Abstract: With the capacity change of distributed generation, the original power supply structure of microgrid has undergone modifications, altering the size, direction, and power flow configuration. This transformation poses a challenge in swiftly detecting and pinpointing the area of short-circuit faults within the microgrid.. In this paper, a low-voltage AC microgrid model is built in MATLAB/ Simulink. The current detected at the common connection point (PCC) between microgrid and large power grid is decomposed by high-scale wavelet energy spectrum algorithm, and the short-circuit fault eigenvalues suitable for different capacities are extracted, which realizes the early detection of short-circuit fault in microgrid under different capacities. By using wavelet energy spectrum characteristics and orthogonal least square (OLS)-radial basis function (RBF) neural network algorithm, a short-circuit fault location method suitable for microgrids with different capacities is proposed and verified by simulation. On this basis, the short-circuit fault protection hardware system of microgrid in grid connection mode is designed and verified by experiments. The results show that the designed protection system can quickly and accurately realize the early detection and regional location of short-circuit faults in grid-connected AC microgrid at the same time.