Short-term Earthquake Prediction Based On Data Mining

The earthquake disaster is one of the worst natural disasters which humanity is faced with.And it is also one of the most important scientific challenges. Human beings give a lot of concernabout earthquake prediction, and have devoted huge efforts about it. Since the mid-20th century,many experts from the related disciplines launched a lot of researches about earthquake prediction.With the rapid development of electromagnetic satellite technology, finding out earthquakeprecursors from the ionospheric parameters, this new technology has become a current researchfocus. The domestic and foreign scholars have conducted extensive researches on the states of theionosphere before and after the earthquake, and the results show that ionosphere disturbances areexisted before large earthquakes. And it’s also a common phenomenon. Ionosphere disturbanceswhich are existed before the earthquakes, provide strong support for the probility of short-termearthquake prediction, and for our own R&D electromagnetic satellite.After a large number of studies about previous earthquake ionospheric precursor processingmethod based on the DEMETER satellite, we use ionospheric parameters with data miningmethods, and try to find out a method for short-term earthquake forecast.The main works of this paper are as follows:(1) Analysis of the current results of the earthquake ionospheric anomalies and data miningmethods in earthquake prediction.(2) Taking into account the short durations and complex forms of earthquake precursors, andthe discontinuity of the satellite data, we propose an earthquake prediction algorithm which isbased on frequent itemsets. The algorithm ignores the time-ordered of the satellite data, looks forthe relationship between the earthquake and the combination of multiple ionospheric parameters.The experiment shows that the method of the seismic data and the determination of non-seismicdata accuracy are about80%.(3) Trying to use a mature method about voice signal processing to deal with the earthquakeprecursors, using speaker recognition model to solve the short-term earthquake prediction. Takinginto account the priori knowledge is less, we try to use self-adaptability methods to solve thisproblem. First, we use HHT to extract features parameters of the earthquake, then use GMM andHMM to train the model, finally have experiment on test data.The result proves that the method isfeasible, then we analysis the deficiency of this method.