類神經網路學習法與ARMAX模式整合技術應用於陀螺儀之識別

Abstract

本論文的目的是在於找出天美時積分陀螺儀的線性函數，其功能是在輸入 相同的情況下，函數之輸出值可以非常近似陀螺儀的輸出值，一般可以應 用在設計控制器的模擬上。本論文使用AR MAX-ANN方法來作系統識別，其 函數架構是AutoRegressive Movi ng Average with eXternal input模式 ，函數的參數是使用類神經網路中的誤差回饋演算法得到。另外我們加入 兩種傳統系統識別方法來與新方法做比較，其中之一是AutoRegressive with eX ternal input model estimator，在這種模式下我們使用最小平 方法得到其模式的參數。另一是Box-Jenkins model estimator，在此使 用來求得其模式參數值的方法是參數推算法。最後再由三種模式的輸出與 天美時積分陀螺儀的輸出做比較，判斷其五千筆資料(最大值：41.2003 deg/sec；最小值： -4 1.1372 deg/sec)之誤差狀況，得到ARMAX-ANN模 式之平均誤差為 3.3599 deg/sec， BJ模式為3.7644 deg/sec，ARX模式 為3.6993 deg/sec。 The objective of this paper is trying to search for the linear parametric model of TIMEX IG-10 integrating gyroscope, whose function is to approximate the measured output of gyroscope at the same input. In this work, an ARMAX-ANN method is presented which employ AutoRegressive Moving Average with eXternal input (ARMAX) model and error back-propagation learning rule. Two traditional methods of system identification, AutoRegressive with eXternal input (ARX) model estimator and Box-Jenkins (BJ) model estimator, has been used for comparing with the ARMAX-ANN estimator. Finally, we computed the average errors of ARMAX-ANN estimator and BJ estimator and ARX estimator at the same input, pseudo- random binary signal. We have the results that the average error of ARMAX-ANN estimator is 3.3599 deg/sec and that of BJ estimator is 3.7644 deg/sec and that of ARX estimator is 3.6993 deg/sec.