Research On Detecting IoT Data Anomaly Based On Serverless Architecture

Anomaly detection of IoT data is one of the application scenarios where technologies such as IoT,big data and machine learning are integrated.Anomaly detection is important to improving security and stability of IoT system as well as the accuracy of the collected data.However,the traditional IoT cloud platform does not support the deployment of machine learning models,which puts shackles on the development and utilization of IoT big data.The current mainstream solution is to deploy algorithms on Iaa S cloud host to enable the anomaly detection server,but this approach has problems like “idle time billing”,high cost,poor scalability and versatility.Therefore,how to innovate the existing anomaly detection algorithms and achieve realtime,universal,accurate and low-cost anomaly detection in IoT scenarios is the main challenge of this paper.To solve the above problems,this paper innovatively proposes an IoT data anomaly detection system based on serverless architecture,with the following contributions1.Introduction of the deep embedding clustering into the field of anomaly detection in the unsupervised scenario with an unsupervised algorithm based on it,and improvement of the existing ADOA(Anomaly Detection with partially Observed Anomalies)anomaly detection algorithm in the semi-supervised scenario.The anomaly detection performance in both scenarios is improved.2.The innovative combination of serverless architecture and IoT scenario improves the pain point that traditional IoT cloud platforms are difficult to deploy machine learning models.The features of serverless architecture like event-driven,automatic scaling,and “pay-as-you-go” greatly reduce service costs while improving the performance of IoT systems.3.A generic,accurate,real-time and low-cost IoT data anomaly detection system is developed based on OneNET IoT cloud platform and Alibaba cloud function compute platform with a case study on a municipal landfill sensor dataset.To verify the superiority of the proposed algorithm and system,experiments are conducted on ODDS datasets and real IoT datasets,and both are deployed in Iaa S cloud host and serverless architecture for comparison.The results show that the proposed algorithm and system are more advantageous in terms of accuracy,response delay,throughput rate and service cost of anomaly detection...