Modelling and controlling of integrated photovoltaic-module and converter systems for partial shading operation using artificial intelligence

The thesis has three main themes: analysis and optimal design of Cuk DC-DC
converters; integration of Cuk DC-DC converters with photovoltaic (PV) modules to
improve operation during partial shading; and an artificial intelligence model for the
PV module, permitting an accurate maximum power point (MPP) tracking in the
integrated system.
The major contribution of the thesis is the control of an integrated photovoltaic
module and DC-DC converter configuration for obtaining maximum power
generation under non-uniform solar illumination. In place of bypass diodes, the
proposed scheme embeds bidirectional Cuk DC-DC converters within the serially
connected PV modules. A novel control scheme for the converters has been
developed to adjust their duty ratios, enabling all the PV modules to operate at the
MPPs corresponding to individual lighting conditions.
A detailed analysis of a step-down Cuk converter has been carried out leading
to four transfer functions of the converter in two modes, namely variable input -
constant output voltage, and variable output - constant input voltage. The response to
switch duty ratio variation is shown to exhibit a non-minimum phase feature. A
novel scheme for selecting the circuit components is developed using the criteria of
suppressing input current and output voltage ripple percentages at a steady state, and
minimising the time integral of squared transient response errors. The designed
converter has been tested in simulation and in practice, and has been shown to
exhibit improved responses in both operating modes.

A Neuro-Fuzzy network has been applied in modelling the characteristics of a
PV module. Particle-Swarm-Optimisation (PSO) has been employed for the first
time as the training algorithm, with which the tuning speed has been improved. The
resulting model has optimum compactness and interpretability and can predict the
MPPs of individual PV modules in real time. Experimental data have confirmed its
improved accuracy. The tuned Neuro-Fuzzy model has been applied to a practical
PV power generation system for MPP control. The results have shown an average
error of 1.35% compared with the maximum extractable power of the panel used.
The errors obtained, on average, are also about four times less than those using the
genetic-algorithm-based model proposed in a previous research.
All the techniques have been incorporated in a complete simulation system
consisting of three PV panels, one boost and two bidirectional Cuk DC-DC
converters. This has been compared under the same weather conditions as the
conventional approach using bypass diodes. The results have shown that the new
system can generate 32% more power.