Study on Four Parameters Estimation’s Methods of Time-Series Model in Modeling the Signal of a Quartz Flex Accelerometer

Abstract

Box-Jenkins method of time series in modeling the signal of quartz flex accelerometer is studied in the paper. Firstly, a JSD-I/A quartz flex accelerometer is placed on a level test bench, and the output signal of the JSD-I/A quartz flex accelerometer is acquired. Secondly, the acquired signal of the JSD-I/A quartz flex accelerometer is preprocessed by Db3 wavelet denoising, trend items exacting, and standardized processing. Thirdly, The statistical characteristics of autocorrelation function (ACF) and partial autocorrelation function (PACF) of the processed time series data are analyzed, and the results show that ACF presents tailing characteristic and PACF presents censoring characteristic after 12th order. So AR(12) model is suitable for modeling the processed data of the JSD-I/A quartz flex accelerometer. Fourthly, the AR(12) model’s parameters are estimated by four methods, named least square method (LSM), Yule-Walker method, LUD method and Burg method, respectively. The fitting effects by residuals sum of squares (RSS) of the above estimation methods are compared and the results show that LSM outperforms the other three estimation methods.