The extracted partial discharge (PD) signal is easy to be covered by all kinds of noise and interference in the insulation condition of mine high voltage cable online monitoring system and have a great impact on the accuracy of the monitoring system .This paper puts forward an algorithm, based on feature vector (EVA), which integrates improved automatic blind equalization (BE) and automatic morphological threshold (AMT) methods for noise reduction. Choosing a balance signal to reveal the PD pulse by automatic blind equalization method firstly, and then using AMT to determine the threshold according to the balanced signal，so as to achieve the purpose of noise reduction. Through the laboratory and field practical tests, the results show that the proposed method is efficient and saves signal integrity as far as possible for extracting PD signals in severe noise pollution signals.