Abstract: Aiming at the problem of main steam temperature fluctuation faced by waste incineration boilers due to the lag of combustion control, a prediction modeling method of main steam temperature in the next 6 minutes based on distributed control system (DCS) data and combustion images is proposed. First, DCS data and combustion images are collected, and spectral norm features of images are extracted. Then, mutual information and conditional mutual information are used to select DCS feature variables highly correlated with main steam temperature and with low redundancy among variables. Then, the time delay of DCS feature variables and image features is estimated based on MI algorithm, the lagged features are eliminated and the delay compensation of advanced features is carried out. Finally, DCS feature variables and image features are used as input, and a prediction model of main steam temperature is established based on long short-term memory. The experiment shows that the root mean square error (RMSE) of the whole prediction is 1.4722, and the RMSE of the first 2 minutes is as low as 0.61. Time delay compensation and the addition of images effectively reduce the prediction error. The model prediction RMSE with time delay compensation is reduced by 22.61%, and the model prediction RMSE with image input is reduced by 11.79%. The model prediction effect is good, which can provide reference for production regulation.