|
I have been reading Fuzzy Belief Revision and trying to determine how the methods could be used to combine readings from conflicting sensors. For example, if one sensor reads 70 degrees and another reads 75 degrees. This paper uses examples that combine subjectively defined properties. For example, a car may have the properties expensive, economical, and powerful with subjectively defined values 0.75, 0.50, and 0.75, respectively. I can't see a way to apply these techniques to quantified data but this paper, and many other papers on Fuzzy Systems, make reference to applying Fuzzy techniques to "imprecise sensors". Is there a paper that deals directly with combining sets of conflicting sensor readings using Fuzzy methods? Or otherwise combining quantified information rather than subjective properties? |
|
As far as I know, fuzzy techniques only combine conflicting data if you think that hardcoding decisions about how to deal with conflicting data is combining conflicting data sources. I think you would have far better luck using a probabilistic model where there are many (possibly biased) sources of data and you want to combine them, possibly considering reasons for their biases in the model. For example, you can easily build a model that says that there are many sensors around an object, the response of each sensor gets less accurate (maybe adding random noise, maybe being multiplied by a number between 0-1, etc) the further it is from the object. Then you can integrate everything approximately and get estimates for things like the actual temperature of the object, the bias of the sensors, if any sensor is defective, etc. |
