In recent days, smart grids are enhanced globally, providing viable solution for efficient energy management in power systems and reliable energy supply. However, the microgrid integrated with multiple energy sources leads to multiple faults occurrence. The traditional methods for detection of faults in microgrid have faced significant challenges like inability to handle various fault scenarios. Therefore, this research proposes modified dragonfly algorithm with adaptive neuro-fuzzy inference system (MDA-ANFIS) for real-time fault detection in microgrid using power line communication (PLC). The proposed MDA-ANFIS is employed on PLC noise dataset that consists PLC noise signals acquired from two medium coupling outlet (MCO) ports. These gathered signals are preprocessed by using phasor measurement unit (PMU) to estimate the characteristics of signals. After that, discrete orthogonal stockwell transform (DOST) is employed to extract optimal features from these preprocessed signals. Finally, MDA-ANFIS is employed to effectively detect and diagnose the types of faults in smart grid. The proposed MDA-ANFIS achieved better results in terms of accuracy (99.93%), precision (99.12%), recall (99.92%), and F1-score (99.10%), when compared to existing convolutional neural network with gated recurrent unit (CNN-GRU) method.