Abstract: Under the “dual carbon” strategy background, integrating technologies such as artificial intelligence, big data and cloud computing to achieve the intelligence, greenness and low-carbonization of municipal solid waste incineration power generation presents a significant challenge. In this study, the cross-correlation method was employed to analyze the correlations among 110 variables in a waste incinerator, and the variables with high correlations with three predicted variables, namely main steam flow, furnace average temperature and average oxygen content, were obtained. Based on this correlation analysis, the time delay steps between these three predicted variables and the equipment control parameters were calculated. Meanwhile, four time-series forecasting models were utilized to predict the main steam flow, furnace average temperature and average oxygen content. The results demonstrated that PatchTST had the lowest prediction error on the test set (the NMAE of main steam flow, furnace average temperature and average oxygen content was 0.072 9, 0.056 0 and 0.140 6， respectively), which exhibited superior generalization and predictive capabilities. It could better eliminate the time lag of data and reduce the error caused by the subjective judgment ability of operators, providing data support for real-time online operations.

Key words: grate-type waste incinerator, artificial intelligence, deep learning, neural network, time series prediction