Due to the influence of various noises and interferences on the fault current signal of distribution network, the characteristics of the fault signal are complex and varied, making fault location difficult. Therefore, a fault location algorithm for distribution network automation based on high-performance neural network technology is proposed. The feeder terminal units (FTUs) is installed in each section of the distribution network to collect fault current signals and implement filtering processing to remove noise. The marginal spectral features are extracted from the processed fault current signal, which can effectively reflect the frequency domain characteristics of the fault. A high-performance neural network model is constructed and trained. In this process, the marginal spectral features serve as the inputs for the trained high-performance neural network, with the aim of outputting the state types of distribution network sections. Subsequently, the sections with faults are identified as the localization results. The experimental results show that under the application of the studied method, the high-performance neural network localization results are consistent with the actual fault location, proving that the neural network model exhibits high accuracy and reliability in fault localization tasks.