In order to search of ionospheric anomaly information before the Wenchuan earthquake, this paper selects 10 days DEMETER satellite field data before Wenchuan earthquake as the research object and select mean, variance, skewness and kurtosis four kinds of random signal digital features as the input layer. After a number of samples training, Self-Organizing Map neural network clustering model is established. Calculation results show that, before the Wenchuan earthquake, earthquake signal acquisition of satellite exist some abnormal data exist some abnormal data, which may contact with ionospheric disturbances caused by electromagnetic wave radiation before earthquake.