Comparative Assessment of Supervised Learning ANN Controllers for Grid-Connected VSC System

Abstract

The widely used PI-controller imposes certain performance limitations when used for the grid-connected VSC systems. Model predictive controllers (MPC) can eliminate these inherent issues of PI-controller but in trade-off with an increased computational burden. Research focuses extensively on finding an intermediate solution by employing artificial intelligence (AI) techniques. This paper discusses the implementation and performance of an artificial neural network (ANN) as an inner current controller. Firstly, the PI and MPC schemes are implemented as a supervisory controller for the grid-connected VSC system. Afterwards, a feed-forward ANN structure is trained with the data set of PI and MPC schemes. The resulted ANN structures are termed ANN-PI and ANN-MPC, respectively. For a fair comparison, Levenberg-Marquardt based backpropagation algorithm is used to train both the ANN controllers with the same number of neurons and hidden layers. All four controllers are implemented in the Matlab/Simulink environment to analyze the performance under different dynamic test scenarios. Further, the performance of all the controllers is evaluated using the OPAL-RT based real-time simulator.