# Startup Drift Compensation of MEMS INS Based on PSO–GRNN Network

## TL;DR

This paper proposes a new method using a PSO-GRNN model to reduce startup drift in MEMS INS, significantly improving navigation accuracy and reducing errors.

## Contribution

A novel PSO-GRNN model is introduced for startup drift compensation in MEMS INS, achieving better performance than traditional methods.

## Key findings

- The proposed method reduced standard deviation by over 80.6% and peak-to-peak value by over 72.7% compared to uncompensated MEMS INS data.
- Compared to traditional methods, the new approach reduced standard deviation by 54.5% and peak-to-peak value by 42.8% on average.
- Navigation experiments showed a 36.4% improvement in speed error and 41.1% improvement in position error.

## Abstract

The startup drift phenomenon that exists in MEMS INSs increases the navigation error, prolonging the start-up time. Aiming to resolve this problem, a startup drift compensation method based on a PSO-GRNN model is proposed in this paper. We adopted a correlation analysis to determine the input parameters of the PSO-GRNN model that mainly affect startup drift. In the process of training this model, we used the PSO algorithm to optimize the spread parameter of the PSO-GRNN model. The information transmission function between particle swarms was used to find the best spread parameter by iterative optimization, the particle’s position was mapped to the GRNN model, and the GRNN model was constructed with the optimal position of the swarm as the spread parameter. This method can effectively compensate for startup drift and improve navigation accuracy. Startup drift compensation experiments were carried out at different ambient temperatures. Compared with the MEMS INS data without compensation, the standard deviation of the MEMS INS data with the proposed method decreased by more than 80.6%, and the peak-to-peak value of the MEMS INS data decreased by over 72.7%. Compared with the traditional method, the standard deviation of the MEMS INS data compensated via this method decreased by 54.5% on average, and the peak-to-peak value decreased by 42.8% on average. Meanwhile, the performance of this method was verified by navigation experiments. With the proposed method, the speed error improved by over 36.4%, and the position error improved by over 41.1%. The above experiments verified that the method of this paper significantly improved navigation performance.