Research On The Early Judgment Of Earthquake Sequence Types Based On Artificial Neural Network And Pattern Recognition Method

The research of seismic sequence types and its formation mechanism is one ofthe basic questions in seismology; meanwhile, it has theoretical significance for therevelation of the earthquake preparation and developing process of physical nature.For a complete seismic sequence, it has mature methods and parameters to recognize.The study of the thesis is to make a rapid determination for the seismic sequence typeswhen the earthquake happens, and the sequence is still not completed, well, it is aproblem for early recognition of seismic sequence types.In this paper, based on summarizing current researches of seismic sequence types,the previous single and comprehensive methods to judge the seismic sequence typesearly are discussed in detail, which includes the grey relational analysis method andthe CORA-3pattern recognition algorithm method and so on. Besides, the paper hasnot only affirmed the important role they play in the early trend after earthquake, butalso analyzed their shortcomings. And in the article the author has analyzed theadvantage of artificial neural network and support vector machine (SVM) method indealing with the complex nonlinear mapping problem, has collated180moderate-strong seismic sequences in china, and constructed two early predictionmodels of seismic sequence types by the BP neural network and support vectormachine (SVM) method.The major research results of the thesis are:①The paper has collated180seismic sequencesin mainland area of China, and has done the initial classification on complete sequences and hasdivided them into three types on the basis of classic discriminative indexes, such as energy ratio、magnitude sent；②Through the continuous experimental research, the paper has structured theseismic sequence’s early prediction model of BP neural network by the following steps：confirmthe structure、choose activation function、set up the training parameters and so on, then hascarried on the internal and extrapolation inspection respectively according to the training samplesand the test samples, the experiments show that it has better result in recognition，the extrapolationinspection discrimination of1day (24hours) can be up to78%；③By selecting the kernel functionand the parameters optimization method （c&g）, the paper has structured the seismic sequence’searly prediction model of SVM, and has also carried on the internal and extrapolation inspectionrespectively by this model, and it is proved to be a effective recognition, the discrimination of1day can be up to82.2%. In addition, the main shock type and isolated type of five time periodsinclude1、2、3、5、7days have been provided to keep higher identification accuracy；④Throughthe way of statistical analysis, the paper has analyzed classification effects of the twoclassification methods, and considers two methods is feasible and practical for earlyprediction of seismic sequence. The methods have good discrimination ability，whichcan judge sequence types well by using1day sequence after the earthquake. It isshown by comparisons that the recognition results of SVM model are better than theBP neural network model constructed in this paper.The innovations of this paper are as follows:①With relatively large samples, theauthor has carried research on early prediction of seismic sequence types by using BPneural network and support vector machine methods, and the research has coveredmore wide-ranging sequence types, which made the recognition results morepervasive;②Most previous research merger the sequence types for two kinds：mainshock type and earthquake swarm type. This paper is composed of three parts: mainshock type、earthquake swarm type and isolated type，which lead the classificationmore elaborate;③The methods of BP neural network and support vector machine used to predict the earthquake sequence types early have the advantage of lowcomputation complexity, and the arguments are further confirmed that1day sequencematerials after the main earthquake can distinguish the sequence types accurately, andthe strengths and weaknesses of two aforementioned methods is comparative studiedas well.