A Computer Immune Model For AETA Earthquake Prediction

Earthquake hazards are serious,but timely and accurate prediction can effectively reduce damage.The observation and monitoring of precursor anomalies,which are often present in various types and quantities before the occurrence of an earthquake,is an important tool for earthquake prediction.The existence of electromagnetic and geoacoustic data anomalies before an earthquake has been confirmed by a large number of earthquake cases,and their acquisition and analysis have gradually become a hot research topic in the field of earthquake precursor prediction.The Acoustic & ElectroMagnetic Testing All in One System(AETA)is a high-density,large-scale monitoring station deployed throughout the country,and each station consists of a high-precision inductive magnetic sensor and a piezoelectric thin-film geoacoustic sensor.AETA provides long-term,stable and real-time observation data for earthquake precursor prediction.When making earthquake prediction based on AETA data,two difficulties are faced:1)AETA data class imbalance is a serious problem.The existing prediction methods are often ineffective without sufficient anomalous samples for training;2)it is difficult to distinguish between seismic and non-seismic foreshadow anomalies;the non-seismic anomalies generated by AETA devices when they are disturbed by themselves or external environment are similar to the seismic anomalies,which makes it difficult to accurately distinguish between seismic and non-seismic anomalies.The existing methods make it difficult to accurately distinguish between seismic and non-seismic abnormalities.Inspired by the mechanisms of natural killer cell concentration activation,negative selection,clonal selection and multicellular collaboration in the immune system,a computerized immune model was developed to improve the performance of AETA earthquake prediction.The main work is as follows.Drawing on the natural killer cell concentration activation mechanism and combining with the characteristics of AETA data,a safe region is generated for each abnormal sample and assigned different weights,and a variable number of new samples are generated within its safe region according to the weight size to alleviate the class imbalance problem in AETA data.The negative selection algorithm can detect a large number of anomalies without relying too much on the anomalous samples,and it has been widely used in the field of AETA seismic prediction with sufficient normal samples for training.However,the negative selection algorithm is "noise-sensitive",so the algorithm is improved by introducing the concept of "density" and using the clonal selection algorithm for parameter tuning to achieve better prediction results.Introduce multi-cell collaboration mechanism.It is difficult for a single cell or a certain cell to make a correct immune response to a pathogen,and multi-cell collaboration can make up for the shortage of single cell immune response.Inspired by this,AETA electromagnetic and geoacoustic data are divided into different feature subsets in the time domain and frequency domain,and modeled separately as different kinds of immune cells,and the immune cells recognizing time domain anomalies and recognizing frequency domain anomalies collaborate with each other,which can better distinguish the seismic and non-seismic anomalies and finally make predictions of earthquake magnitude range.