Research On Intelligent Computation Offloading Mechanism For Fog Computing

Since the commercial construction of 5G technology,the transformation process of information and communication technology industry has been gradually accelerated,giving rise to various emerging applications such as high-definition video slow live stream,immersive interactive experience and intelligent face recognition,which will pose great challenges to the computing and storage resources of terminal devices.And thus the fog computing mode with decentralized autonomous data centers is introduced to keep multi-user connection status over the same period,reduce device load pressure and alleviate information overload phenomenon.Furthermore,in view of the resource constraints of boundary fog nodes,related researches with computation offloading have been widely carried out,where the comprehensive consideration of device self-power capacity,resource allocation fairness and collaborative processing energy efficiency is not actually explored.In order to remedy the above defects and create better energy-saving service,this thesis studies the intelligent computation offloading scheme for fog computing,and its main contributions can be summarized as the following three aspects:1)Radio frequency energy harvesting-based collaborative energy-saving computation offloading mechanism: In order to fit the differentiated energy demands in vertical markets and ensure that Internet of Things(Io T)devices can hold an efficient and sustainable operation mode,this thesis studies a radio frequency energy harvesting-based collaborative energy-saving computation offloading mechanism.Specifically,a system energy consumption minimization problem is formulated under the joint optimization consideration of computation offloading decision,uplink bandwidth resource allocation,downlink bandwidth resource allocation and base station power splitting.Meanwhile,by combining the concept of penalty function,a new evaluation index is introduced,and then an adaptive particle swarm optimization-based collaborative energy saving computation offloading(APSO-CESCO)algorithm is proposed to solve such problem.The proposed algorithm constructs dynamic inertia weight and linearly adjusted penalty factor,which can alternate the spatial distribution density of particle community in real-time during the iterative search process,and the optimal computation offloading policy with tolerable punishment can be well-generated.Furthermore,to prevent particles from exceeding exploration range,the velocity boundary is introduced which can also reduce the generation probability of invalid solutions and improve the actual exploration effectiveness.Finally,the simulation results show that the proposed algorithm can achieve higher convergence efficiency and solution accuracy,and compared with other common benchmark schemes,the system energy consumption can be reduced by 34.09%,14.72% and 6.86%,respectively.2)User satisfaction-based energy-saving computation offloading mechanism: In order to improve the resource accordance in fog computing offloading scenario and establish better energy-aware service,this thesis studies a user satisfaction-based energy-saving computation offloading mechanism.Specially,through the joint optimization consideration of service decision,task offloading ratio,uplink bandwidth resource ratio and computing resource ratio,an energy consumption minimization problem is formulated on the premise that the user satisfaction should be preferentially fulfilled.That is,first,constructing a satisfaction function which can unite with the historical energy consumption distribution,so that the target users’ perceived response to energy-saving service can be well reflected,and then,a user satisfaction-based service decision(US-SD)algorithm is developed to achieve the optimal service decision.Correspondingly,with the goal of minimizing the processing energy consumption,a subtask partition and resource allocation-based intelligent computation offloading(SPRA-ICO)algorithm is proposed.In such algorithm,there it is with an innovative actor-critic network structure design,and the noise is added to the output continuous action,by which the randomness of deterministic policy exploration can be well guaranteed.Meanwhile,the experience replay buffer mechanism and parameter soft update operation are comprehensively employed to reduce the mutual guidance of training samples and improve the function convergence performance.Finally,the simulation results show that the proposed mechanism can realize good convergence speed and user retention rate,and compared with other benchmark schemes,the total energy consumption can be reduced by 14.76% and 6.52%on average,respectively.3)Twin delayed deep deterministic policy gradient-based intelligent computation offloading mechanism: In view of the multi-users’ randomness distribution in the dynamic large-scale Internet of Things scenario,how to comprehensively formulate available resources of fog nodes in the area and achieve the computation service with low cost has become a great challenge.As a result,this thesis studies an efficient and intelligent computation offloading mechanism with resource allocation.Specifically,an optimization problem is formulated to minimize the total energy consumption of all tasks under the joint optimization of computation offloading decision,bandwidth resource and transmission power.Meanwhile,a twin delayed deep deterministic policy gradient-based intelligent computation offloading(TD3PG-ICO)algorithm is proposed to solve such optimization problem.By combining the concept of actor critic algorithm,the proposed algorithm designs two independent critic networks that can avoid the subjective prediction of a single critic network and better guide the policy network to generate the global optimal computation offloading policy.At the same time,this algorithm introduces a continuous variable discretization operation to select target offloading node with random probability,and the available resources of target node are dynamically allocated to improve the model decision-making effect.Finally,the simulation results show that this proposed algorithm has faster convergence speed and good robustness.It can always approach the greedy algorithm with lowest total energy consumption,and compared with full local and deep Q-learning network(DQN)based computation offloading scheme,the total energy consumption can be reduced by 15.53% and 6.41%,respectively.