Flue gas oxygen content series prediction model for municipal solid waste incineration process

Flue gas oxygen content (FGOC) is the core index for measuring combustion state and pollutant generation, and its accurate prediction is the prerequisite for optimizing municipal solid waste incineration (MSWI) process control. However, the MSWI has complex characteristics such as multivariable strong coupling, large lag, and non-linearity, making it difficult for traditional modeling methods to effectively capture its dynamic evolution. Therefore, this article proposes a series prediction model of multi-point FGOC for both the boiler outlet (BO) and chimney outlet (CO) location. Firstly, a systematic data preprocessing process is constructed, in which abnormal samples are cleaned using the interquartile range (IQR) algorithm. The data quality is further improved through Z-score standardization. Secondly, the mutual information (MI) algorithm is introduced to select features highly correlated with the FGOC at different locations, thereby effectively capturing non-linear dependencies. Finally, a series prediction framework based on Transformer-Long short-term memory (Transformer-LSTM) is designed, with serial prediction models constructed for the BO and CO. The prediction results of the previous stage are used as inputs for the latter stage, realizing high-precision terminal prediction under the "prediction-correction" mechanism. Evaluated on operational datasets collected from a practical MSWI power plant in Beijing, the proposed method shows partial improvement for the specific FGOC prediction scenario and provides a feasible reference for multi-point flue-gas-oxygen-content forecasting.