Abstract: With the continuous expansion of grid-connected distributed photovoltaic capacity in urban and rural distribution networks, high-penetration new energy sources have completely changed the traditional single-source radial power supply mode of 10 kV distribution networks, transforming the grid structure into a complex multi-power hybrid supply system. After grid reconfiguration, the characteristics of short-circuit faults differ greatly from those in the past. Conventional sectional overcurrent protection frequently suffers from maloperation, failure to operate and tripping escalation, which seriously endangers the safe and stable operation of power grids. Aiming at the distortion of short-circuit fault characteristics and frequent failures of traditional protection caused by the integration of high-proportion distributed PV, this paper quantitatively reveals the influence laws of three typical effects: current shunting, current infeed and reverse current. An improved random forest fault identification model is established by combining current amplitude, phase difference between current and voltage, and transient energy entropy. An adaptive feature weight adjustment mechanism is introduced to address the low identification accuracy of single features. Furthermore, an intelligent protection architecture adopting sectional adaptive setting + local- regional distributed coordination is proposed. The proposed scheme has been applied and verified in the reconstruction project of 10 kV urban distribution feeders. The average isolation time of short-circuit faults is reduced from 38 min to 8 s. Maloperation and failure to operate of protection devices are completely eliminated, and the power supply reliability reaches 99.97%. The research results can provide technical references for the safe grid connection of high-proportion new energy in urban and rural distribution networks.