In order to address the issues of poor noise resistance and robustness resulting from a single source of fault diagnosis data, this paper proposes a method for fault location in micro-grid relay protection that integrates multiple sources of alarm information. The faults in the micro-grid are modeled using the symmetrical component method, and the characteristics of short-circuit faults are analyzed by solving the positive and negative sequence network differential equations. Similarity computation is used to process and visualize the data, and convolutional neural network (CNN) is employed to identify the fault information, thereby achieving intelligent generation of alarm information. The switch function method is utilized to weight and fuse the multiple sources of alarm information, and an improved quantum-behaved particle swarm with binary encoding (IBQPSO) is applied to solve the fault model. Finally, a case study is conducted in an improved IEEE 33 system. The results show that the proposed method can accurately generate micro-grid fault alarm information and quickly locate the faults, with high positioning accuracy even under conditions of multiple point information distortion.