Artificial neural network based auto-tuned PI compensator to enhance the dynamic response of the DC-link voltage in a grid-connected voltage source converter

This paper proposes an artificial neural network (ANN) based auto-tuned PI compensator to enhance the dynamic response of the DC-link voltage in a grid-connected voltage source converter (GVSC). The converter's model is linearized using the Jacobian method and the optimized parameters of the PI compensator for many local operation regions are determined. The collected data are thereafter used as features and target for the learning process of an ANN. Two ANN structures are designed; first only one input is used, that is, the reference of the DC-link voltage. In the second case, an additional input is added, that, is the active component of the grid current. The auto-tuned PI compensator with optimized parameters provided by the ANN is tested with numerical simulations conducted on a GVSC operating as an active rectifier. The results show that the ANN based auto-tuned PI compensator provides a better dynamic response of the DC-link voltage than the conventional PI controller, that is a lower settling time with practically no overshoot.