A Full-ANN Control Method of Three-Phase Grid-Connected Inverter

To break free from the confines of the d-q control framework and traditional control techniques, such as phase-locked loop, as well as proportional-integral/proportional-resonant controllers, this paper proposes a three-phase inverter grid-connected control method based on a full-artificial neural network (FANN-TPIGCC), which leverages the nonlinear approximation capabilities of artificial neural networks to directly obtain the modulation wave required for inverter control. First, the inverter circuit principle is analyzed and two sets of auxiliary variables are chosen. Subsequently, the first and second sets of auxiliary variables are separately input into two echo state networks based on the least square algorithm. These networks respectively output the reference modulation wave and the amplitude ratio along with the phase difference. Finally, utilizing the amplitude ratio and phase difference, a modulation transformation method is devised to fine-tune both amplitude and phase of the reference modulation wave, resulting in the ultimate modulation wave utilized for inverter grid-tied control. Simulink simulation results show that the FANN-TPIGCC method can achieve flexible switching between different reference current under both strong and weak grid conditions. It exhibits lower total harmonic distortion and ideal steady-state performance, meeting the control requirements of three-phase inverters.