A Framework for Model-Based Adaptive Training

Scheduling System - Principled Knowledge

The individual elements that make up a chunk of knowledge represent the principles for this chunk of knowledge. In the MOBAT training framework, the principles represent the theory and fundamental laws underlying the operation of the device or physical system on which the training is based.

It is the most general form and hence underpins all other representations. Different mechanisms can be used to generate qualitative or quantitative solutions for the principled knowledge representation. For example, an explicit domain representation can be achieved with a Component Based Language [ITSIE 1992 D7].

Alternatively, the principled knowledge can be represented in a numerical simulation model as shown in the MOBIT boiler plant simulation model [MOBIT 1994]. In Proto-MOBAT, the principles are the fundamental elements that make up and control the function of the individual components in the boiler plant.

In Scheduling-MOBAT, the qualitative knowledge for principles is realised in a problem-space representation at a detailed level for the elements that make up the scheduling rules. A subset of the order scheduling rules (discussed earlier in this chapter) for the “validate-date” principles are represented in a sub-problem space as shown in the extended problem space map in Figure 5-8.

Figure 5-8 Extended Scheduling Problem Space Map

Figure 5-8 Extended Scheduling Problem Space Map

The scheduling problem space map looks like a typical task-subtask hierarchy. This map represents the logical structure, not the dynamic subgoal hierarchy. In operation, problem solving within a problem space involves all types of knowledge.

In response to a difficulty in problem solving, a sub-problem space to any other problem space can dynamically be created. In SOAR, when problem solving returns from a sub-problem space, a chunking mechanism is used to capture the elements (i.e., the principles) of a chunk of knowledge for future use.

In Scheduling-MOBAT, a sub-problem space is representing a set of scheduling principles for its parent problem space. The principles are the elements, variables and equations that govern the behaviour of a chunk of knowledge. The representation of principles enables an expert system to act as an intelligent assistant capable of explaining its reasoning at a fine level of detail.

The alternative model representations of procedural, associative and principled knowledge enable the trainer to select the most effective knowledge representation level for a given training objective.

Flower Show
© | | Sitemap