Application And Research Of Data Collaborative Computing Between Devices And Devices In Edge Computing Environment

With the enhancement of the intelligence of the edge user equipment,more and more delay-sensitive services are migrating to the edge.The data generated by the user is processed at the edge,which can effectively reduce the delay and enhance data security.In the era of big data,the legal protection of private data is becoming more and more perfect.It is of great significance to effectively mine the value of data and apply it on the premise of protecting private data.This research combines the federated deep reinforcement learning with the recommendation system,and proposes to apply a personalized recommendation algorithm based on federated deep reinforcement learning in the edge computing environment of the user device.The algorithm can continuously improve its recommendation agent during the interaction with the user.The main work of this research includes:1)Study the interaction mode between the current information flow recommendation system and the user,compare the interaction mode between the agent and the environment in reinforcement learning,use Markov decision process to model the interaction process between the recommendation system and the user,and make interactive recommendations The system environment is organically combined with the user-side edge computing environment.2)According to the characteristics of the edge-end recommendation environment and the established Markov model,combined with reinforcement learning to design a feedback-based personalized recommendation algorithm at the edge-end,the reinforcement learning agent recommends items to users based on user interaction data.3)Introduce federated learning for training acceleration and privacy protection of reinforcement learning models,reduce latency through edge caching,and ease cloud aggregation pressure through edge computing nodes with model aggregation functions.The experimental results show the effectiveness of the edge-end federated reinforcement learning recommendation algorithm,which provides a direction for follow-up research.There are 17 pictures,6 tables,and 73 references.