Study on gray neural network drift modeling for piezoelectric gyro

The piezoelectric gyro's drift has a multi-valued nonlinear behavior in different temperature and operation time. It can not be described by using temperature input neural network model and time sequence model (ARMA). A single-mapping based on the three dimension coordinates was presented. Temperature and run time were designed as input, gyro's stationary null voltage and scale factor were designed as output in the tree dimension coordinates. Grey accumulate operation (AGO) was used in the processing of acquired data. Then, the RBF neural network model was presented to approximate the gyro's drift. The simulation results show that the new approach for modeling is effective and of high precision.