Research On Earthquake Disaster Determination Technology Based On Communication Data Anomaly Detection

Earthquake is one of the most serious natural disasters that has caused huge human and property damage to society.The earthquake is severely destructive and unpredictable.After the earthquake,it is of great significance to quickly obtain disaster information and determine the disaster situation in the earthquake area.Communication data has the characteristics of reflecting changes in human society and human activities.The mining of communication data can effectively determine the disaster situation of earthquakes.Past experience also shows that the scope of the disaster area can be judged more accurately through communication data.Therefore,it is of practical significance to study the use of communication data to determine earthquake disasters.This paper studies the earthquake disaster determination based on communication data,including preprocessing of communication data,anomaly detection and earthquake disaster determination,and concept drift detection.The details are as follows:(1)For the problem that the original communication data can not directly reflect the regional communication situation,the research uses Geohash coding to aggregate the data with the user and time as the key to a data stream with the Geohash string as the basic unit to facilitate the extraction of the entire region.Communication situation data;(2)Aiming at the category imbalance existing in the communication data stream,an oversampling method based on the spatial distribution of the samples is studied to improve the comprehensive performance of the learning algorithm.The experiments on the UCI dataset show that the method is compared The previous algorithms can more effectively improve the comprehensive performance of the model;(3)In order to explore the applicability of the anomaly detection algorithm to the problem of seismic communication anomaly determination,the K-sigma algorithm,anomaly detection based on K-means clustering,and The applicability of three different algorithms for LSTM anomaly detection to this problem on earthquakes,and the performance of these algorithms are compared on the experimental data set;(4)Aiming at the concept drift in the communication data stream,this paper studies Degree integrated classification algorithm to detect concept drift through the performance change of integrated learning models To dynamically adjust the model to adapt the concept of change of the data stream.Experiments on real-world data sets and artificial data sets verify the effectiveness of the concept drift detection method.