Earthquake Disaster Assessment Based On Anomaly Detection Of Communication Data Flow

As a serious geological disaster,earthquakes will cause certain casualties and property losses every time they occur.A reasonable assessment of the disaster situation after the earthquake can effectively provide a decision-making basis for earthquake relief.With the development of science and technology,the number of mobile phones in our country is increasing,which can collect a large amount of communication data,and the communication data can reflect the changes in human activities.The use of abnormal changes in communication data can be used to evaluate the disaster situation,obtain the grid disaster degree,and display it on the map to provide auxiliary support for emergency rescue.Based on the abnormal changes in the communication data flow after the earthquake,this thesis evaluates the extent of the disaster in the disaster situation.The main research work is as follows.First,the data processing flow is designed according to the characteristics of the communication data.The feature reduction method based on neighborhood rough set is improved,and feature reduction is performed on communication data.The reduction effect of the method is verified on the test data set.Aiming at the anomaly detection problem of communication data flow,a data flow anomaly detection model based on LSTM+difference normal is designed.Use LSTM to predict the data,model the normal distribution of the prediction difference in the set window,and judge the abnormality according to the reciprocal of the probability density of the current prediction difference in the normal distribution as the abnormality score.Experiments on the NAB data set verify that the model's anomaly detection ability is significantly better than the K-sigma model and the Meanshift clustering model.Furthermore,in view of the changes of data distribution in the data stream over time,this thesis designs a data distribution detection method based on SVM model parameters.The parameter difference of the SVM is trained through the data of the two windows before and after,and the change of the data distribution is detected.The data and categories of the two data windows are input into the SVM model,and the classification plane vectors of the data of the two windows before and after are calculated,and the performance change is judged according to the cosine value of the angle between the vectors.Compared with ADWIN and DDM methods on data sets such as electricity and air pollution,the accuracy of this method is better.Finally,this thesis uses the change trend of communication data over time to adjust the disaster assessment.Based on regional spatial distribution statistics,a method for adjusting the degree of damage to the G statistic is designed.Analyze the G statistic of the degree of grid abnormality.A clustered grid will be presented,and the degree of damage will be reassessed using the centrality of the feature vector.Through time and space analysis,the accuracy of disaster assessment has been improved.Based on WebGIS,disaster data storage and disaster display are realized,and the grid damage degree is displayed on the map.