Research On Reliability Anomaly Detection Methods And Key Technologies For Edge Services In Urban Brain Applications

Empowering urban management with digital technology is an important trend in continuously advancing the modernization level of social governance systems and governance capabilities.As a complex intelligent information management system for cities,the urban brain can assist urban managers in allocating public resources,making scientific decisions,and improving governance efficiency.The construction of urban brain depends on the technological support of Big data,artificial intelligence,blockchain,digital twins and other technologies,as well as the computing support of mobile edge computing(MEC).However,in the mobile edge computing system,due to the dynamic and uncertainty of the internal operating state and external operating environment,the runtime reliability of the mobile Limbic system will fluctuate with uncertainty,which will affect the service quality of the city brain.Therefore,in order to meet the low latency service requirements of the urban brain,how to accurately detect runtime reliability anomalies of edge services in real-time is a fundamental scientific issue to ensure the high latency sensitive applications mentioned above.In the dynamic and changeable mobile edge computing system operating environment,the real-time change characteristics of edge service reliability are extremely complex,with massive,continuous,unstable and other characteristics.In the field of data mining,the phenomenon of unpredictable changes in the distribution pattern of data over time is called concept drift.So this article proposes a reliability anomaly detection method called B-Detection based on concept drift to detect negative changes in the real-time reliability data flow distribution of edge services.The B-Detection proposed in this paper is an anomaly detection model based on Long short-term memory(LSTM)automatic encoder.In addition,considering factors such as system architecture upgrades and software version updates that can lead to changes in the normal operating state of the system,there is a risk of model performance degradation.In response to this issue,this article designs a Boosting algorithm for improving model performance for the B-Detection method.On this basis,this article also proposes a computation offloading scheme after the occurrence of anomalies,and provides management suggestions to improve the quality of urban brain system services.The main work of this article is as follows:(1)In order to meet the modeling challenge of complex distribution characteristics in reliability data stream,the B-Detection method in this paper uses the encoding and decoding structure of LSTM Autoencoder to learn and characterize the reliability data distribution of edge services.Firstly,a weighted(random)reservoir sampling algorithm is proposed to extract samples of normal reliability data streams under stable operation of edge services.Then train the Autoencoder to let the model learn the normal data distribution characteristics,so that the model has a good reconstruction ability for normal data,while it is difficult to reconstruct abnormal data.Finally,by reconstruction loss,B-Detection was able to detect anomalies in data with significant reconstruction loss.(2)In order to guarantee the real-time detection performance of B-Detection in edge computing scenarios,this paper proposes a performance enhancing algorithm.Its implementation mechanism is: test the detection effect of the anomaly detection model according to the verification set,set rewards and penalties for the samples in the original Sample space,carry out resampling and model retraining,and constantly improve the real-time performance of the anomaly detection model iteratively.In addition,after detecting anomalies,this article will further analyze the current state of edge services and propose a computation offloading strategy that comprehensively considers resource utilization and geographic distance based on comprehensive analysis to ensure the normal operation of edge services after anomalies occur.(3)This article conducts extensive experiments on the real dataset service Survey to evaluate the detection performance of the B-Detection method,and compares it with six other different anomaly detection methods.The proposed computation offloading strategy is compared and analyzed with the other two methods.The sufficient comparative experimental results have verified the effectiveness of the proposed method in this paper.(4)Based on the above research,this article proposes management suggestions to improve the quality of urban brain system services.The work in this article has important practical application value for the operation,maintenance,and quality assurance of urban brain information systems.It has important reference significance for improving the modernization level of urban governance system and governance capacity.