To improve the efficiency of transient simulation and analysis of wind power grid-connected systems

a wind farm equivalence model is needed. The existing direct-driven wind farm equivalence model has problems such as it is difficult to be applied to post-fault analysis and low computational efficiency. To address this problem

this paper proposes a novel direct-driven wind farm clustering and equivalenting method driven by a mechanism-data hybrid approach. Firstly

the post-fault transient response characteristics of direct-driven permanent magnet synchronous generators (PMSG) were analyzed

and a method of wind turbine group clustering based on the initial wind speed and the post-fault steady-state voltage at the terminal of each PMSG was derived. Secondly

an efficient prediction method of the post-fault steady-state voltage at the terminal of each PMSG based on multi-layer graph convolutional networks was proposed

and the prediction model can be adapted to different wind farms and power grid topologies. Further

a multi-machine equivalent model was constructed for direct-driven wind farms. Finally

the correctness of the proposed subgroup clustering and equivalenting method were verified by simulation.