AGI-Fedavg: A UAV Power Line Inspection Algorithm Based on Federated Learning

As the most important power grid equipment, transmission line data is important privacy of each power company. However, due to the differences among power regions, the disclosure of data has become a more difficult problem. Meanwhile, the data available among power regions are limited, which is far from enough to train an accurate machine learning model. To solve these problems, this paper proposes a new federated learning algorithm AGI-fedavg, which is based on the classical federated learning algorithm. For the non-IID algorithm, the node eigenvalues are introduced on the basis of the Fedavg algorithm. Experimental results on public data set show that our proposed algorithm effectively solves the problems of reduced training accuracy of the Fed-avg algorithm on the non-IID problem.