Power quality enhancement in grid-tied photovoltaic systems through artificial neural network-based current control strategies

Abstract

The control of power electronics converters represents a critical issue for integrating renewable sources into electrical grids. Classical controllers face challenges in addressing system nonlinearities, often requiring several complex mathematical procedures, complex tuning, and limited robustness under varying operating conditions. Artificial neural networks (ANNs) have recently demonstrated an intelligent approach to replacing existing control methods. This paper aims to develop an ANN design method for the current control loop to ensure the efficient operation of grid-tied photovoltaic (PV) systems with high performance in terms of active/reactive power commands. The proposed ANN method employs the Levenberg–Marquardt (LM) backpropagation algorithm, renowned for its rapid convergence and robustness, along with a hyperbolic tangent sigmoid activation function, which enables it to model complex nonlinear behaviors effectively. The proposed neural network controller is trained to operate with constant switching frequency over a wide range of operating conditions. The experiment setup was developed using the dSPACE MicroLabBox platform. The experimental validation demonstrates that the designed controller significantly enhances the performance and ensures reliable operation across a broad range of real power outputs. Furthermore, the simplicity and low-cost effectiveness of the proposed inverter design make it particularly attractive for energy generation in grid-tied photovoltaic systems, offering a viable solution for sustainable energy production.