Aiming at the problems of poor real-time performance and low time-frequency resolution in existing detection algorithms for composite power quality disturbances, a real-time power quality disturbance detection method based on improved adaptive S-transform (IAST) is proposed. A globally adaptive Gaussian window is constructed as the kernel function of the IAST, allowing the effective window length and frequency spectrum to adapt dynamically with the detection frequency. This avoids the need for frequent switching of window parameters to improve time-frequency resolution, thereby reducing algorithm complexity. The window parameters are optimized with the objective of enhancing signal energy concentration, ensuring accurate time-frequency positioning of various types of disturbances. An automatic thresholding method is used to determine the dominant frequency points of the actual disturbance signals, which are then subjected to time-frequency transformation to further improve computational efficiency. Simulation and experimental results show that, compared with existing algorithms for detecting composite power quality disturbances, the proposed method offers superior real-time performance, strong time-frequency resolution, and low computational complexity, making it suitable for accurate real-time detection of complex power quality disturbances.