Research On Task Offloading In Industrial Internet Of Things Based On Privacy Preservation And Queuing Theory

Opportunistic computation offloading is an effective way to improve the computational performance of(Industrial Internet of Things)IIo T devices.However,as more and more computational tasks are offloaded to mobile edge computing servers for processing,this inevitably leads to privacy and security issues for users,such as personal usage habits.A listener can listen to a user’s offloading habits to infer the user’s location and the user’s privacy patterns.Nowadays,most studies have been conducted on the basis of traditional cloud computing scenarios in terms of data encryption,access control,and authentication,and fewer studies have been conducted on the privacy protection of IIo T with MEC architecture.In addition to user privacy,it is equally important to ensure stable system operation(e.g.,data queue stability and stable server operation).However,most existing DRL-based approaches do not impose long-term performance constraints.Instead,they use heuristics that discourage unfavorable behavior in each time frame by introducing penalty clauses related to,for example,packet loss events and energy consumption.This paper focuses on the study of task offloading in mobile edge computing considering privacy protection and stable system operation.First,it is used to solve the privacy leakage problem caused by users performing computation offloading in mobile edge computing.The basic idea is that in order to reduce the probability of being listened to by eavesdroppers,and we introduce the concept of privacy metric and set a security threshold for each IIo T user,which increases the probability of secure transmission of IIo T user in MEC scenarios.Second,it is used to solve the system instability problem caused by the computation offloading process.The basic ideas are: 1)To maintain the stability of IIo T data queues and minimize the energy consumption,we introduce a Lyapunov optimization model and define the total queue backlog,where virtual energy queues and data queues are introduced.We further define the Lyapunov drift plus penalty for the total queue backlog.According to the upper bound of the drift-plus-penalty expression on each time frame,the IIo T data queue can be maintained stable while minimizing energy consumption.2)To avoid overloading the edge servers,we propose a joint constrained optimization model for latency and energy consumption based on queuing theory.First,we aim to design a Lyapunov-based privacy-aware framework that defines the amount of privacy of IIo T users and designs a "privacy reduction" mechanism.On the one hand,we define the cumulative privacy amount of each IIo T user and trigger the privacy protection mechanism when the cumulative privacy amount exceeds a set privacy threshold.Then,the offloaded data generated by IIo T users are transferred to local processing,and finally,the cumulative privacy amount of IIo T users is reduced.This model ensures that the cumulative privacy volume of all IIo T users remains stable.On the other hand,we use an order-preserving quantization approach,which balances the drawbacks of the actor-critic network employed in our work.At the same time,it allows the model to converge rapidly during the training process.We further combine the advantages of Lyapunov optimization and actor-critic networks to address the problem of how to make the model learn the optimal strategy and maintain the minimum energy consumption in the long run.In particular,this framework integrates model-based optimization and model-free deep reinforcement learning to handle offloading with very low computational complexity,and Lyapunov optimization ensures that the framework achieves minimal energy consumption while stabilizing the data queue.It is demonstrated through experimental simulation results that the proposed scheme can maintain data queue stability and minimize energy consumption under strict security.Second,we propose a joint constrained optimization model for delay and energy consumption based on queuing theory,this model can effectively solve the task offloading problem in IIo T.Subsequently,we improve the structure of Multi-Agent Approximate Policy Optimization(MAPPO)algorithm to form a lightweight optimal task offloading algorithm,namely Multi-Agent Deep Reinforcement Learning(MAQDRL)based on queuing theory,which is more suitable for IIo T.Then,we obtain dynamic and stochastic multi-user offloading by using Multi-Agent Deep Reinforcement Learning(MADRL)to obtain the optimal offloading policy in dynamic and stochastic multi-user offloading environments.We also improve the neural network structure of MADRL by analyzing the structural features of the input data.As a result,our proposed algorithm shows good convergence and excellent performance in terms of task arrival rate,bandwidth,energy consumption,latency,and other metrics.The simulation results indicate that compared with other classical algorithms,MAQDRL is effective for solving the EIIo T offloading problem.