Research On Emergency Sensing And Prediction Of Large-scale Activity Based On Spatial-temporal Data

With the rapid developments of China's economy,large-scale activities have been held more and more frequently.Once an emergency occurs in an activity,and it will make massive casualties and economic losses.How to accurately detect and predict the crises of large-scale activities in real-time has been become a critical issue in research area,which needs to be resolved urgently for the safety management of large-scale activities.In terms of practical applications,such as the Internet of Things,cloud computing,big data,and smart cities are rapidly developing.Moreover,the mobile communications,BDS(Bei Dou Navigation Satellite System),and GPS(Global Positioning System)global positioning,video surveillance,and other sensing technologies and equipment are popularizations.These technologies not only made large-scale activity venues accumulate a large amount of spatiotemporal monitoring data,but also brings new challenges and opportunities for the detecting and prediction of activity emergencies.In research area,several mature models,technologies and methods have been raised in the fields of big data analysis,mining research,and applications.These achievements provide a solid theoretical basis for analyzing and researching based on big spatiotemporal data.These results will do much help for the emergency perception and prediction.It can not only avoid the loss of personnel and property to the utmost extent,but also greatly help emergency decision-making and resource scheduling.Unlike traditional event perception and prediction methods supported by empirical data,we focus on four critical issues of large-scale activities emergency sensing and prediction.Using machine learning and data mining methods to discover patterns of emergencies based on big spatiotemporal data.The main work and innovations are as follows:(1)Modeling and classifying historical data of large-scale activities based on time series.Firstly,modelling the large venues based on graph structures.These large venues have the same physical structure,Secondly,modelling historical data for large-scale activities.Moreover,proposing a crowd density model for large-scale activities based on time series.Then,giving the activity similarity measurement method.Finally,clustering the data at specific time intervals and spaces based on KMeans.Furthermore,finding the patterns to realize automatic classification of largescale activities.(2)Emergency sensing based on Markov Model.Existing studies often simply use historical data to determine the current state,which makes it difficult to sense the process of event state changes.In this paper,we firstly proposed a regional temporal state model based on the crowd density time series.Secondly,the Cartesian set of crowd density level and the time span is used as the state,in order to realize the model expression of crowd density state over time.Finally,designing the emergency sensing algorithm based on the above model.Comparing the current activity state with the patterns summarized by the model.Then we can discover the abnormal to realize the activity emergency sensing in a specific area.(3)Emergency prediction based on spatiotemporal context time series.Existing researches of emergency prediction mainly focus on the micro-events in some specific fields.Applying existing results directly to predict the critical situation in large-scale activity is a big challenge.In this paper,we show a novel method,which integrates relevant research results into a unified spatiotemporal model.Firstly,constructing the historical spatiotemporal context time series based on historical activity data.Then,dividing the time series into time period and time window.Finally,exploiting the time series' spatiotemporal patterns based on KNN(K-Nearest Neighbor),in order to predict the emergency of current activity.(4)Implementing the prototype system of emergency sensing and prediction.To better verify the research algorithm and results of this paper,and solve the problem of applying the results in the real world,we implement a prototype system for large-scale activities emergency sensing and prediction,which is designed based on HTML5 technology.In this paper,we realized the construction of the prototype system and Web application development based on the proposed emergency sensing and prediction algorithm.Then,demonstrating the function of the prototype system in the background of an example.