Research On Task Reliable Offloading Method For Fiber-Wireless Heterogeneous Access Networks

With the rapid development of 5G and the Internet of Things(IoT),large-scale device connectivity and data transmission have posed challenges to the coverage and communication bandwidth of the IoT.Fiber-Wireless Heterogeneous Access Networks(FiWi HetNets)are considered one of the solutions that can achieve large-scale,low-cost deployment of the IoT due to their ability to simultaneously utilize the advantages of wired optical fiber networks in low latency,high reliability,and large capacity,as well as the advantages of wireless networks in flexible coverage.However,traditional FiWi HetNets,which adopt a treelike deployment structure and centralized task processing methods,have led to significant communication and computational overheads.Mobile Edge Computing(MEC)technology,which can offload caching and computing capabilities to the edge side closer to users,has been introduced to improve the service capability of edge networks by relieving the pressure on the core network through task offloading to the edge network.FiWi HetNets integrated with MEC have gradually become a hot research topic in the industry.Task computing offloading for FiWi HetNets mainly includes two stages:task scheduling and task processing.In the task scheduling stage,a network access strategy and task scheduling strategy need to be developed based on task service requirements and network connection status.However,most existing methods ignore the task interruption caused by wireless transmission risks and the low computational efficiency caused by centralized policy solving.To address the above problems in task scheduling,a method based on Federated Reinforcement Learning(FRL)for rapid and reliable scheduling of FiWi Hetnets tasks is proposed.First,a reliable offloading framework for FiWi heterogeneous networks with multiple wireless access modes is constructed to meet the capacity and coverage requirements of the IoT.Then,a risk measure model is introduced by introducing conditional risk values to measure wireless transmission risks caused by error transmission and collisions.Based on the risk measure model,an adaptive access and scheduling decision algorithm based on FRL is proposed.A double-layer federated learning model selection and aggregation scheme based on cumulative reputation value is designed to achieve distributed and reliable training of the model.A deep reinforcement learning model is deployed at the end to solve adaptive wireless access,task scheduling,and resource allocation decisions in complex scenarios,enabling rapid scheduling decision-making and improving the reliability of task transmission.Simulation results show that the proposed method has better comprehensive performance in energy consumption and service delay,which helps to improve the success rate of task processing.In the task processing stage,the service capabilities of edge nodes need to be considered to develop reasonable resource allocation and task offloading strategies.However,existing research has not considered the task processing failure caused by link and node degradation,and has ignored the supporting role of service software in complex task processing,making it difficult to provide reliable services for computationally intensive IoT tasks such as intelligent inspection and robot navigation.To solve these problems,a sequence subtask cloud-edge collaboration offloading method based on latency and reliability is proposed.First,the task is divided into multiple subtasks according to functionality,and a sequence subtask description model is constructed.Then,a resource-aware reliability enhancement model is established,and a multi-mode risk subtask cloud-edge backup offloading mechanism is designed to improve service reliability by providing dedicated backup and joint backup for key subtasks with high interruption probability or maximum reliability improvement potential.Finally,a subtask sequence partition algorithm based on latency and reliability is designed to accurately divide and allocate the subtask sequence of the task in the cloud-edge collaboration network,further optimizing the task processing latency and reliability.Simulation results show that the proposed method can ensure reliable task processing while reducing the energy consumption of edge nodes and improving the service delay of tasks.