Reconfigurable intelligent surface (RIS) breaks through the limitation of passive adaptation in traditional wireless environments and is one of the promising key technologies in 6G. The paper delves into the analysis of a device-to-device (D2D) communication network augmented by a RIS. In order to ensure link fairness, we first propose a joint optimization of RIS beamforming design and bandwidth resource allocation to maximize the minimum rate of all devices. Then, we introduce Dirichlet distribution into the proximal policy optimization (PPO) and propose a D-PPO algorithm. After that, we design a Markov decision process based on the D-PPO to address the above joint optimization problem. Finally, the simulation outcomes demonstrate that the efficacy of our proposed scheme is better than the baselines.