Abstract: With the continuous improvement of the automation level of distribution network, the sensing terminals and access data have increased sharply. Problems such as slow data transmission and information missing and false alarms have seriously affected the speed and accuracy of fault handling. Edge computing can quickly and efficiently process data and make decisions on the spot. To this end, an optimized configuration method of edge computing units for rapid processing of distribution network faults is proposed, including system architecture, edge generation, and division methods. Firstly, the objective function is constructed by economic and communication real-time indicators, and reliability and system requirements are constraints to solve the optimal number of edge. Then, spectral clustering combined with k-means algorithm is used to divide the jurisdiction and location of edge, and partitions are modified according to balance and actual requirements. Finally, the IEEE 33-node and IEEE 69-node systems are verified as examples. The fault simulation model based on Matlab/Simulink is established. Compared with traditional methods, this method can deal with distribution network faults correctly and quickly, and effectively guarantee the safe and stable operation of distribution network.