计算机科学
分布式计算
云计算
聚类分析
GSM演进的增强数据速率
建筑
块链
数据聚合器
数据挖掘
资源(消歧)
边缘计算
计算机网络
无线传感器网络
机器学习
人工智能
计算机安全
操作系统
艺术
视觉艺术
作者
Huifeng Yang,Liandong Chen,Kai Cheng,Lei Zhang,Peipei Shen,Jiewei Chen,Shaoyong Guo,Qichen Li
出处
期刊:Lecture notes in electrical engineering
日期:2022-01-01
卷期号:: 188-197
标识
DOI:10.1007/978-981-19-6901-0_21
摘要
AbstractPower system contains a lot of user data. When using distributed machine learning for joint modeling, there is a risk of data privacy leakage. These problems mainly show that data is exposed in the public environment, leading to attackers using big data analysis and other means to mine important information they are interested in. At the same time, complex edge wireless environment, such as unreliable wireless environment and limited network resources, leads to a long process of federated learning and training and low resource utilization. To solve the problems of data privacy, limited power network resources and unclear ownership of power data, we adopt a hierarchical federated learning architecture based on blockchain. In this architecture, we propose a balanced clustering approach that distributes edge nodes into different clusters. Edge nodes in the same cluster send their local update information to the cluster’s leader, which is then aggregated by each cluster’s leader and sent to the cloud server. Then the global aggregation is implemented on the cloud server. We propose an algorithm to determine the optimal tradeoff between local update and global aggregation to minimize functional loss with limited resource budgets. Finally, the corresponding data is verified and tracked on the blockchain. The performance of the algorithm is evaluated by a large number of experiments on real data sets, and studied on different models and data sets, and a large number of simulation results are obtained.KeywordsFederated learningBlockchainResource constraintsData sharing
科研通智能强力驱动
Strongly Powered by AbleSci AI