碩士 / 中原大學 / 電機工程研究所 / 89 / In this thesis, we propose a Takagi-Sugeno fuzzy model-based parallel distributed Kalman filter for GPS/INS sensor fusion. First some Global Positioning System basic concepts are given. The advantages are all weather, global 24 hour positioning capabilities. The disadvantage is slow update of measurements and the possible loss of signal acquisition do to being obscured. Then important mathematical equations, for example pseusdorange, is given. For inertial navigation systems, the advantages are fast update of sensor measurement and independent positioning without use of external signals. The disadvantage are the integration process may cause divergence due to measurement error. Due to the above facts, we then give the motivations for GPS/INS sensor fusion and some typical fusion structures are shown. In the next part of this thesis, we give a thorough analysis on the Kalman filter estimation algorithm. Both discrete-time standard and extended Kalman filters are compared. Using a mobile robot model as an example, we illustrate the estimation of state for the EKF. Following, we present the main result of this thesis, ``T-S fuzzy parallel distributed Kalman filtering'. To carry out this approach, we first need to model a nonlinear system into T-S fuzzy model local linear representations. In the second step, we use a standard Kalman filter to estimate the state for each linear subsystem. Using the fuzzy inferred output, the overall estimate of state for the original nonlinear system is obtained. This methodology will greatly reduce the computational complexity compared to directly using the EKF due to the fact that standard KF only need to compute the state estimation on-line. In contrast, the EKF computes the state estimation, estimated error covariance, and filter gain online. This parallel distributed Kalman filter is then illustrated in numerical simulations on the mobile robot. A special tuning method for the EKF error covariance matrix is introduced where matrices chosen according to 1) Gaussian white noise; 2) hard bound or soft bound on propagating modeling uncertainty is given. Finally, an experiment of GPS and INS fusion using the parallel distributed Kalman filter with some concepts on positioning and navigation from a practical implementation point of view.
| Identifer | oai:union.ndltd.org:TW/089CYCU5442014 |
| Date | January 2001 |
| Creators | Yi-Wen Liao, 廖翊汶 |
| Contributors | Kuang-Yow Lian, 練光祐 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | en_US |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 138 |
Page generated in 0.0023 seconds