الفهرس | Only 14 pages are availabe for public view |
Abstract In this thesis we consider the problem of estimating the state noise covariance Q and the measurement noise covariance R. One of the approaches used in the litrature estimated one value of the Q and R matrices after a long interval, i.e. after the Kalman filter reached the steady state conditions. In this thesis we mOdified the above approach to obtain estimated values of the Q and R after a shorter interval, i.e. estimation occurs during the transient period of the Kalman filter; only a short non-estimation start-up period is essential. As an example; estimating R can start after 3 sampling periods. Also the first estimated values of Q matrix can be obtained beginning from sample number 5, and at any sample after that. Also the suggested approach can be modified to suit any system having time varying parameters \ r - 2 - After that we checked this approach by solving two examples, one for a first order system and the other is for a second order system, to obtain the estimated values of R and Q matrices. Different computational tests were performed to observe the effect of various parameters on the estimates of Rand Q. The effect of the following parameters were considered 1) The number of delayed samples The values of initial guesses for measurment noise cova- riance (RE) , and state disturbance covariance (QE). 3) The value of assumed actual noise. The thesis consists of six chapters. Chapter I is the introduction. Cba.pter II presents both tne Kalman filter and ru~ovatj.on approaches. Chapter III illustrates the causes of divergence of l;;bE:Kalman filt9r and the methods used to overcome this. Ohapter IV presents the approach used to estimate the value of the Rand Q matrices. Chapter V, indicates the results obtained from the different computational tests Which were done by applying the suggested a?proach to solve two examples. Chapter VI is the conclusion. |