Anomaly Detection Research Based On Drosophila Olfactory Circuit And Stream Clustering

With the rapid development of 5G communication technology,industrial Internet of Things(IIo T)systems are widely used in national infrastructure.Among them,the occurrence of anomalous events seriously affects national security and people’s livelihood.The massive amount of data generated by large-scale connected devices has transformed the object of anomaly detection from a traditional static data set to a dynamic data stream with infinite,high-dimensional and fast transmission characteristics,which poses a great challenge for real-time and effective streaming anomaly detection in IIo T.Therefore,it is very urgent and meaningful to research an effective anomaly detection study in real time to ensure the security of IIo T.Although there are many relevant research results in industry and academia,there are still some problems with anomaly detection techniques when facing data streams,as follows.First,the conceptual drift of data streams causes some data to be misclassified as anomalous,resulting in a higher anomaly detection rate than the actual value;second,the "dimensional catastrophe" of data streams,in which anomalies are usually localized in a high-dimensional space and a large number of redundant calculations are performed during the computation process;third,for complex log streams Third,there is no reliable and specific solution for detecting hidden anomalies in complex log streams.Therefore,how to efficiently perform stream data anomaly detection has become an urgent problem.To address the above problems,this paper presents a detailed analysis of current stream anomaly detection techniques,and proposes an anomaly detection algorithm based on Drosophila olfactory circuit and stream clustering based on stream clustering anomaly detection mechanism,and builds an industrial Io T real-time log stream implicit anomaly detection system based on distributed system model.The superiority of the proposed study in this paper is demonstrated by system implementation and full experimental comparison.The main contributions of this paper are as follows.(1)For the industrial Io T data stream concept drift problem,a cache structure-based industrial Io T stream clustering anomaly detection algorithm is proposed to alleviate the concept drift problem by combining the cache structure with a dynamic recursive update function to improve the detection capability of whether the data belongs to anomalous data or a concept drift has occurred.(2)For the problem of "dimensional disaster" in industrial Io T data streams,this paper proposes a clustering anomaly detection algorithm for industrial Io T streams based on the "winner-takes-all" strategy of Drosophila olfactory circuit mechanism and the habituation property,which enhances the high aggregation of similar data and the low coupling of different classes of data.(3)For the problem of lack of specific solutions for complex log stream data implicit anomalies,this paper constructs an industrial Io T real-time log stream implicit anomaly detection system on the model of distributed system.The system uses different rule bases to vectorize log stream data and uses an adaptive algorithm selection mechanism for anomaly detection of data sources,which effectively alleviates the problem that specific algorithms cannot solve complex log stream implicit anomaly detection.