Abstract: Voltage stability of distribution network is of great significance for safe and reliable operation of distribution system. Aiming at the unexpected and concealment characteristics of the uncertainty of wind power, component fault and load fluctuation in the complex active distribution network, and the concealment of the voltage instability, a method of rapid perception and early warning for the risk situation of voltage instability in distribution network based on data mining and deep learning was proposed in this paper. Taking into account the uncertain factors of source network load and other factors, through feature extraction and derivative methods, the environment factors, power flow state and equipment state which may cause voltage instability are analyzed, and multi-dimensional feature sets are established, feature importance evaluation and redundant redundancy are carried out by improved information entropy judgment method. Integrated distribution network voltage instability probability and severity calculation method, the distribution network instability risk is divided into three warning levels. A deep neural network algorithm is introduced to establish the situational awareness model of voltage instability risk and to conduct early warning and grading of voltage instability risk. The actual example and the improved IEEE37 system example verify that the proposed method has a high early warning accuracy and can provide an effective basis for operation and maintenance of distribution network.