To address quality issues caused by reliance on operator experience and inaccurate parameter settings during the cooling process of hot‐rolled thick plates, this paper proposes an intelligent control system based on Twin Delayed Deep Deterministic Policy Gradient (TD3) and dual residual neural networks integrated with a convolutional block attention module. The system uses postcooling temperature and plate‐shape data, together with basic steel plate information, as state inputs, and outputs the optimal cooling process parameters for the next steel plate, thereby achieving closed‐loop control. The residual network structure strengthens feature extraction, while the attention mechanism enhances sensitivity to key spatial features, thereby improving the perception of cooling effects by the model. In addition, a quantitative quality evaluation system is established based on plate flatness and temperature uniformity, which guides policy optimization through a reward function. Model training results demonstrate that the proposed model reduces the standard deviation of steel plate shape to 3.41 mm and temperature to 1.73°C, improving control accuracy and product consistency, with strong predictive performance in practice. This approach reduces dependence on manual tuning, enhances process stability, and provides strong technical support for intelligent manufacturing and high‐precision quality control of hot‐rolled thick plates.
Zhang et al. (Mon,) studied this question.