Research And Application Of Artificial Neural Network Based On Rough Set In Earthquake Prediction

At present, earthquake prediction is still a worldwide scientific problem, human have no idea of the causes and the regular pattern of the earthquake. As a result, the research on earthquake prediction is still at the exploratory stage. The practice of earthquake prediction research proves that the birth and occurrence of the earthquake is very complicated. It is a large-scale and underlying change process in the macro nature. And it may involve various anomalies in geophysics, geology, surveying, geochemistry, meteorology and other disciplines. In the process of exploration and study, people gradually found that there is a certain relationship between the magnitude and the types of seismic anomalies, the duration of seismic anomalies before the earthquake. This relationship is a strong uncertainty and nonlinear mapping which is difficult to analyze by simple expression. It makes big difficulties and limitation for earthquake prediction. Therefore, in the field of earthquake prediction study, faces with lots of observational data accumulated, it becomes an urgent problem that how to accurately identify the core anomaly indicators, how to build a generalization model to simulate the nonlinear relationship between the anomalies and the magnitude.Rough Set (RS) is a novel and effective soft computing method, which can analyze and handle the uncertain problems. The main idea of RS is to delete irrelevant or unimportant attributes, and meanwhile reserve the core attributes which plays a decisive role in classification, in the condition of maintaining the same classification. Artificial Neural Network (ANN) is a simulation of the brain biological neural network, which is a network model constituted of large numbers of processing neurons nodes connected by a certain structure. ANN is used for dealing with vague, imprecise data and complex nonlinear mapping problem. Therefore, bringing the theory of RS and ANN into the field of earthquake prediction is a good choice, especially to identify the key seismic anomalies indicators, and to analyze the nonlinear relationship between the anomalies and the earthquake.China Earthquake Case is a series of books, which recorded earthquake cases study report, contains all the earthquake cases reports which occurred in China from 1966 to 2002, and whose magnitude is greater than 5.0. (Later add four earthquake cases between 4.5 and 4.9). The book gives a detailed summary and analysis on the basic seismic parameters, geological background, anomalies of each earthquake case. So far, it is the most important, complete and comprehensive materials for earthquake regular pattern exploration and earthquake prediction research.Taking into account of the puzzle in earthquake prediction, according to the fact that RS and ANN can complement each other and is good at resolving the related problems, this paper proposed a model based on the combination of RS and ANN. Firstly, it systematically sorted out the data of all the earthquake cases that was recorded by China Earthquake Case, and stored them in the database established using SQL SERVER 2000, as the data source of analysis and experimental prediction. Secondly, the system called the database to obtain the earthquake cases data including abnormal information and it is necessary to do some data preprocessing such as data discretization. Thirdly, the system implemented the attribute reduction algorithm based on the discernibility matrix using Java, to extract the core attribute from a large number of seismic data. Finally, this paper builds a nonlinear mapping relationship between seismic anomalies and magnitude through constructing a BP Neural Network in the environment of Matlab. In this way, it would be an attempt to predict earthquake. Through the study and implementation of the system, this paper expected to provide more reasonable and objective suggestions for earthquake prediction, in order to improve the prediction accuracy.