在MEMS器件中,经常涉及到它的力学性能.对实验测得的数据,通过对非线性部分做基于Mtlab的最小二乘法的多项式拟合和BP神经网络逼近,得到如下结论:最小二乘法多项式拟合的误差较大,拟合精度低,而BP神经网络逼近误差小,精度高.同时最小二乘法拟合的多项式次数不同逼近的误差也有所不同. Of MEMS devices, it often involves its mechanical properties.For the experimentally measured data, the nonlinear part is subjected to polynomial fitting and BP neural network approximation based on Mtlab’s least square method, and the following conclusions are obtained: Least Squares Polynomial The fitting error is large and the fitting accuracy is low, while the BP neural network approximation error is small and the precision is high, while the error of the approximation by the least-squares fitting polynomial is also different.