Abstract:
This thesis presents two islanding detection techniques for distribution network hav-
ing high penetration of Distributed Generations (DG's) system. Various islanding and
non-islanding events for di erent values of active and reactive power mismatches are
simulated using Real time digital simulator (RTDS) software. One of the proposed tech-
nique is based on the machine learning technique i.e and other technique is based on
the passive islanding detection technique. The machine learning tool, Random forest
technique is used as a classi er to classify the islanding events and the non-islanding
events. The RF classi er utilizes sequence components of voltages which are obtained
by acquiring voltage samples of all the phases from point of common coupling (PCC) of
the targeted DG. The passive detection technque is based on the oscillation frequency
of the synchronous generator. The change in the magnitude of the oscillation frequency
is used to di erentiate the islanding events from the non-islanding ones. Validity of the
proposed schemes has been evaluated on large number of islanding/non-islanding cases
which are generated by modeling standard IEEE 34-bus system for the technique based
on Random forest classi er and by modelling standard IEEE 123 Bus System for the
technique based on oscillation frequency of the synchronous generator. The simulation
results indicate that the proposed schemes are capable to provide e ective discrimination
between islanding situation and non-islanding events. Moreover, they also gives satis-
factory results during perfect power balance situation. In addition, it provides better
stability in case of non-islanding events and hence, avoids nuisance tripping.
