Research On Fog Computing Based Energy-Efficient Intelligent Computation Offloading

With the rapid popularization of 5G technology and the rapid development of the Internet of Things(Io T),various emerging applications such as face recognition,augmented/virtual reality,etc.have emerged,and these applications are usually computationally intensive and delay-sensitive,which puts forward high requirements on the computing power and battery capacity of the Io T devices.As an efficient and instant computing paradigm,fog computing can offload computation tasks of Io T devices to nearby fog nodes to provide them with additional computing resources,thereby it can reduce the computing burden of Io T devices and extend battery lifetime.Currently,there are large numbers of literatures on computation offloading in fog computing networks,but the joint consideration of fairness,energy efficiency,and reasonable allocation of resources in the computation offloading process is not full explored.In order to solve the above problems,this thesis studies the fog computing based energy-efficient intelligent computation offloading scheme.The main contributions include the following three points:1)Energy Minimization and Fair Computation Offloading Mechanism for Fog Computing Networks: In order to ensure the fairness of each fog node while reducing the total energy consumption of Io T devices,this thesis proposes an energy minimization and fair computation offloading mechanism for fog computing networks.Specifically,an optimization problem with the minimization of total energy consumption for all tasks is formulated,in which the optimization allocation of task offloading ratio,transmission power and fog node selection are jointly considered.According to such optimization problem,a candidate set generation algorithm of destination node for task offloading is developed,and the lowest energy consumption of each fog node under the corresponding delay constraint,the corresponding offloading ratio and transmission power are obtained through the bisection method.Furthermore,in order to make a tradeoff between the low energy consumption and fair selection of destination node,based on fair scheduling indicator,a fair selection algorithm of destination node is proposed to realize the allocation of computation tasks in a low-energy and fair manner.Finally,the simulation results show that this mechanism can ensure the fair selection of the fog node under the condition of low total energy consumption,and the survival rate of fog node is enhanced by 10.9% on average as compared with the maximum equivalent processing rate mechanism.2)Simultaneous Wireless Information and Power Transfer(SWIPT)based Rechargeable Fog Computation Offloading Mechanism: In order to further reduce the energy consumption of Io T devices and extend the battery lifetime,this thesis proposes a simultaneous SWIPT based rechargeable fog computation offloading mechanism.Specifically,an optimization problem that minimizes the total energy consumption for completing all tasks in a multi-user scenario is formulated,and the joint optimization of task offloading ratio,transmission time and power split ratio is fully considered.Based on above non-convex optimization problem,a difference-of-convex programming(DCP)and accelerated gradient based alternating optimization algorithm is proposed.This algorithm converts the solution of non-convex optimization problem into two convex optimization sub-problems by employing DCP and alternating optimization theory.At the same time,combined with the accelerated gradient descent method,the optimal solutions of task offloading ratio,transmission time and power split ratio can be achieved with a fast convergence speed.In particular,the integration of SWIPT technology further reduces the energy consumption of smart devices and extends their service lifetime.Finally,the simulation results indicate that the proposed algorithm has the characteristics of low energy consumption and fast convergence speed,and compared with other scheme,the energy consumption is reduced by an average of 12.8%.3)Deep Reinforcement Learning based Intelligent Rechargeable Fog Computation Offloading Mechanism: In order to satisfy the dynamic and self-adapatation requirements of complex fog computing networks and achieve intelligent optimization of computing resources,this thesis proposes a deep reinforcement learning based intelligent rechargeable fog computation offloading mechanism with SWIPT technology.Specifically,an optimization problem that minimizes the total energy consumption for completing all tasks in a multi-user scenario is formulated,and the joint optimization of task offloading ratio,uplink channel bandwidth,power split ratio,and computing resource allocation is fully considered.To solve the above-mentioned non-convex optimization problem of continuous action space,an intelligent computation offloading algorithm that integrates communication,computing and energy harvesting is proposed.The design of the inverting gradient update based dual actor-critic neural network structure can effectively improve the convergence rate and stability in the training process.Finally,the simulation results show that the proposed algorithm can converge faster and effectively reduce energy consumption under the delay constraint.