A key challenge in federated learning is statistical heterogeneity, which affects the generalization ability of the global model on each client. To address this issue, we propose a method called Partial Parameter Enhancement and Local Adaptive Aggregation for Personalized Federated Learning (FL-PPELA). The key component of FL-PPELA is the adaptive local aggregation module PPELA. In the PPELA, partial parameters in the global model are enhanced in the direction beneficial to the local client. Next, adaptive aggregation of the local model and the global model before local training. Finally, build different models for each client’s local dataset. To evaluate the effectiveness of FL-PPELA, we conducted experiments on four real datasets. FL-PPELA demonstrated superior test accuracy compared to 12 advanced baseline methods.