Lowing ambiguity level in object state estimation in a case-based control system.
This work continues the publication series that is devoted to the investigation of methods of controlling complex objects basing on cases. In situation when it is hard or even impossible to use the exact mathematical model of object behavior the case-based reasoning control method becomes adequate. Instead of using the model we may use the accessible information about object under control state, controlling actions and their results. This means that we are using “cases”. In the theory of deduction this idea corresponds to three constituents of the case-based reasoning: description of a problem, solution or action applied, and outcome that is the result of applying the solution. This approach is based on separating of object states into classes, in each of which all the states are equivalent to each other. After that the object under control state is comparing with cases in the case base which was collected in advance. In order to choose the controlling action for the current case some metrics of similarity is used to find a case with similar initial and final states in the case base. The controlling action is extracted from this case then. The way of estimating of similarity of the current case and its precedent is critical for a system that uses the case-based reasoning technique. In previous publications the method was presented that used to separate cases from the case base into classes of equivalence. To estimate not fully described case the projection of classes onto the features space was used. Incompleteness in object description brings to the ambiguity in object estimation especially while one has the lack of time and resources. Classes in the intersection of which the object appears, are forming so called differential set of the object. If there is no feature that may separate the classes the choice of controlling action is hardly being made. The method offered by authors helps to low the ambiguity level on the basis of object prehistory investigation (previous states found and actions used). As a result of taking the prehistory into account the number of possible choices of actions is decreasing. The way of lowing the ambiguity level is described in the article. The notion of “prognostic” differential set of object states is presented. The final state estimation is produced on the basis of comparison of the current and prognostic sets.
Proceedings of the Institute for System Programming, vol. 26, issue 2, 2014, pp. 231-244.
ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).
DOI: 10.15514/ISPRAS-2014-26(2)-10Full text of the paper in pdf (in Russian) Back to the contents of the volume