dc.description.abstract | This research attempted to control a Four-Leg Inverter (FLI) on microgrid rooftop solar (MGRS), which
connects to a distribution network (grid) via a distribution transformer. The connected load on an MGRS system
comprises two loads: nonlinear load and unbalanced linear load. Rooftop Solar (RS) injection current on each grid
phase fluctuated depending on irradiation value. Load and irradiation fluctuations and RS capacity differences on every
phase caused the transformer’s current unbalance and harmonic. Since the pulled current load varied between grid
phases, the current load’s instantaneous fundamental power demand (active and reactive) also differs for each phase.
Optimized Constructive Neural Network (OCNN) with single-phase PQ theory was utilized to independently control
FLI in every phase determined by fundamental power demand. Therefore, a transformer would perceive load and RS
injection as balanced despite varied and unbalanced conditions. OCNN builds networks by self-constructive methods.
Each training session enables the addition of new hidden layers and neurons inside each layer. The OCNN network
compares the error value associated with the training results to the error value associated with the temporary best
network (TBN). Throughout each training session, this comparison is made to determine the network with the lowest
error value or the global best network (GBN). The frequent irradiation fluctuation indicated that the system often
stayed in a transient rather than a steady-state. In high transient conditions, the performance of the proposed controlling
method had been tested in simulations. The result revealed that the OCNN controller obtains the lowest peak values
under high transient conditions, namely 2.62% for PCU and 6.73%, 7.33%, and 6.63% for THDi, respectively, at
phases A, B, and C. | en_US |