Neural Network Methods for Improving Linear Regression of Sensor Output Characteristic

BP neutral network structure and LM study rule of neutral network are applied to compensate and adjust sensor′s nonlinear errors. Training programs are done by MATLAB language. Two different initial rules and network architectures are compared in the aspect of network performance, especially of the converging speed. The results of computer simulation illustrate that the method which uses Nguyen Widrow initial rule and improved network architecture has a high converging speed and good precision.