Information Technology Reference
In-Depth Information
distinction can be made from the learning based on perfect domain theory when
consider the value and transmission of confidence.
9.9.3 Deep knowledge based approach
Domain theory of fault diagnosis is often imperfect. How to refine the knowledge
base of fault diagnosis is an important problem. A learning model that used for
malfunction diagnosis of distillation columns based on deep knowledge is
proposed (Cuiying Lv,1994). The construction of the model can be divided into
four steps: instance explanation, hypothesis generation, hypothesis confirmation
and extension. The model can discover new fault which the existing knowledge
can not find and pursue correct diagnosis. The process of learning based on deep
knowledge can be generalized as:
1. An instance is presented by the environment.
2. Explanation of instances: explain the instance based on EBL. The explanation
may succeed or not. Success means it is explainable according to domain
knowledge of the system; otherwise it can not be explainable by means of
domain knowledge. If successful, go to step 5 for extension, else go to step 3
to generate hypothesis. The explanation tree is created to explain. If
successful there will be a perfect explanation tree otherwise the explanation
tree is imperfect.
3. Hypothesis generationSystem attempts to confirm the absence of knowledge
based on current instance. When the knowledge absence is found, a
hypothesis that can remedy the knowledge absence is created, the goal-end of
the hypothesis is replaced with the absence-end and the data-end of the
hypothesis is substituted with data-end of the input instance. Then go to step 4
hypothesis confirmation.
4. Hypothesis confirmationDeep knowledge base is used. Search for the
relationship between the goal-end of hypothesis and data-end in the deep
knowledge base. That is attempting to confirm whether there are some
common properties between he goal-end of hypothesis and data-end in the
deep knowledge base. This could be successful or not. If it fails, return to step
3 hypothesis generation and get a new hypothesis, if it is successful, go to step
5 extension.
5. ExtensionExtend the confirmed hypothesis. The consequence of extension is
that some more generalized hypothesis can be acquired. The whole process of
extension is made up of two phases: the first one is to vary constants of
hypothesis maximumly while the second step is to extend the hypothesis to
get one or more generalized concepts.
Search WWH ::




Custom Search