IMU data denoising method based on recurrent neural network

Abstract

The invention discloses an IMU data denoising method based on a recurrent neural network, and the method comprises the steps: building an IMU data time sequence sample data set, building the recurrentneural network, and carrying out the training optimization to obtain a weight file; and finally obtaining a deployment model. According to the method, a time sequence prediction method is introducedinto measurement data of a micro-electromechanical system inertial measurement unit (MEMS IMU), the measurement data is regarded as a time sequence, a long-short-term memory network (LSTM) is designedto extract a nonlinear relation of the time sequence, and the effect of removing noise is achieved through a sequence prediction method. Compared with a traditional denoising method based on a statistical model, the algorithm has better robustness and generalization ability, and the randomness of original data is not changed. The method has important application value in the aspects of inertial technology, integrated navigation and the like.

Description

technical field [0001] The invention belongs to the field of micro-inertial instrument detection and deep learning, and relates to an IMU data denoising method based on a cyclic neural network. Background technique [0002] The MEMS inertial measurement unit (IMU) is a device that measures the three-axis angular velocity and acceleration of an object. It can independently calculate the position, heading and velocity of the object through time integration without being affected by the surrounding environment. Due to its low price, size advantage, and independent measurement, it has been widely used in integration with the Global Navigation Satellite System (GNSS) to provide integrated navigation solutions, and it can independently output positioning information when the satellite signal is weak and blocked. However, since the original measurement of the inertial measurement unit contains various nonlinear and random errors, and the integral will become larger and larger over ...

AI Technical Summary

Problems solved by technology

Traditional methods mostly use signal analysis methods such as median filtering and wavelet transform for modeling, but from the perspective of data, the sensor data we obtain is essentially a set of data that changes with time, and the analysis of error data should also consider the correlation with time Sex, poor precision, poor effect of noise removal

Images