Citation
Sabri, Sabri Yasameen Hussein
(2018)
Custom Neural networks modelling for semitransparent thin film photovoltaic.
Masters thesis, Universiti Putra Malaysia.
Abstract
Thin-Film solar module of cadmium telluride (CdTe) is one of the Semi-transparent
PV (STPV) that can be employed in a wide application range as a means to sunlight
permeability while supplying solar electrical energy with some shading which also
preferable in hot areas. The power generated by solar photovoltaic (PV) is highly
affected by the weather environment. The prediction of a PV harvested energy and
the system performance requires an accurate and reliable modelling as a formula and
simulation design before installation. Silicon-based PV module with specifications
equivalent to that for the STPV for comparison purposes. The proposed approach
analyses the empirical data of a Thin-Film STPV module of CdTe type towards
modelling. A developed Custom Neural Network (CNN) has been functioning for
modelling the PV generated power based on laboratory and in-situ measurements.
Experiments for different PV panel installation topologies have been conducted for
performance analysis. Several standard single independent variable fitting modelling
equations have been addressed as a basic modelling for I-V and P-V characteristic
curves such as; Polynomial, Exponential, and Gaussian as parametric models. The
developed CNN modelling has been implemented on both; I-V, P-V characteristic
curves, and to simulate the power pattern of the PV module by adopting three factors;
a minimum number of the hidden neurons, the use of all measured data for training
the network weights, and linear output activation function, these factors were
examined to reduce the complexity of solving the network equations. Silicon-based
PV has been used in all modeling stages to validate the proposed methodology. The
simulation has been performed by the MATLAB-Simulink environment. The result
highlights the limit at which the STPV starts generating power via comparing with
its equivalent silicon-based PV module. The proposed CNN modelling has the best
goodness-of-fit than other relative models, and it is verified by the comparison
between the measured and modelled outcomes which shows reasonable R-square value. The experiments have been conducted on different Thin-Film STPV modules;
48W and 40% transparency, 62W and 20% transparency, and 72W and 10%
transparency. The results show that for a single module, the daily harvested energy
is =190.01Wh, while that double module is= 218.48Wh, which satisfies the analysis
of the single module measurements and that each individual thin film module can
only generate power at a high certain level of irradiance. The results of the proposed
CNN attain a correlation coefficient of 0.986 and show different fitting accuracy
depends on several factors for each individual method. The proposed approach can
facilitate the modelling strategy for other types of PV modules.
Download File
Additional Metadata
Actions (login required)
|
View Item |