Abstract: In order to solve the problem of low prediction accuracy of traditional modeling methods in establishing a prediction model for the main steam temperature (MST) of incinerators, a waste incinerator MST prediction method based on improved gray relevance analysis (IGRA) and convolutional-long short-term memory (CNN-LSTM) neural network was proposed. Firstly, DCS variables with high correlation degree with MST were selected by IGRA. Secondly, principal component analysis (PCA) was applied to extract principal component features containing most of the information of the incinerator combustion image. Then, based on IGRA and particle swarm optimization (PSO) algorithms, the delay between DCS parameters and the MST was estimated and compensated. Finally, a CNN-LSTM model with the input of time series matrix composed of DCS variables and image features was constructed to predict the change trend of MST in the next 6 minutes. Results show that compared with the existing MST prediction model, the proposed model reduces the mean absolute error M_AE by 13.07%, reduces the root mean square error R_MSE by 13.89%, and improves the determination coefficient R² by 13.08%.