Neural Network Control of AC/DC Converters Robust to AC Grid Faults

This work presents the experimental application of artificial neural networks to improve the control of grid-connected AC/DC converters. The challenging context of AC grid faults is addressed and a general-purpose three-phase PWM rectifier equipped with IGBT technology, operating at 10 kHz switching frequency, is considered. A power supply facility with grid simulation capability emulates the most common failures. The estimation of the grid angle, critical for the efficient control in the synchronous d-q reference frame, is achieved by a set of neural networks depending on the type of fault, also identified by neural networks. The MATLAB/Simulink environment trains and simulates the neural networks and builds the control code for the target Texas Instruments Delfino F28379S microcontroller. The consolidated quadrature phase-locked-loop approach for grid synchronization is considered for comparison. The results demonstrate superior reliability when neural networks are used, especially when grid faults occur.