Research On Earthquake Prediction Based On SOM Artificial Neural Network

The earthquake is one of the most destructive nature disasters, which threatens the social economic development and the human lives and property safety. China is one of the countries with fierce earthquake activity and serious earthquake disaster in the world. Therefore, earthquake prediction is an important subject of contemporary seismology. Because of the diversity of factors caused earthquake, the complexity of birth mechanism and high nonlinearity. it is difficult to use kinetic equations to describe. Artificial neural network has strong nonlinear ability, and has predominance over traditional prediction methods. It was applied to earthquake magnitude prediction in the thesis, combining SOM with RBF, which improved the prediction accuracy effectively and certainly has practical apply value on earthquake predictionThe main work of the thesis includes:(1) It discussed the principle and presented the research of earthquake prediction and artificial neural network, surrounding the problem in the earthquake prediction. And in order to solve this problem, it presented a new method of applying artificial neural network to future earthquake magnitude prediction, which aimed to reduce the influence of subjective factors and improve the accuracy and reliability of earthquake prediction.(2) The select of predictors has an important influence on prediction accuracy. The thesis used the most effective seismology precursors as predictors, choosing the border area among Fujian,Guangdong and Jiangxi Provinces and southeastern coast area as the research object. The data were both calculated by software and checked artificially to ensure their accuracy and reliability.(3) Because of the small number and uneven distribution of samples, SOM neural network was used to classify the predictors firstly and the samples were studied and predicted by establishing RBF network, which can prevent the BP network from falling into local minimum. (4) The normalized test samples were brought into the artificial neural network model established by SOM and RBF, and were classified, studied and predicted until the error expectation was reached. Finally the desired predictions were anti-normalization. The research indicated that the method obtained good result on future earthquake magnitude prediction and is feasible and reasonable in earthquake prediction.