With the advancement in electronic attack technologies, intelligent jamming poses a significant challenge to the reliable transmission of wireless communications. Traditional anti-jamming methods often fail to adapt to dynamic nonlinear jamming environments. This paper addresses nonlinear swept-frequency jamming by modeling anti-jamming communication as a sequential decision-making problem and proposes an intelligent anti-jamming method based on proximal policy optimization (PPO) to optimize dynamic channel selection. Firstly, the channel selection problem is formalized as a Markov decision process (MDP), where a state space integrating jamming patterns and communication status is designed, the channel set is defined as the action space, and a multi-objective reward function trades off jamming avoidance against switching overhead. A dual-network architecture comprising a policy network and a value network is constructed, and the PPO algorithm is employed for policy updates, where a clipping mechanism is used to enhance training stability. The system optimizes the anti-jamming strategy online through a closed-loop process of “sensing–decision–learning–communication”. Simulation results demonstrate that compared to conventional methods, the proposed method significantly improves key performance indicators such as packet success rate and throughput. It can rapidly track changes in jamming, exhibiting excellent real-time performance and environmental robustness, and thus provides an effective solution for reliable communication in dynamic jamming environments.