The Earthquake Prediction Model Based On The Adaptive Immune Mechanism

Earthquake is a devastating natural disaster,which cause great economic losses and casualties to the whole world every year.It is very important to predict earthquakes.But it is very difficult to predict earthquakes because of the inaccessibility of the earth,the frequency of large earthquakes and the complexity of the causes of earthquakes.The three elements of earthquake prediction are time,location of earthquake center and magnitude.Usually,the time and location of the three elements are fixed,and whether an earthquake within a certain magnitude range will occur are predicted.This problem can be abstracted as a classification problem.At present,the main research directions are divided into earthquake prediction based on precursor data and earthquake prediction based on historical data.Because of the high integrity and reliability of historical data,this paper chooses earthquake prediction based on historical seismic data as the research direction.Although some achievements have been made in earthquake prediction,the following problems still exist at present.Because of the complexity of the causes of earthquakes,the main influencing factors of earthquakes occurring at different times and locations may be different,and the seismic indicators describing different earthquakes are also different.At present,most methods use the same seismic indicators for all seismic data to make the representative indicators are not representative and the prediction effect is poor.At the same time,due to the lack of seismic data,especially large earthquake data,supervised machine learning methods which need a large number of abnormal data sets for training is easy to lead to fall into local optimum,resulting in over-fitting problems.In view of the above two problems,this paper inspired from the high adaptability and specificity of the biological immune system,and draws lessons from the specificity,memory and clonal selection mechanism of the adaptive immune mechanism to solve these two problems.In order to solve the problem of poor representation of feature indicators,specific mechanisms and memory mechanism are used for reference,the optimum indicator subset and the combination of prediction sub-models are set for each type of seismic data,the corresponding relationship between the processed seismic data and the combination of the optimum prediction sub-models are recorded at the same time,so it can be processed by the same prediction sub-models combination when the same kind of seismic data is input.Using clonal selection mechanism for reference,we use clonal selection algorithm to optimize the combination of prediction sub-models for each type of seismic data,and find the optimum combination of prediction submodels globally to avoid falling into local optimum.Based on the above bio-immune inspiration,this paper proposes an earthquake prediction model based on adaptive immune mechanism.The model consists of three parts: pre-processing stage,primary immune response stage and secondary immune response stage.In the pre-processing stage,the seismic indicator data are first calculated,and eight classical characteristic indexes are adopted.Then the feature set is segmented,and the feature subset data matrix is established by data mapping.In the stage of primary immune response,the immune cell pool was first established,then the immune cells were activated,and the combination of immune cells was optimized by clonal selection,and finally form immune memory.In secondary immune response stage,memory B cells were first matched and then activated to process antigens.Finally,the historical seismic data of Sichuan Province are used as input to predict whether earthquakes of magnitude 4.5 or above will occur in the next month.The prediction effect is evaluated by using detection probability,false alarm rate,accuracy rate,R score,ROC curve and area under the curve of ROC as evaluation indexes.The experimental results are compared with the mainstream method of earthquake prediction,neural network.The experimental results show that the accuracy rate and R score of the proposed algorithm are higher than those of the neural network.