Prediction of Sector and Region for Three-Level NPC Inverters Using Artificial Neural Networks

The performance of inverters, the most important hardware unit of renewable energy systems, depends on the sector and region values in which they operate in relation to the reference voltage vector. The accurate identification of sectors and regions is crucial. This study aims to overcome the shortcomings of sector and region identification based on classical mathematical models. To this end, the identification of sectors (6 classes) and regions (4 classes) is predicted with highly accuracy using artificial neural network (ANN) architectures. In this context, Narrow, Medium, Wide, Bilayered and Trilayered architectures were used and systematically compared. For sector detection, Narrow NN and Wide NN showed the highest performance with 99.97% accuracy. For region detection, Wide NN has the highest performance among the other architectures with 98.81% accuracy. The proposed architecture is modelled in a simulation environment and analyzed in terms of inverter sector and region, output current and voltage values. The simulation results show that the models based on classical and artificial neural networks are compatible and provide a solution to reduce the processing load.