A privacy enhanced privacy protection method for longitudinal and transverse federated learning (PEVHFL) is proposed to address the issue of improving privacy protection performance. PEVHFL achieves knowledge transfer through weighted aggregation of horizontal model parameters and transmission of vertical embedding layer information. Then, a vertical and horizontal double differential privacy mechanism was designed to improve data privacy protection performance. In addition, the SAM optimizer is used to flatten the convergence domain of the convergence algorithm and improve its convergence accuracy.