With the popularization and development of new energy grid-connected technology, the large number of non-linear devices connected to the power grid have an impact on its power quality. Thus it is necessary to detect and analyze that impact. To help overcome the shortcomings of existing detection and recognition methods in anti-noise and accuracy, a modified LMD method is proposed in this paper. This approach first studies the mechanism of the selection process for the adaptive decomposition method, and then analyzes the degree of extreme point fitting distribution that is susceptible to high frequency and intermittent signal interference. It uses the noise-assisted decomposition method to add controlled Gaussian white noise to the original signal and then perform LMD decomposition. Then, taking into account end-point energy leakage in the feature parameter extraction, an empirical modulation decomposition method is proposed to detect instantaneous parameters. Simulation results show that the proposed method is able to effectively suppress mode mixing and endpoint effects. Finally, experimental data from the built power quality disturbance platform demonstrates that the proposed method is capable of accurately extracting all disturbance parameters of voltage sag. This also provides a novel method for power quality disturbance analysis. This work is supported by National Key Basic Research Program of China (No. 2016YFB0101800).