Research And Implementation Of Anomaly Detection System Based On Edge-Cloud Collaboration

With the development of edge computing technology,the equipments at the edge can provide more and more powerful capability.Therefore,anomaly detection can be carried out on the edge based on the running state data collected by the terminal device.Compared with anomaly detection relying on the cloud platform,anomaly detection at the edge can effectively reduce the data processing delay and pressure of the cloud platform.However,the implementation of traditional edge anomaly detection mainly relies on rule analysis,which is difficult to meet the requirement of complex anomaly detection.This thesis aims to design and implement an anomaly detection system based on edge-cloud collaboration.The system accomplishes anomaly detection of running data through the collaboration of edge and cloud.The edge nodes receive real-time data which are collected by the devices.These data are analyzed to detect anomalies based on rules and models.The data analysis module based on rules can process the data at a shallow level,while the module based on models can conducts in-depth analysis.The cloud platform receives the results of exception and sends alarm messages to the manager.In the data analysis module about the model,this thesis proposes a novel algorithm called A Deep Learning Framework for Continually Learning in Anomaly Detection based on Cloud-Edge Computing(CLAD).The method considers both the minimum data transmission and the continuous adjustment of models.A fast anomaly detection method is designed based on autoencoder and ridge regression classification on the edge side.It uses fewer parameters to extract rich data features and uses the results of cloud platform detection to adjust the training of edge model.A novel anomaly detection method with high accuracy is proposed based on multi-head attention mechanism and feedforward neural network.It increases the local feature of the input signal to enrich the information focused by the model.Experimental results show that CLAD outperforms CTF,OmniAnomaly,US AD and other baseline methods in terms of model performance.This thesis firstly introduces the research background of anomaly detection system based on edge-cloud collaboration.Then,the thesis proposes the requirements of the anomaly detection system based on the cooperative evaluation of rules and models,which relies on the investigation and analysis of existing anomaly detection systems.Next,CLAD is introduced in detail.Moreover,this thesis describes the design and implementation of an anomaly detection system which is based on cooperative evaluation of rules and models.Finally,a series of functional tests and non-functional tests are carried out to verify the effectiveness of the system.Test results show that the system runs smoothly and can be widely applied in real scenarios.