Research On Edge-Cloud Collaborative Task Offloading And Resource Allocation In Industrial Application Scenarios

With the transformation and upgrading of traditional industries,an increasing number of devices are being connected to the industrial Internet of things(IIoT),resulting in a vast amount of industrial data.IIoT requires in-depth data mining and analysis to optimize production and operational decision-making.Edge-cloud collaboration technology combines the big data analysis capabilities of cloud computing with the rapid response capabilities of multi-access edge computing,providing differentiated services for industrial data processing.Proper scheduling of edge and cloud resources can effectively reduce data analysis latency,improve data analysis accuracy and reliability,alleviate bandwidth transmission pressure on core network nodes,and ensure the privacy of industrial data.However,in the process of edge-cloud collaborative resource scheduling,resource scheduling needs to be performed based on data task requirements,edge and cloud resource states,and optimization goals.This dissertation starts by thoroughly studying the task requirements from the perspective of task offloading and resource scheduling efficiency,and proposes a predictivebased task offloading solution.Building upon this foundation,in order to ensure that industrial resources can provide low-latency and highly reliable services for tasks,this dissertation further investigates the synergy of computing,caching and communication resources in the industrial domain.It presents a multi-hop computing offloading scheme and a multidimensional industrial edge-cloud resource coordination strategy.Finally,from the perspective of task offloading and resource scheduling security,this dissertation proposes a blockchain-based secure resource scheduling approach,progressively addressing the task offloading and resource scheduling techniques for the IIoT.The main contributions of this dissertation are listed as follows:Firstly,to address the problem of real-time task demand acquisition in IIoT,this dissertation investigates a prediction-based intelligent edge-cloud collaborative resource scheduling technique.An artificial neural networkbased prediction algorithm is proposed to forecast the computational resource requirements of tasks.Based on the prediction results,as well as real-time collected task data,maximum tolerable latency,and system resource status,the edge-cloud resource scheduling problem is formulated as a Markov Decision Process(MDP).Then,a parallel reinforcement learning algorithm based on collaborative clustering is proposed to solve the aforementioned edge-cloud collaborative resource scheduling problem.Simulation experiments validate that the proposed edge-cloud collaborative resource scheduling algorithm effectively improves the training efficiency of the resource scheduling model and significantly enhances task processing efficiency.Secondly,to address the issue of horizontal and vertical collaboration of edge-cloud resources in the industrial Internet,this dissertation investigates a microservices-oriented multi-hop multi-path edge-cloud collaborative resource scheduling technique.A joint modeling of computational and communication resources is proposed,considering multi-path data transmission schemes,and establishing a multi-hop multipath optimization problem with the objective of minimizing latency.Then,considering the dynamic nature of computational and communication resource capabilities,this dissertation reformulates the optimization problem as MDP and presents a knowledge-assisted reinforcement learning-based multi-hop multi-path resource scheduling approach.This algorithm leverages the resource scheduling strategy generated by the linkstate routing algorithm as prior knowledge and combines it with the interaction data of resource scheduling to train the reinforcement learning model,effectively improving the efficiency of reinforcement learning.Simulation results validate that the proposed method exhibits faster convergence speed and superior convergence performance compared to traditional approaches,leading to enhanced timeliness of the model and improved performance of the resource scheduling scheme.Thirdly,in response to the diversified resource requirements of various computational tasks in industrial application scenarios,including computing,caching,and communication resource capacities,this dissertation investigates a joint resource scheduling technique based on deep reinforcement learning.A joint modeling of the states of computing,caching,and communication resources,task requirements,and the popularity of task data flows is performed.The multi-dimensional resource joint scheduling problem is formulated as MDP with the objective of optimizing long-term rewards.Then,a deep reinforcement learning-based algorithm for joint scheduling of computing,caching,and communication resources is proposed.This algorithm takes into account the mutual influence between resources and the long-term nature of scheduling decisions,enabling adaptive adjustment of resource scheduling decisions in dynamic environments with changing resource states,thereby improving system rewards.Simulation results validate that the joint resource scheduling method outperforms traditional approaches that consider individual resource types in terms of optimizing rewards.Finally,addressing the security concerns in the process of industrial edge-cloud collaborative resource scheduling,this dissertation investigates a resource scheduling technique based on blockchain and distributed reinforcement learning.The problem of edge-cloud collaborative resource scheduling in industrial Internet of Things is formulated as MDP.An approach based on parallel reinforcement learning is proposed to allocate resources and determine the optimal number of parallel servers.Then,a blockchain-based training and deployment scheme for parallel reinforcement learning is proposed.To enhance the efficiency of the blockchain consensus,this dissertation improves traditional consensus algorithms and presents a picture delegated proof of state+suspicious practical Byzantine fault-tolerant consensus algorithm.Simulation results demonstrate that the proposed approach effectively improves the efficiency of the resource scheduling process while ensuring the security of the resource scheduling process.