Neural Identification of Average Model of STATCOM using DNN and MLP

Abstract

Modeling of STATCOM is conventionally performed in the time-domain. Amongst them, dq-theory is well-known in which state-space equations are used for the analysis. Power systems, however, use the frequency-domain information in phasor-related studies such as load flow analysis. Because time-domain models of FACTS controllers cannot be directly applied to the power system analysis, an intelligent model can usefully bridge the time-domain information to the corresponding frequency-domain data. This paper proposes two neural network identifiers based on the existing time-domain average model of STATCOM. Extended resultant bridge presents an average-neural model of STATCOM, which can be analytically applied to power systems. To this extent, design and development of two neural network identifiers are performed using the dynamic neural network (DNN) and the multi-layer perceptron (MLP). To verify the developed models, the exact solutions obtained from the average model of STATCOM are compared with the outcomes of the DNN and the MLP identifiers. Moreover performance of the two identifiers is accordingly compared as well.