Abstract：The modeling of distributed power sources in DC microgrids serves as the foundation for large-scale

power system simulations. However，the difficulty in obtaining electromagnetic parameters of diesel generators

hampers the accuracy of their electrical characteristic modeling. To address this issue，a data-driven equivalent

modeling method for diesel generators in DC microgrids was proposed. The method utilizes the voltage and current

outputs after rectifier supply conversion，along with the engine's output speed as inputs，and the root mean square

values of the phase voltage and phase current as outputs，based on the long short-term memory（LSTM）neural

network，voltage and current prediction models for the diesel generator were constructed. The temporal logic and

nonlinear mapping capabilities of the LSTM network were leveraged to describe the electrical characteristics of the

diesel generator. Furthermore，a reduced-order fuel consumption model was developed using polynomial fitting

based on the engine load characteristics of the diesel generator. Comparisons with actual operational data from DC

microgrid diesel generators validated the proposed modeling method's reasonableness and accuracy，demonstrating its practical application value.

      Key words：DC microgrids；equivalent modeling；long short-term memory（LSTM）neural network；diesel

generator