Herein, we propose an algorithm to address the issue of communication resource limitations in vertical federated learning. The vertical federated learning algorithm is designed to simultaneously optimize transmission power, user selection, and channel estimation with a hybrid system combining visible light communication (VLC) and radio-frequency (RF) communication. The first step involves constructing a VLC/RF hybrid system by introducing a VLC link in a traditional RF link. Following this, we introduce a channel estimation algorithm based on multilayer perceptron to improve the accuracy of transmitted data. The final step involves establishing an optimization problem to minimize the longitudinal federated learning loss function. This problem is then solved by co-optimizing transmission power and user selection. The simulation results show that the accuracy of the proposed algorithm is improved by 7.2% and 18.2%, respectively, compared with the existing method.