Time Series Analysis Of Seismic Data Based On Neural Network

China is earthquake-prone country, and the occurrence of the earthquake brought to mankind a serious loss of life and property. Therefore, earthquake prediction is a very important topic. Earthquake caused by associated factors, the complexity of the mechanism of its birth to the nonlinear process and the difficulty of understanding issues make it difficult to establish a more complete physical theory model. Compared with traditional methods, neural network in dealing with these issues has its own unique advantages, and therefore it has wide application in many fields. This article aims to predict earthquake applying the neural network, in the analysis of seismic time series, based on the proposed new method. And finally it can be achieved with MATLAB Toolbox.The mass of seismic data which is nonlinear, high dimension, lack of values and interference characteristics, resulting in the use of a variety of analysis currently in using forecasting methods is difficult to draw a correct conclusion. One of the important tasks in this article is data pretreatment. Firstly, file format is transferred into a format commonly used, followed by processing of the noise data. In addition, it would limit the scope of fixed area, and we can select seismic data through ranging any time, any location and any magnitude through the program which can reduce the workload.For neural networks, the selection of predictor is very important. In this paper, seismic rates between different earthquake magnitudes as the inputs of neural network and time intervals of two major earthquakes as the output of neural network to predict the next the occurrence of a major earthquake. We build a new model of BP neural network in MATLAB. Meanwhile, the nonlinear input and output of the neural model are trained, and then make the final simulation prediction. Those earthquake-prone areas are selected as the research objects, for example, the Taiwan region, the Sichuan region and the East China province. In the course of the study, in order to obtain more satisfactory results, we select seismic data sequences in appropriate years. In the first place, it need to complete selection of the region to study earthquake frequency at this area in recent decades. In the second place, the associated input factors are calculated. Because seismic rates between different earthquake magnitudes as the neural network input, in this paper, we introduce the label to determine the sub-magnitude of earthquake magnitude. And then the data samples normalized are brought into the BP neural network model established. By continuous parameter adjustment, continuous learning, until the error is reached. Finally, the desired predictions are anti-normalization. The research found a potential discipline with seismic rates, finally practical application value of the model is verified.