Abstract:

In order to make the precision of low-cost MEMS gyroscope data higher, this paper proposes a hybrid kernel function support vector regression (SVR) MEMS gyroscope random error prediction model, and through the particle swarm optimization (PSO) algorithm for model parameters and kernel. The function parameters are optimized. At the same time, the MEMS gyro random error data before and after prediction is analyzed by the Allan variance method. The experimental results show that the hybrid kernel function SVR can predict the random error of MEMS gyroscope up to 99.99%. When the MEMS gyroscope is in different state but the noise characteristics are the same, a unified SVR prediction model can be used to predict the random error. The results provide a basis for further real-time error compensation for MEMS gyroscopes.

Key words: support vector regression, PSO algorithm, Allan variance, random error