In order to reduce the noise component in output signal of micro-electro-mechanical system（MEMS）gyroscope

a denoising method combines improved least mean square（LMS）algorithm with moving average filtering（MAF）is proposed.Firstly

preprocess the gyroscope data by improving the LMS algorithm

and then analyzes quantitatively the effect of different sliding window length of MAF and filter convergence factor on the noise suppression of gyroscope.Finally

a certain type of MEMS gyroscope inertial measurement unit is used to build the test environment.The proposed algorithm is applied to the actual noise reduction research

and the algorithm is compared with wavelet noise reduction and traditional LMS.Allan variance（AV）is used to identify the random error parameters of MEMS gyroscope.The results show that the error coefficients of gyroscope after filtering by LMS-MAF algorithm are significantly reduced

and the random walk of angle

quantization noise

zero deviation instability

speed slope and speed are all reduced by more than 90 %

which greatly reduces the noise of gyroscope and effectively improves the accuracy of gyroscope.