在有限元素法中，用理论分析建立的有限元模型很难完全与实际相符，这便带来了如何寻找有限元模型的误差源和怎样对它进行修正的问题。用静力试验修正刚度矩阵，再以修正后的刚度矩阵为基准修正其它矩阵是方法之一。目前存在的问题是该方法的修正结果对静位移中各元素相对误差的不一致具有高度敏感性。本文应用神经网络修正有限元模型中的刚度矩阵，既消除了上述敏感性，同时具有可修正性。 In the finite element method, the finite element model established by theoretical analysis is difficult to exactly match with the reality, which brings up the problem of how to find the error source of the finite element model and how to correct it. It is one of the methods to modify the stiffness matrix with the static test and then to correct the other matrices with the modified stiffness matrix as a benchmark. The existing problem is that the modified result of this method is highly sensitive to the inconsistency of the relative errors of the elements in the static displacement. In this paper, the neural network is used to correct the stiffness matrix in the finite element model, which not only eliminates the above sensitivity, but also has the correctability.