Research On Intelligent Analysis Technology Of Edge Computing Data Security

As an emerging business format and application model formed by the deep integration of a new generation of information technology and manufacturing,the Industrial Internet is the key infrastructure for realizing the connection of the entire industrial system,the entire industrial chain,and the entire value chain,as well as providing guarantees for the development of industrial intelligence.In the development process of the Industrial Internet,along with the tremendous progress and growth of computing power and data volume,cloud computing and big data technologies have gradually opened up the edge side,the cloud,the enterprise production and management systems,promoting the development of intelligent data analysis technology.intelligent data analysis requires analysis and research on a large amount of rich data.However,security issues such as user privacy and business secrets have become barriers to data cooperation and sharing among enterprises,and data security and data silos exist widely.The combination of edge computing and federated learning is a powerful means to solve the problems of data security and data silos.The edge computing platform centered on virtualization technology and resource management and orchestration technology provides computing and storage services for intelligent data analysis.On the premise of satisfying privacy protection and data security,federated learning promotes cooperation between all parties and improves the effect of intelligent data analysis.However,there are still privacy and security problems caused by gradient leakage attacks in federated learning.The combination of federated learning and reinforcement learning on the edge computing platform is an effective way to solve the problem of multi-party collaborative intelligent decision-making.In order to allow the model to make more accurate decisions,the federated reinforcement learning algorithm adopts an online training strategy,that is,the model is trained while deploying and making decisions.Therefore,the current federated reinforcement learning algorithm has the problem of slow decision-making.In response to the above problems,this paper designs a federated learning scheme for adaptive data generation,and conducts experimental verification and evaluation of this scheme through specific problems.The main work and progress are as follows:1)Research and propose an adaptive data generation federated learning scheme(AGFL).The scheme combines relevant content of federated learning,generative adversarial networks and meta-learning,and the scheme includes a federated training module(FPM)and a decisionmaking method.FPM includes four stages:data normalization,data generation,federated pre-training,and model fine-tuning.2)By applying the AGFL scheme to the spectral signal denoising problem,a federated denoising algorithm based on AGFL(AGFL-DAE)is proposed to verify the ability of the scheme to solve the privacy and security problems caused by gradient leakage attacks.In the experiment,the data privacy security of AGFL-DAE is evaluated,and the comparison with the experimental results based on Dropout technology and gradient compression technology respectively shows that:In contrast,the mean square error of AGFL-DAE is increased by about 2.64%when it is attacked by gradient leakage.The experimental results show that the AGFL scheme can effectively solve the privacy and security problems caused by gradient leakage attacks.3)By applying the AGFL scheme to the road network proximity detection delay optimization problem,a federated decision-making algorithm based on AGFL(AGFL-DDPG)is proposed to verify the ability and to solve the privacy security caused by gradient leakage attacks,and the ability of making decisions faster of federated reinforcement learning.In the experiment,the data privacy security and decision-making speed of AGFL-DDPG are evaluated.The comparison with the experimental results of the Model-Replacement algorithm shows that:In contrast,the mean square error of AGFL-DDPG recovered by gradient leakage attack increases by about 10.98 times.the decision-making latency is reduced by 90.72%.The experimental results show that the AGFL scheme can greatly improve the decision-making speed of federated reinforcement learning,and the scheme can effectively solve the privacy and security problems caused by gradient leakage attacks.