AbstractThe structure complexity and the extended geographical area are among the factors that create challenges in accurate fault location and detection in distribution networks. The single‐phase to ground (SPG) faults are considered as most frequent reasons for distribution network interruption, which can threaten the network's reliability. Signal processing methods are usually used as common approaches to distinguish and locate faults in distribution networks. This paper proposes a novel scenario to select optimal wavelet packet transform (WPT) coefficients for SPG fault location and a method based on superimposed voltage energy to distinguish the faulty phase according to these coefficients. Furthermore, employing the energies of the superimposed voltages helps to minimize the effects of high resistance faults (HRFs) and load encroachment on the efficiency of the faulty phase detection part. Comparing the obtained results with other scenarios demonstrates the considerable efficiency of the proposed scenario. Finally, with the help of general regression neural networks (GRNN) as machine‐learning tools, a new algorithm is derived for detecting and locating the SPG fault and capacitor bank switching overvoltage (COV). Simulation results are implemented on the IEEE 34‐bus standard distribution network, demonstrating the efficiency and superiority of the suggested method.