In a smart substation and due to the low sampling rate, travelling wave fault location cannot be applied. Therefore, a novel algorithm for transmission line fault location in smart substations is proposed. The algorithm can select the best fault location method among the conventional single-ended frequency fault location methods, and then get a value that is the closest to the fault. The conventional frequency fault location methods have various source of error, and fault location result is affected by power source, the opposite system impedance, transition resistance and other parameters. In different fault cases, every method has different location accuracy, namely the most accurate location method always varies with varying parameters. In this paper, firstly we construct a large number of training samples of transmission line fault, secondly the rough set theory is used to reduce training samples and find the intrinsic relationship between fault location and system parameters. When transmission line has a fault, this algorithm can search for the most accurate fault location method by k-Nearest Neighbor method and finally give the most credible fault distance. The results of ATP simulation and RTDS simulation show that the algorithm can successfully avoid the method of larger system error and select the optimal method and greatly improve the fault location accuracy. This work is supported by Science and Technology Project of State Grid Gansu Electric Power Company (No. 522722140043).