The coal gasification process is characterized by strong nonlinearity, strong coupling, and multiple conflicting objectives, which make it difficult for traditional mechanism-based optimization methods to achieve efficient and robust performance. As a widely applied representative gasifier type in China, the Jinhua gasifier urgently requires intelligent modeling and decision-support techniques for operational optimization. This paper proposes an operation optimization approach for the Jinhua gasifier based on a dual-channel convolutional neural network-long short-term memory-attention (CNN-LSTM-Attention) predictive model. The proposed model employs a dual-channel structure to integrate process features with historical sequence information and leverages a hierarchical attention mechanism to enhance the representation of critical features. In the tasks of predicting hydrogen and carbon monoxide fractions, the constructed dual-channel CNN-LSTM-Attention model achieves coefficients of determination of 0.9322 and 0.9637, respectively, demonstrating high accuracy and robustness. Building upon this, the model is coupled with a particle swarm optimization algorithm and used as a surrogate model to intelligently search for optimal operating variables. Experimental results show that, compared with the original operating conditions, the optimized scheme increases the hydrogen fraction by 1.22%, the carbon monoxide fraction by 1.51%, and the overall effective gas content by 1.38%. This study provides effective support for intelligent modeling and operating condition optimization of the Jinhua gasifier and offers valuable insights for the efficient operation of representative coal gasification technologies.