A model-based architecture for explaining uncertain data /
Marine oil spills, such as the one resulting from the
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1995.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=742744991&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | Marine oil spills, such as the one resulting from the wreckage of the Exxon Valdez in 1989, are a source of major ecological and economical disruption. When such an incident occurs, rapid response is of critical importance to contain the spilled oil, since the actions taken in the first eight hours of the incident often determine the success or failure of the response. Thus, decision makers are often forced to respond to the oil spills before a clear understanding of the situation can be obtained. This dissertation contributes to the development of computational tools to support such decision-making problems. An oil spill model is one such tool. The input of the model consists of the location of oil spill source, the type and original volume of oil spilled, the temperature, and the wind and current vectors at each location for each time step. The model generates the state of the oil, which is the oil distribution, for each next time step. Given an initial state of the oil, we can apply to the model the inputs at different times iteratively to predict the oil's future state. Data on the state of the oil and the inputs at different times is provided by observers, and may have large uncertainties associated with them. Before decision makers can use the model to predict the future state of the oil, they must find inputs that agree with the observations and cause the model to match the past behavior of the spill. We wish to automate this problem solving process. In this dissertation, a model-based architecture is proposed to explain uncertain observations in oil spill tracking. Explanation of uncertain observations for each time step goes through an iterative process of two phases. The two phases are implemented by two modules: error detection and fix generation. The error detection module looks for inconsistencies between the modeled and observed oil distributions. The fix generation module removes the inconsistencies by adjusting the input of the model in order to match the modeled oil distribution with the observed oil distribution. When an inconsistency can not be removed for some time step alone, the fix generation module backtracks in time such that the inputs in the previous time steps are readjusted in order for the inconsistency for that time step to be removed. Fuzzy logic techniques are employed to deal with the uncertainty in the observations. The validity of the architecture is successfully demonstrated through implementation. |
|---|---|
| Item Description: | Vita. "Major Subject: Computer Science". |
| Physical Description: | xvi, 109 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilms Inc. |
| Bibliography: | Includes bibliographical references. |