III. EXPERIMENTAL
In this work, one type of commercial grid connected
single phase inverters, rating at 5000 W is implemented and
its model is determined. The experimental system composes
of DC power supplies, a digital power meter, a digital
oscilloscope, resistive (R), inductance (L) and capacitive (C)
loads, a AC power system and a computer. Waveforms are
collected by an oscilloscope and transmitted to a computer to
calculate power waveforms in single input and single output
(SISO) type batch processing. In order to obtain the model
which can validate in the any situation, the Cross-validation
technique is applied. In the experimental, testing is done
when the inverter operates in six steady state conditions.
These conditions compose of three voltage change
conditions an three current change conditions. Then two
experiments are designed follow as system modeling using
cross validation and no cross validation technique. The
power waveform data are divided in two groups. One group
is used as data to estimate model or training model whereas
the other group is used to validate models. The waveform
from inverter which operates in normal condition is used to
be a training set and validation set. The six conditions data
are used to verify as independent data sets. The accuracy of
each case has been observed. In case of cross validation, the
training data is obtained from cascading six subsamples in
one group and validation data is constructed from expansion
of another group from training data. The six conditions are
verified as independent data sets again one by one. Then, the
accuracy of no cross validation and using cross validation
technique has been compared. The comparison of both case
are evaluated following to the system identification process
in Fig. 2. The implementation of the process has developed
by programming using the MA TLAB software in order to
check accuracy of waveforms and find waveforms of the
maximum accuracy compared with actual waveforms. This
is executed by selecting model structures and adjusting the
model order of the linear terms and nonlinear estimators of
nonlinear system identification. The next process, iterative
simulation waveform and experimented have been compared
until the best model performance is derived by results
include a quantitative measure of the model quality in terms
of goodness of fit to estimation data. The percentage of best
fit accuracy is obtained from comparison between
experimental waveform and simulation modeling waveform.