| Accurate forecasting of earthquakes(time,location,and magnitude)is a very difficult problem.And it is more difficult to achieve this aim under current conditions.However,when an earthquake occurs,it can greatly reduce human casualties and economic losses if the hazard level and probability of this earthquake can be quickly predicted,and the damage level of this earthquake can be recognized early to help formulate rescue plans effectively.Intensity is a parameter that indicates the damage level of an earthquake and is highly related to the depth of the seismic source,magnitude,and epicenter distance.However,the existing earthquake intensity prediction methods suffer from insufficient adaptivity,few effective data,difficulty in obtaining the seismic source depth,and low accuracy of intensity prediction.The computer immune method is a heuristic algorithm inspired from the working principle of human immune system with a set of adaptive hazard perception and defense system,which has been successfully applied to network security intrusion detection,fault detection,earthquake prediction,anomaly detection and other fields.The process of earthquake hazard prediction is similar to the process of immune system’s danger perception and response to pathogen intrusion.The model consists of a danger perception layer,an innate immune system layer and an adaptive immune system layer,each layer is independent of the other,and the layers are sequentially executed.In the earthquake hazard prediction,the seismic source depth is crucial,and the acquisition of the seismic source depth requires a relatively dense earthquake station network.However,the existing earthquake network is sparsely distributed,so the seismic source depth is not considered in most earthquake hazard predictions.The Acoustic & Electromagnetic Testing All in one system(AETA)aims at collecting the acoustic and electromagnetic seismic data.More than three hundred stations have been deployed across the country,especially in Sichuan Province,where one to five stations are guaranteed to be deployed in each county,and more than five years of data have been collected so far,which can be used for the prediction of earthquake source depth.This thesis focuses on the core problem of how to construct a three-layer sequential immune model for predicting earthquake hazard level and probability based on AETA data,and proposes a set of artificial immune heuristic algorithms based on electromagnetic and geoacoustic data in AETA.The main line of research is “A threelayer sequential immune model for hazard prediction based on AETA data?A earthquake danger perception model based on danger theory?A memory natural killer cell model for predicting the depth of seismic source?A memory lymphocyte model for predicting earthquake hazard”.The "three-layer sequential immune model for hazard prediction based on AETA data" is the overall framework,which contains the remaining three sub-models.The main research content of this thesis includes the following four aspects.First,a three-layer sequential immune model based on AETA data for predicting earthquake hazard levels and probabilities was proposed.AETA data are considered as "pathogens" that invade the organism,which are characterized by fine granularity and large invasion volume.This thesis constructs a three-level sequential immunity model to assess the risk level and probability of the pathogen based on the three-level process of the immune system,namely the danger perception level,the innate immunity level and the adaptive immunity level.The danger perception layer is responsible for the preprocessing and signal extraction of the "pathogen";in the innate immune layer,the "pathogen" stimulates the innate immune cells to produce "cytokines"(seismic source depth),which activate the adaptive immune layer;in the adaptive immune layer,the adaptive immune cells assess the risk level(intensity)and probability(confidence)of the "pathogen".Second,a feature space definition method(Danger Theory based Feature Extraction,DTFE)is designed for describing the danger perception layer of AETA data with respect to the depth of the earthquake source.The danger perception layer is based on the idea of danger theory,i.e.,the immune system responds to substances that may produce "danger" in the body,and extracts the "danger" signal from the perspective of "change".Digital differentiation is used as a means of acquiring "changes",adaptively extracting relevant features of "pathogens(i.e.,AETA data)" based on "changes" and downscaling the features and mapping the danger signals.Seven machine learning algorithms are used to estimate the depth of the seismic source after extracting AETA features by DTFE method and the most widely used principal component analysis(PCA)method,which shows that the accuracy of the machine learning prediction of the seismic source depth is higher after DTFE extraction.Third,the Memory Natural Killer Algorithm(MNKA)of the innate immune layer is designed to predict the depth of the earthquake source using AETA data.Natural killer cells in the innate immune layer are able to perform induced proliferation operations on pathogens that are less concentrated and difficult to identify,establish phenotypic detectors to initially label pathogens,and generate corresponding cytokines(corresponding to different seismic source depths)to stimulate the adaptive immune layer.Based on this principle,this thesis proposes a memory natural killer cell model,which consists of four stages: pathogen proliferation,phenotype establishment,phenotype evolution and cytokine secretion,alleviating the problem that the current mainstream neural network methods for predicting seismic depth depth are dependent on the amount of effective data in predicting seismic depth depth and are prone to overfitting and local optimality.The results of seismic source depth prediction on AETA dataset by MNKA and three immune multi-classification algorithms and seven machine learning algorithms show that the proposed method obtains the best results in predicting the seismic source depth on AETA dataset with the characteristics of processing few effective data volume and adaptivity.Fourthly,the Lymphocyte Cell Differentiation Algorithm(LCDA),a memory lymphocyte differentiation model based on the seismic source depth and magnitude to predict the earthquake hazard was designed.The adaptive immune layer consists of lymphocytes that build lymphocyte detectors for specific antigens(pathogens processed by the danger perception layer and the innate immune layer)based on danger signals from the danger perception layer and cytokines produced by the innate immune layer.The danger level and probability of the antigen(corresponding to the harzard level and probability of an earthquake)are predicted.The proposed memory lymphatic differentiation model consists of three stages: lymphocyte differentiation,optimization and classification,which solves the problem that the statistical model-based seismic hazard prediction method lacks self-adaptation and needs to set parameters in advance for a specific region;the artificial neural network model is less accurate due to the small amount of effective seismic data,which is prone to local optimum and overfitting problems.Two experiments are designed in this section: the results of Experiment 1show that the LCDA algorithm proposed in this thesis achieves the best results in predicting earthquake hazards based on AETA data,LCDA has advantages over other methods in dealing with AETA datasets with little effective data,and the introduction of clone selection increases the adaptiveness of the model.And the results of Experiment 2 show that the seismic hazard prediction results with the seismic source depth considered are more accurate than those without the seismic source depth,and the integration of the seismic source depth parameters in the seismic hazard prediction process is achieved. |
