Enhanced Current Loop PI Controllers with Adaptive Feed-Forward Neural Network via Estimation of Grid Impedance: Application to Three-Phase Grid-Tied PV Inverters

Abstract

This paper describes a single-stage grid-connected three-phase photovoltaic inverter feeding power to the grid. Using the Recursive Least Squares (RLS) Estimator, an online grid impedance technique is proposed in the stationary reference frame. The method iteratively estimates the grid resistance and inductance values and is effective in detecting inverter islanding according to IEEE standard 929-2000. An Adaptive Feedforward Neural (AFN) Controller has also been developed using the inverse of the system to improve the performance of the inner-loop Proportional-Integral controllers under dynamical conditions and provide better DC link voltage stability. The neural network weights are computed in real-time using the controller sample time, making the system highly compliant to abrupt changes in grid conditions. Additionally, the proposed controller significantly improves reference tracking ability when subjected to variations in irradiation and temperature. The current control technique proposed results in better control approaches for microgrids, energy storage systems, variable frequency drives, and distributed generation.