考虑误差影响的力学参数反分析神经网络方法研究

通过求偏导技术将反分析的参数误差分别表示为模型误差与测量误差的函数，利用这些函数研究了模型误差与测量误差在力学参数反分析神经网络方法中的传递过程以及影响反分析结果的影响因素．研究结果说明：输入变量的选择严重影响力学参数反分析的精度，选取力学参数灵敏度大的测点位移作为神经网络的输入变量，可以减小模型误差与测量误差对反分析结果的影响．这对提高力学参数反分析的实用性与可信性有重要意义．

Research of a neural network method for mechanical parameter back analysis considering influence of the errors

In this paper, functions of the mechanical parameter errors with model errors and measure errors were built with derivative. With these functions, the transferring process of model errors and measure errors in neural network methods and the factors to affect the outcome of back analysis are researched. The research shows that if nodal displacements of high mechanical parameter sensitivity are chosen as input variables of the neural network, the influence of model errors and measure errors on the results of mechanical parameters back analysis can be diminished. This result will advance mechanical parameter back analysis.