To improve the low-voltage ride-through capability of wind farms, this paper takes grid-connected medium and large-scale wind farms as the research object. Starting from the entire wind farm level, it analyzes the transient characteristics and LVRT capabilities of doubly-fed wind turbines under fault disturbance scenarios. The RBF neural network is used to fit the transient process of the DFIG rotor current after removal, and a crowbar protection adaptive switching strategy is proposed. The simulation results of the calculation example show that the RBF neural network is more accurate in predicting the maximum current on the rotor side after a fault, and can effectively avoid repeated switching of the crowbar. This strategy further improves the reactive power support and LVRT capabilities of wind farms under fault scenarios, and significantly improves the safety and stability of wind farm operation. This work is supported by the National Natural Science Foundation of China (No. 51877088) and Enshi Science and Technology Bureau (No. RZ1900001902).