In order to overcome the unsatisfactory training results caused by differences in data distribution between platforms,a momentum-accelerated federated learning algorithm with client drift control was proposed.The improved algorithm was verified by the collected waybill data.Experiment results show that under different data distributions,the improved federated learning algorithm is improved in performance compared with the traditional federal average algorithm(FedAvg),among which the convergence speed is increased by up to 36％,and the F1 score is increased by up to 5.7％.
维普
万方数据