Optimal deep learning control for modernized microgrids

In this study, a new control approach is introduced for active/reactive power control in modernized microgrids (MMGs). The dynamics of MMG are considered to be unknown and a fuzzy reference tracking linear quadratic regulator (FRT-LQR) is designed. To tackle the effect of uncertainties and faults such as short-Circuit, weak connection, unbalanced grids, an optimal H-infinity-based deep learned control (OHDLC) is presented. The main contributions are: (1) The dynamics are unknown, and are online identified by the restricted Boltzmann machines (RBMs). (2) The parameters in hidden layers are tuned by the unsupervised contrastive divergence (UCD) algorithm, and the parameters in the output layers are tuned by the designed Lyapunov based learning rules that ensure the stability. (3) The designed H-infinity-based supervisor compensates the perturbations. (4) Several simulations, comparisons, and real-time examination as Hardware-in-the Loop (HiL) setup verify the applicability of the suggested control method. A comparison between the suggested approach and related controllers shows that the designed controller is more robust and accurate. In the suggested method, besides the fact that the deep learning approach improves the accuracy, the designed H-infinity-based supervisor also enhances the robustness.

Automated summary

This study introduces a new control approach for active/reactive power control in modernized microgrids. The control method utilizes a fuzzy reference tracking linear quadratic regulator and an optimal H-infinity-based deep learned control to handle uncertainties and faults. The study presents several contributions and verifies the applicability of the suggested control method through simulations and real-time examination. A comparison with related controllers shows that the designed controller is more robust and accurate.