Model-Free Deadbeat Predictive Current Control for Grid-connected Inverters using Autoregressive Model and Recursive Least Squares

This paper presents a novel Model-Free Deadbeat Predictive Current Controller (MF-DBPC) tailored for grid connected two-level inverters incorporating resistance-inductance (R-L) filters. Unlike traditional approaches, the MF-DBPC leverages a data-driven model derived solely from current and voltage measurements, eliminating the need for explicit systems parameter inputs. Central to the MF-DBPC’s functionality is an Auto-Regressive with Exogenous Input (ARX) model, complemented by a Recursive Least Squares (RLS) estimator for real-time parameter identification. This strategy offers enhanced adaptability to dynamic system conditions and achieve robustness against parameter mismatches inherent in grid-connected inverter systems. To demonstrate this, two distinct model-free predictive control strategies have been benchmarked: one grounded in the deadbeat control principle and the other utilizing a rolling optimization technique. Simulation analyses demonstrate that the MF-DBPC, driven by the deadbeat principle, yields superior current waveform quality while requiring a sampling frequency five times lower than its rolling optimization technique. Experimental validation further confirms the efficacy of the MF-DBPC across steady-state and dynamic performance metrics. Notably, its robustness against filter inductance mismatches is highlighted, showcasing resilience under challenging real-world conditions.