阐述了岩石力学问题的几种神经网络建模方法 :非线性力学参数关系、利用历史信息预测未来信息、利用局部信息预测整体信息、利用小尺度下的信息预测大尺度下的信息等建模方法。并指出了合理的学习样本、学习参数、学习算法、网络结构等是提高神经网络模型的外推预测能力与非线性表达能力的关键。 Several modeling methods of neural network for rock mechanics are described: the relationship between nonlinear mechanical parameters, the prediction of future information by historical information, the prediction of global information by using local information, the prediction of large-scale information by using information under small scales . It is pointed out that reasonable learning samples, learning parameters, learning algorithms and network structure are the keys to improve the ability of extrapolation and nonlinear expression of neural network model.