Research on dynamic real-time error correction method using Wiener-based neural network

摘要

A novel structure of Wiener-based neural network is proposed and applied to correct the dynamic real-time error for the improvement of the sensor's dynamic performance. First, the compensation filter was established based on the principle of inverse model and was described by a dynamic linear-static nonlinear cascade (Wiener model). Then, the neural network structure was devised and the network weights were accord with the parameters of the compensation filter. Followed that, some experimental devices were designed for dynamic calibration of the infrared temperature sensor. Finally, the identification of compensation filter was achieved by network iteration and the actual calibration data of the were made use of in the testing experiments. The results show that the stabilising time of the sensor is reduced to less than 7 ms from 27 ms and the dynamic performance is obviously improved after compensation.