With the rapid expansion of offshore oil and gas development in China, offshore platforms are increasingly exposed to safety issues caused by harsh operating environments and the progressive manifestation of inherent equipment defects. How to select appropriate risk control measures based on the normalized risk values of key components in critical equipment, so as to achieve a balance between safety and economic benefit, has become a major concern in risk control strategy design. To address this issue, a deep reinforcement learning-based decision-making method for risk control measures in subsea tree system on offshore oil and gas platforms is proposed. Taking the subsea tree system as an example, the increase in risk value induced by external uncertain factors is quantified using a Poisson process. With production revenue and the costs of risk control measures incorporated into the reward function, a deep reinforcement learning-based decision-making framework is established. By continuously updating the neural network parameters through interactions with the environment, the method learns risk control decisions with favorable revenue per unit time. The proposed method is compared with threshold-based risk control strategies, and the effects of different failure penalties as well as preparation costs for preventive maintenance and replacement operations on revenue per unit time are further investigated. The results show that the proposed method achieves performance close to that of the best threshold-based strategy in terms of average revenue per unit time, while exhibiting a smaller reduction in revenue under varying failure penalty conditions, indicating a certain degree of flexibility and robustness. This study is positioned as a simulation-based validation of a risk control method and is currently intended for upper-level offline decision support rather than online automatic closed-loop control in field production systems.