将多层前向神经网络技术和计算力学方法相结合应用于非均匀介质参数识别，提出了一种新的Ｓｉｇｍｏｉｄ函数，使前向网络能较好地实现全局最优逼近．采用有限元法获得神经网络训练所需的样本集，然后通过实测点的位移值用三层人工神经网络对非均匀介质进行参数识别．计算机数值实验表明，这种方法是有效的，所得结果具有相当的精度． A new Sigmoid function is proposed based on the combination of multi-layer forward neural network and computational mechanics for the identification of inhomogeneous medium, which makes the forward network can achieve global optimal approximation well. The finite element method is used to get the sample set needed by neural network training, and then the parameters of the inhomogeneous medium are identified by the three-layer artificial neural network by the displacement value of the measured point. Computer numerical experiments show that this method is effective, the results obtained with considerable accuracy.