The Earthquake Prediction Methods Research Based On Improved Support Vector Regression Machine

The destructive force of earthquake disasters is extremely great,it always occur inadvertently, it comes with great economic loss and great number of casualties. Also, China is one of the countries who has much earthquake activities. Doing well in earthquake prediction and disaster reduction is a necessary choice to sustain the development in our country. In order to predict the occurrence of earthquakes, researches from every country are all study on the reason why earthquake occur and its regularity. After so many years of research, it shown that earthquake has certain regularity in a region of strong earthquake event, which has certain periodicity. The discovery of such a rule for earthquake prediction research work provides a very good idea.There are many methods can be applied to earthquake prediction, wavelet function can reflect the non-stationary signal characteristic of the time-frequency domain. In this paper, at first, using wavelet transform in the time series of energy which the moderately strong earthquake activity release, then get the characteristics of earthquake activity in different time scale, then using the wavelet coefficient and the active period of the earthquake which obtained through the wave transform to do the further research.The sample of earthquake is dispersed and the number of it is quite small, it is difficult to use the exact formula or methods to expression its regularity. While the support vector machine overcome the disadvantage exist in traditional methods, and it has better prediction abilities, it fits the research on modeling and prediction of earthquake discrete data, the study also found that the predictive ability of SVR is better than neural network algorithm and other traditional statistical learning method. In order to simplify the complexity of calculation, This paper used the improved support vector regression machines to predict the maximum magnitude,the improved SVR simplifies the complexity of the calculation,and also gain good effect. It is a good choice for earthquake prediction.The main works of this paper are as follows:(1) At first, using the active periods of the earthquake which obtained through the wave transform to do the further research in China. Since the activity of earthquake has periodic regularity, the wavelet analysis algorithm can get the characteristics of earthquake activity in different time scale.(2) The improved support vector regression machines reformed the expression of hyper-plane,simplifies the complexity of the calculation.(3) We first use the active period which obtained from the wavelet transform as the time window,and then use the improved support vector regression machines to predict the maximum magnitude. Then use the effective frequency N、the maximum magnitude Mmax、the average magnitude (?)、the equivalent energy n* as the input parameters of SVR. Compared with the real magnitude,the result is quite good. When compared with neural network algorithm,this combined algorithm gain good effect.