Boiler combustion optimization plays an important role in the economic and stable operation of power plant boilers, and NO_x emission prediction is one of the basic links, so a method based on the improved dung beetle optimization algorithm to optimize convolutional neural network (CNN) and long short term memory (LSTM) combined model hyperparameters for super supercritical boiler NO_x emission prediction was proposed. First, the feature parameters related to NO_x emission were determined by Pearson correlation. Second, the CNN-LSTM prediction model was established, the hierarchical data structure was extracted by using the convolutional neural network CNN, and the long short-term memory network mined the long-term dependency, then the dung beetle algorithm was improved by combining the good point set and t-distribution variance strategy, and the LSTM hyperparameters were optimized by the improved algorithm to obtain the final prediction model. Finally, it was validated in comparison with other neural network models. The prediction is carried out with the actual data of a 660 MW unit boiler deep peaking, and the average absolute error between the actual value and the predicted value of NO_x emission concentration is 3.351 6, and the average relative error is 2.466 7, which shows that the prediction model has a more accurate prediction effect.