In order to identify the short-circuit fault of high-voltage transmission lines fast and accurately, a novel method for fault detection and phase selection is discussed. Based on shift invariant features of sinusoidal signals after matrix transformations, applying matrix transformation function with 4 sampling points ultra-short data window, the short-circuit fault is detected by dynamic sliding window. The sampling values of each phase current and voltage are used to calculate their K values after short-time dynamic denoising, through comparison of the K value and the threshold, the fault type is identified. A short-circuit fault model of 220 kV dual-source power transmission system is established in MATLAB/Simulink environment, simulation results show that the method has high sensitivity and reliability and it can accurately identify the short-circuit fault under different conditions, theoretically, the time of fault detection and phase selection does not exceed 5 ms.