Abstract: To solve large voltage dynamic deviation in the transition process caused by power disturbances of the islanding DC microgrid, this paper proposes a neural network control strategy to optimize the voltage dynamic track. The characteristics of droop control and its improvement methods are analyzed, and the optimal voltage dynamic track of the islanding DC microgrid is proposed. Based on this, a neural network is used to generate the reference voltage and minimize the dynamic voltage deviation of the common DC bus through nonlinear control. Based on the vector space isomorphism principle, a neural network reduced-dimensional state feedback construction method is proposed, and the neural network control boundary is determined according to the maximum source-load imbalance condition. Considering the dynamic voltage deviation and its motion trend, a voltage perturbation hierarchical reward function is constructed. The static reward guides the neural network to reduce the voltage deviation, and the dynamic reward improves the neural network to correct the voltage motion trend, so as to realize the optimal control of the whole process of voltage dynamic track. Finally, the results show that the proposed strategy can effectively suppress the dynamic voltage deviation caused by the power imbalance and ensure the fast stabilization of the common DC bus voltage by using MATLAB/Simulink for case verification.