Abstract：The stability analysis of power system transient voltage，power angle，and frequency is often based

on short-time-scale electromechanical transient simulation，which requires a relatively accurate large-scale

simulation model，which not only has low simulation efficiency but also takes a long time，thus giving rise to the

demand for dynamic equivalence of large power grids. However，as the proportion of new energy increases，the

power system gradually presents some unique characteristics of new power systems，with more significant

differences in system inertia，frequency spatiotemporal distribution，and frequency modulation resource

distribution. Conventional dynamic equivalence methods based on slow coherence theory are gradually difficult to

ensure accuracy. To solve these problems，based on previous work，a large power grid dynamic equivalence method based on random forest algorithm was proposed. First，the large power system was divided into the study area and the external area. Then the external area on the opposite side of the study area tie line was equivalent to several nonlinear dynamic loads controlled by the random forest algorithm. The random forest algorithm was trained using the original model tie line's voltage/reactive power and frequency/active power relationships. A joint simulation model based on PSS/E and Matlab platform was established to achieve equivalent simplified electromechanical transient simulation，and Python environment was used to achieve information exchange and simulation control between them. Simplified results for the Northeastern United States power grid was presented and compared them with existing methods. The results show that the proposed method can effectively balance simulation accuracy and control stability.

      Key words：large-scale power system；dynamic equivalence；random forest algorithm；parallel simulation