Research Of The Intelligent Anomaly Detection Technology For Industrial Internet Of Things

The development and application of IIo T（Industrial Internet of Things）greatly improve the industrial production efficiency,which also bring more severe network security challenges.As an important means of maintaining network security,anomaly detection for IIo T becomes increasingly valued.This thesis researches on the intelligent anomaly detection technology for IIo T,summarizes the current research status and challenges,refers to and studies a variety of existing anomaly detection technologies of IIo T,researches on the real-time detection technology and the time series technology for online and offline scenes,respectively,and combines the online detection technology with offline detection technology to further improve anomaly detection efficiency.The main research contents of this thesis are as follows:· Applying to the online real-time detection scenarios,this thesis researches the real- time detection technology based on online analysis from the real-time requirements of IIo T,designs a feature filter and proposes an elastic lightweight broad learning system（EL-BLS）based on the BLS（Broad Learning System）,which further re- duces the time overhead of anomaly detection on the basis of maintaining the ability of rapid detection and incremental learning,and has better classification detection capabilities.· Applying to the offline time series analysis scenarios,this thesis starts from the time-dependent perspective of IIo T,researches the offline time series technology based on offline analysis,and proposes a feature-attended multi-flow long short- term memory（FAMF-LSTM）model,which has high generalization and stabil- ity,and achieves excellent detection effect.Optimizing on it,a feature grouping algorithm is designed and a feature-grouped multi-flow long short-term memory（FGMF-LSTM）model is proposed,which greatly reduces the calculation overhead.· Applying to the online and offline comprehensive detection scenarios,this thesis researches the online and offline collaborative intelligent detection technology,de- signs an architecture of online and offline collaborative intelligent anomaly detec- tion system,and proposes a feature automatic updating online detctor（FAUOD）, which combines the real-time detection model and the time series analysis model proposed by expanding the feature filter,and makes them cooperate together under the architecture to further enhance the detection capability.The experiment of the above content is conducted on four popular Io T related datasets with evaluation criteria such as accuracy and recall rate,and compared with a variety of mainstream detection methods.The experimental results show that the proposed anomaly detection technique can effectively improve detection efficiency.