The load-side interface converter plays an important role in suppressing strong uncertainty in DC microgrids

but its input voltage fluctuations

modeling errors

and load changes pose severe challenges to system voltage regulation control. To suppress voltage fluctuations in DC microgrids

the advantages of independent model information and strong anti-interference ability are taken into account

and the self-learning features of actor-critic (AC) algorithm are integrated; thus a DC-DC interface converter reinforcement learning self disturbance control strategy based on AC algorithm is proposed. The parameter tuning problem of active disturbance rejection control (ADRC) is characterized as a Markov decision process

and the complex rules of the microgrid model are continuously learned based on the interaction data between the intelligent agent and the system. A robust optimal parameter dynamic adjustment curve driven by errors is trained and generated. Numerical simulation and physical experiment results show that

compared with traditional linear active disturbance rejection control (LADRC)

the proposed strategy reduces the maximum output voltage offset by 60.8% and 58.6%

respectively

during bus voltage fluctuations and load changes

and shortens the corresponding recovery time by 22 ms and 23 ms

demonstrating superior disturbance rejection performance and strong robustness to changes in capacitance and inductance parameters.