Research On Resource Allocation Algorithm For High Energy Efficiency Federated Learning In Iot Scenario

With the advancement of information and communication technologies and artificial intelligence,the Internet of Things industry has developed rapidly,and various intelligent applications have emerged.Federated learning,as a distributed machine learning technology,uses parameters to replace the transmission of original data for model updates,thereby avoiding privacy leaks and reducing communication costs,and has attracted widespread attention.However,in the IoT scenario,there is a lack of trust between task initiators and computing devices,and the potential of data is difficult to explore,which limits the performance of federated learning.Integrating blockchain with decentralized,tamper-proof,and traceable characteristics into federated learning can achieve distributed and trustworthy model construction.Meanwhile,the computing resources of industrial equipment exhibit significant dynamic and heterogeneity in industrial IoT,and the construction and deployment cycles of decision models are limited.Digital twin accelerates the convergence of decision models through real-time physical state sensing,mapping,computation,and interaction.However,there are various limitations on computing resources,communication resources and energy resources.Therefore,there are key challenges in integrating blockchain and digital twin to improve federated learning performance.Effective and reasonable allocation of multidimensional resources is crucial for achieving highenergy-efficiency federated learning as federated learning performance improvement and system energy consumption are mutually constrained.This paper addresses the problem of ubiquitous node trustworthy collaboration and real-time model parameter interaction in the federated learning training process in the IoT scenario.Combining blockchain and digital twin technologies,we propose a blockchain-based trustworthy federated learning framework and an efficient federated leaning model construction framework empowered by digital twin.Further,considering mutual constraints between federated learning performance improvement and system energy consumption,we study a joint resource allocation algorithm based on deep reinforcement learning,which is of great significance for achieving high-energy-efficiency federated learning.Firstly,we propose a blockchain-based trustworthy federated learning framework and design a local model aggregation scheme based on the committee mechanism,which effectively speeds up model convergence and improves model accuracy.Furthermore,considering the challenges posed by multidimensional resource limitations for the joint deployment,we propose a resource allocation strategy that combines spectrum resource allocation,block size adjustment,and primary node selection to achieve high-energy-efficiency federated learning.The simulation results show that compared with traditional federated learning approaches,the proposed framework improves model accuracy and accelerates model convergence.Moreover,compared with traditional resource allocation algorithms,the proposed algorithm significantly reduces system energy consumption while ensuring the performance of federated learning and blockchain.Secondly,we propose a digital twin-empowered federated learning framework for industrial IoT.An adaptive federated learning strategy is designed,which can minimize system energy consumption through joint optimization of training way selection and spectrum allocation.The simulation results show that compared with the traditional federated learning,the proposed algorithm can adaptively adjust training strategies and spectrum allocation to reduce energy consumption.This provides theoretical guidance for the design of resource allocation algorithms for digital twin-empowered federated learning in industrial IoT.