识研究

Study on Identification Hammerstein Model of Transducer Based on Improved FLANN

针对非线性动态传感器模型辨识问题，提出利用函数连接神经网络算法对非线性系统的Hammerstein模型进行一步辨识的方法。以多项式逼近传感器中的静态非线性环节，同时结合动态线性环节的差分方程，建立关于直接输入输出的离散数据表达式，利用改进FLANN训练求解Hammerstein模型参数。采用变学习因子的方法对FLANN算法进行改进，提高了收敛速率和稳定性。实验结果表明，该辨识方法简单有效且具有更快的收敛速度。

For identification nonlinear dynamic model of transducer, a method for the nonlinear system one-stage identification by using functional link artificial neural network （FLANN） algorithm is proposed. The nonlinear system is described as a polynomial expression, combining the differential equation of dynamic system to build discrete data expression of input to output, solving the unknown parameters of the model by FLANN training. The convergence speed and the stability of convergence of FLANN algorithm is improved through variable learning factor. Experimental results show that the improved FLANN is simple and effective and has higher convergence rate.