Investigations into New Algorithms for Self-Organising Fuzzy Logic Control using Type-1 and Type-2 Fuzzy Sets

The number of applications of intelligent control systems has grown significantly over
the last few decades, and today they are used in various challenging industrial
application domains, where they provide particularly useful solutions. The term
„intelligent controllers‟ describes a field where control approaches are represented by
mechanisms similar to those used by the human brain. These characteristics include, for
example, learning, modification, adaption, effective working with high levels of
uncertainty and coping with large amounts of data.
Intelligent control systems are particularly useful for complex systems such as
biomedical and chemical plants, which are expected to work under optimal conditions.
A good example of an intelligent controller is the so called Self-Organising Fuzzy Logic
Control (SOFLC) proposed by Procyk and Mamdani in the late 1970s. The SOFLC
scheme involves a control policy that allows its structure to be adapted based on the
environment in which it operates.
The SOFLC combines a conventional fuzzy logic controller with a supervisory layer
which monitors and regulates the performance of the system. In this thesis, new
architectures are proposed for single input single output (SISO) and multi-input multi-
output (MIMO) structures to improve on the original SISO SOFLC design in terms of
performance and robustness, as well as extend the analysis and design issues relating to
such algorithms to the MIMO case using hybrid approaches. The work proposed in this
thesis includes: 1. A new development of type-1 and type-2 Self-Organising Fuzzy
Logic Control with a Dynamic Supervisory Layer (SOFLC-DSL) for the SISO case: In
this part of the thesis, the work is mainly focused on designing a sophisticated SOFLC
algorithm by combining a type-1 fuzzy system with a new Particle Swarm Optimisation
(PSO) algorithm, so as to make the SOFLC scheme more flexible and effective in terms
of responding to changes in the process to be controlled or the environment surrounding
it. A new on-line PSO algorithm is developed by using the idea of credit assignment and
fitness estimation to allow the optimisation of the consequent parts of the performance
index (PI) table on-line. The proposed scheme is tested on a non-linear and uncertain Muscle Relaxation Model. Computer results demonstrate that the proposed algorithm
achieve satisfactory performance, and is superior to the standard SOFLC scheme. In
order to enhance the capabilities of the controller to deal with environments where the
level of uncertainties and noise are high, both interval and zSlice type-2 fuzzy sets are
deployed. Simulation results show that the performance of the SOFLC-DSL algorithm
improves in terms of set-point tracking properties and the smoothness of the generated
control signals. 2. A new extension of the SOFLC-DSL to the multivariable case: The
proposed SOFLC-DSL algorithms are applied as the dominating controllers within
multivariable control architectures. In order to deal with the effects of interactions
between the input and output channels, both the relative array gain matrix as well as a
linguistic switching mode compensator are considered. The proposed algorithms are
tested on a drug dynamic process, and the results show they have good control
abilities in terms of maintaining the desired set-points with smooth control effort, as
well as in handling the interaction between different control channels.