Generalized distributed learning-based control for microgrid clusters with spatiotemporal uncertainty
Fabin Cheng, Jingang Lai · 2025
This article studies the distributed voltage restoration problem through small sample data migration for DC microgrid (MG) clusters with spatiotemporal uncertainty. In the entire design process of distributed voltage restoration control, an adaptive fuzzy-neuro learning network framework is proposed. Firstly, a general spatiotemporal uncertainty model is established to describe the impact of voltage and time variables on the security control of MG clusters. Given that the current single network structure does not possess spatiotemporal recognition properties and the complexity of building stacked networks for spatiotemporal feature learning, we adopted a Euler norm-based time feature supremum extraction method and then used boundary mapping to enable the establishment of the general Takagi-Sugeno (T-S) fuzzy inference network to fit any spatiotemporal uncertainty. Secondly, considering the limited data acquisition and difficulty in fully modeling events in some extreme situations in practice, an adaptive critic network learning strategy based on small sample data is introduced to achieve data transfer and improve the generalization of control schemes. Finally, in order to verify the effectiveness of the proposed design scheme, the IEEE 33-node distribution system is tested via real-time testing equipment built on the OPAL-RT platform.