Research On IOT Data Access Technology Based On Edge Computing

With the development of Internet of Things(Io T)technology,Io T system will face a lot of data to the numerous smart terminal devices.These devices are not just smartphones or laptops,but also intelligent connected vehicles,vending machines,smart wearables,surgical medical robots,and more.In the era of centralized big data processing with cloud computing model as the core,the vast amounts of data generated by countless types of such devices needs to be pushed to a centralized cloud for retention(data management),analysis and decision making.The analyzed data results are then passed back to the devices if needed.This round-trip of data consumes a large amount of network infrastructure and cloud infrastructure resources,further increasing latency and bandwidth load issues,thereby affecting critical missions of Io T system.In this case,edge computing technology is rapidly integrated into large-scale Io T monitoring systems.In the Io T system,the edge computing layer is closer to the Io T terminal devices,therefore,it can not only provide local data collection and improve the real-time performance of the service,but also reduce the backhaul delay between the terminal devices and the cloud platform.However,edge nodes have very limited computing,storage,and network resources,which make it difficult to process largescale data.In this paper,a method of data collection using deep learning technology combining cloud and edge nodes is proposed.Using advantage of computing power of the cloud and real-time superiority of the edge node,the neural network model can be trained in cloud and deployed to the edge node and predict sensor data at the edge node.This method can not only increase the speed of computing services and satisfy the realtime requirements of data collection,but also reduces network traffic and eliminates bandwidth load effectively.The main work of this paper includes:Firstly,by analyzing the requirement of traditional IOT system,this paper describes the overall architecture of IOT system based on cloud and edge computing.The system architecture is divided into three parts: the Io T Device Layer,the Edge Layer,and the Cloud Layer.In the Device Layer,the sensor device nodes need to receive sensor data from current environment and push sensor data to the edge layer;the Edge Layer is a real-time data storage and processing center,which is responsible for receiving real-time data transmitted from the Io T Device Layer.It is also responsible for calculating the accuracy value between the predicted data and the real data,and determining whether to upload real data to the cloud based on the accuracy value.The Cloud Layer is responsible for interacting with the Edge Layer and performing the historical data storage,training,prediction and deployment of deep learning models.The method proposed in this paper is used to solve the problem of high load and high delay in the process of data access and transmission in the Internet of things.Secondly,through the analysis and research on the data access technology of IOT,combined with the actual application requirements of sensor equipment and edge nodes,a method of data collection using deep learning technology combining cloud and edge nodes is proposed.Specifically,it can be divided into the part of model training that using Bi-directional LSTM(Bi-LSTM)and deployment in the Cloud Layer and the part of data matching and prediction comparison in the Edge Layer.Thirdly,the Io T sensor monitoring and management system is designed and implemented.Raspberry Pi is used as the edge node,and Raspberry Pi and laboratory servers are used to realize the above architecture.By assembling the sensor,we can get the environmental attributes such as room temperature and humidity in real time;the system docks with the Things Board console,which can directly monitor the real-time data collected by the cloud platform.Finally,experimental verification is carried out.In Experiment 1,the prediction accuracy of the recurrent neural network is evaluated and compared.The results show that the bidirectional LSTM used in this paper is better than the traditional LSTM,GRU and other neural networks in the prediction of time series data.Then,the influence of different batch sizes on the training model under the same parameters is compared in the training process.The results show that when the batch size is 64,the final loss value is small and there is no over fitting.Therefore,the batch size of the Bi-LSTM model trained in this paper is 64;the second experiment is to compare the network bandwidth load between the proposed three layers Internet of things architecture and the traditional sensor-cloud architecture,which using the same server system to run the test experiment.The sensor data is divided into six groups,and the nload tool is used to monitor the bandwidth load in the data collection process.The results show that transmitting sensor data directly to the cloud(sensor-cloud mode)has the largest bandwidth load,while deploying the prediction model to the edge node,using the edge layer and cloud layer for data comparison and prediction can further reduce the amount of data packet collection and achieve smaller network load.Experiment 3 evaluates and measures the end-to-end delay time between layers,the results show that the delay time of data transmission through sensor-cloud mode is longer than that through sensor-edge-cloud mode when the network and equipment are in the same situation.