Data-driven Modeling of Distribution Network using Experimental Data

For over three decades the deep learning (DL) has found its applications in modern power systems with introduction and implementation of renewable energy sources (RES) in distributed generation. As RES are integrated in/out of the distribution grids, swift changes in electrical parameters pose a big challenge on power system quality and reliability. To employ an effective monitoring and control, it is needed to represent a system with an equivalent dynamic model, which may be done through DL. DL indicates a multi-layer neural network, which are extremely effective for modeling of non-linear relationships. In this paper, shallow neural networks are realized for a) three phase, and b) dq -frame experimental data, using two different optimization algorithms. The inferences will throw light on effectiveness of difference of log-sum-exp algorithm (DLSEA) as compared to backpropogation algorithm used by neural network toolbox of MATLAB.