Research On Real-time Identification Method Of Data Stream Time Series Events

In recent years,with the development of data streaming applications,the data scale has grown exponentially,such as in network monitoring,environmental monitoring,sensor networks,weather monitoring and financial services.Compared with traditional static data,data streams are infinite,continuous,large,and abrupt,and can only be read a limited number of times during data stream processing.Traditional algorithms cannot be directly applied to data stream processing.In the data stream,there are various event data that are inconsistent with normal data and are not randomly generated.Real-time identification of events in the data stream has a wide range of applications,such as network intrusion identification,environmental pollution identification,disaster event identification,weather change identification,stock trend identification,and so on.Data streams all have time characteristics.Therefore,the corresponding events are time series events.How to identify the time series events in the data stream in real time has always been a research hotspot of experts in data flow research.In this paper,the time series events constructed by the existing time-sequence event identification method are small,the model is complex and inaccurate,and it is not adaptive update and lag identification.A real-time identification method for data stream time series events is proposed,including the sequence of two-level regression.The event model(TRTM)construction method and the real-time identification method for time-series events based on TRTM ensure the efficiency and accuracy of the model construction,and have better adaptability to the new incoming stream timing event data,and improve the stability and real-time performance of data stream event identification.Firstly,in the construction stage of the time series event model,the historical time event data is normalized,and the event data of different latitudes is transformed into the same latitude,and the data features are extracted by the first-order mobile regression method to solve the problem of large data size.Then,the extracted data features are subjected to two-stage linear regression to construct a time-series eventmodel,which quantifies the trend of time-series events with time,and facilitates real-time identification of time-series events.When new time-series events are accumulated,the time-series event model is self-contained using the region compression method.Adapting to the update,effectively combining historical information and updating information,solves the problem that the time series model of the environment changes inaccurate over time.Then,in the phase of timing event identification,the timing event identification domain is constructed,and the window cross-correlation operation is used to determine the start point and drift range of timing events in the perception data.The concept of confidence factor is put forward to judge the trend of real-time sequential event perception data.Based on the concept of confidence factor,a confidence factor transformation strategy is proposed to judge the similarity between the real-time perception data and the timing event model,and to estimate the danger degree of the trend of the perception data,so as to realize the multi-stage recognition mechanism of timing events from low to high,and solve the problem of traditional time series event identification delay.Finally,for the method proposed in this paper,in the process of constructing the time series model,the error impact of the historical data scale on the time series event model is experimentally analyzed,and the optimal fit of the model is determined through experiments.In the process of time series event identification,the confidence factor transformation strategy is compared with the existing identification algorithm in terms of real-time,efficiency and reliability.Experiments show that the real-time identification method of time series events based on TRTM proposed in this paper can effectively identify time-series events in real time.