This paper proposes a fair federated learning method for handling client participation probability inconsistencies in heterogeneous environments