The Analysis And Implementation Of Earthquake Prediction Algorithm Based On BP Neural Network

Earthquake is the vibration caused by earth’s crust rapid release of energy, producing shock waves during a natural phenomenon. About5.5million earthquakes happen around the world every year. Earth-quakes tend to cause serious casualties and property losses. China is an earthquake-prone country, so the earthquake prediction research has very important significance. The cause of the earthquake is complicated, and understanding of the cause is not deep enough. Seismologists can’t observe the interior of the earth directly, so influence factors in the process of the earthquake observation data is exile. This caused the slow process of earthquake prediction research, and the accuracy is not ideal.With computer technology high speed development, a large number of prediction algorithms are developed, and the prediction algorithm in efficiency and accuracy is better than traditional prediction methods, In this paper, BP neural network algorithm which is very classic in machine learning is used to construct a earthquake prediction model.A very important step in earthquake prediction process is selecting features. The research of this paper is predicting the numbers of earthquake of every magnitude in a certain target area with a given year. Considering the weight of low-magnitude earthquake error in total error will be too large, if the numbers of earthquake is put as the target output of BP network, so this paper adopt the method which put logarithm of the numbers of earthquake as the target output. In order to make the data more suitable for BP network to train and predict, historical earthquake data must be preprocessed before input the data to BP network for training. The main job of this part is reading into the record sequentially, then choosing records with the information of time, latitude and longtude, and get the logarithm of numbers of these records. Finally, the input and output are normalized. Small input values can make the network easier to converge. BP network with sigmoid units output range-from0to1,so normalizing output is also necessary.Finally, the model with year being the input of BP network was trained with data provided by Chuzhou Seismological Bureau. After several times modifying weithts, network weight converge. Then this network is used to predict the numbers of earthquake in Sichuan and Taiwan, and it shows good consequent, proving the rationality and value of this model.