A Federated Learning Algorithm for Bidirectional Selection Between Vehicles and Parameter Servers
The federated learning framework is gradually being widely applied in the internet of vehicles.In response to the mobility characteristics of vehicles,as well as the communication congestion problem caused by the simultaneous access of parameter server interaction parameters during federated learning of a large number of vehicles,a fuzzy logic based vehicle selection algorithm and a parameter server selection algorithm based on evolutionary game are proposed.By designing a fuzzy logic algorithm and considering the mobility of vehicles,equipment conditions,and data size factors,vehicles are selected with stable communication connections,high computing power,and large data size to participate in federated learning.Evolutionary game is used to further characterize the process of selecting parameter servers for autonomous vehicle decision-making.The accuracy of federated learning models and the resulting communication and computational costs is balanced.Thereby communication congestion is avoided and individual and overall profit is maximized.Finally,simulation result verifies the performance of the proposed algorithm in a large number of vehicle scenarios,achieving low-cost and high-precision model training.
internet of vehiclesfederated learningevolutionary gamefuzzy logicreplicator dynamic
万方数据
维普
