Information Technology Reference
In-Depth Information
7
Transparent Fuzzy/Neuro-fuzzy Modelling
7.1 Introduction
Fuzzy logic is a methodology widely applied in model building of dynamic
systems for implementation of advanced control systems. Fuzzy models are
developed using the universal approximation capability of fuzzy logic systems.
Such models differ from other types of model built using
non-symbolic
methodology
, mainly because they can represent knowledge in a transparent
manner using fuzzy IF-THEN rules which are understandable to the human expert
who can directly operate on them. This provides the direct man-machine
communication.
Fuzzy models are generally built by extracting and encoding expert knowledge
into the IF-THEN rules with the linguistic arguments, in this way generating a
transparent knowledge appropriate for its easy inspection, modification, and
maintenance by human experts. However, the process of knowledge acquisition
and building of adequate IF-THEN rules are not trivial tasks, because the experts
are not always available and their knowledge is often incomplete, episodic and
time varying. This was the motivation for switching model building approach from
the seminal ideas of knowledge acquisition described above to a data-driven
approach. Unfortunately, many of newly developed algorithms for data-driven
fuzzy modelling aim at good numerical approximation and pay little attention to
the transparency and computational load of the resulting rule base. In this chapter
we will therefore present a rule base simplification method that can be used - along
with arbitrary fuzzy modelling methods - for obtaining transparent and compact
fuzzy models from data. The efficiency of the approach will be demonstrated on
the example of nonlinear plant modelling and prediction of its future output value.
Search WWH ::
Custom Search