Abstract: The anomaly recognition of spatial electric field is an important issue in studying the ionospheric disturbance caused by earthquakes. As a random digital signal, the Ultra-Low Frequency (ULF) space electric field data can be disrobed by the four features of the mean, variance, skewness and kurtosis. The ULF data recorded before the 2008 Wenchuan earthquake was used as original data to train the improved BP neural network, the identification model is established for the classification of abnormal signal of space electric field, which is verified by SOM neural network. The calculation results show that the abnormal signal is concentrated in the area between 5°~25°N and 88°~120°E, which lie within the influence scope of and may be caused by the Wenchuan earthquake. This fact is consistent with previous research results and proves that the improved BP neural network model is reasonable.

Key words: BP network, Space electric field signals, Random signal features, The 2008 Wenchuan earthquake