The invention discloses an urban medium-voltage distribution cable partial discharge signal denoising method. In order to suppress periodic narrowband interference, a K-means clustering algorithm is used to classify noisy partial discharge signal spectrums. In order to suppress white noise interference, an improved empirical Wavelet Transform (EWT) algorithm based on an Order Statistic Filter (OSF) is provided, and the method comprises the following steps: firstly, estimating an envelope line on a signal frequency spectrum by using the OSF, reasonably segmenting the signal frequency spectrum, then, carrying out adaptive decomposition on a signal by using the EWT, introducing a kurtosis criterion to select a useful modal component, and finally, carrying out adaptive decomposition on the signal by using the EWT. And finally, reconstructing a partial discharge signal through threshold denoising. In combination with the K-means clustering algorithm and the empirical wavelet transform, the method provided by the invention does not need to modify algorithm parameters, can perform adaptive denoising on white noise and periodic narrow-band interference at the same time, can effectively reserve signal details while having a high denoising signal-to-noise ratio, and is relatively high in algorithm efficiency.