The mathematical model built on linear system theory cannot fully contain dynamic nonlinear characteristics of the transducer, which results in some modeling error and makes great restriction for some dynamic error correction methods. For the rapid developing neural network can approach functions nonlinearly and precisely, this paper studies the nonlinear modeling in the test system identification, where the transducer is the key factor, based on the neural network theory, and gets a quite fine transducer nonlinear model with the NARX(nonlinear autoregressive network with exogenous inputs)network, after comparing the advantages and shortages of many network architectures. Meanwhile, it also analyzes the key factors affecting the identification precision based on many simulation experiments, and summarizes proper network architecture & the corresponding training algorithms for the modeling of pressure transducer & accelerometer. Moreover, in applications with the modeling method, it analyzes the nonlinear compensation and correction method to get a mathematical model of the compensation and correction part for the test system to improve the data precision. |