Suppression strategies for pyrolysis gas parameter fluctuations in small pyrolysis-combustion coupled reactors for MSW

• The MI method selects the main parameters that affect the key exported parameters. • A new MI-CNN-LSTM pyrolysis gas parameter prediction model has been constructed. • MI-CNN-LSTM exhibits stability and robustness across different datasets. • Fuzzy RBF for suppressing fluctuations in key export parameters. The small thermal capacity of a Small Pyrolysis-Combustion Coupled Reactor (PCCR) may lead to unstable operation and fluctuations in the parameters of the pyrolysis gas, making dynamic monitoring very difficult and affecting the stability and emissions of the secondary combustion of pyrolysis gas. Analyzing pyrolysis gas parameter variability's impact on secondary combustion stability and developing adaptive compensation strategies are critical for optimizing incineration efficiency and emission control. This study proposes a soft measurement method based on a CNN-LSTM to predict the temperature, calorific value, and flow rate of the outlet pyrolysis gas. MI based feature screening identifies the key variables influencing these parameters, and a CNN-LSTM prediction model is subsequently trained. The deviation between the predicted and target values is input into the FRBF model to dynamically adjust the primary air flow rate and temperature. This dynamic adjustment effectively suppresses the fluctuations in the outlet pyrolysis gas parameters. Research findings indicate that intelligent compensation outperforms manual compensation in precisely adjusting the primary air flow rate and temperature, reducing their standard deviations by 7.33 Nm³/h (44.80%) and 1.17 °C (29.47%), respectively. Additionally, the standard deviations of the outlet pyrolysis gas temperature, calorific value, and flow rate are reduced by 1.42 °C (9.58%), 11.05 MJ/kg (14.36%), and 9.99 Nm³/h (24.30%), respectively. For large incinerators and industrial scenarios (e.g., furnace control, reactor opt.), training/test partitioning ensures generalization, hyperparameter tuning optimizes config, and progressive data training adapts models. Optimized model enables real-time monitoring with fuzzy rules modified to suppress pyrolysis gas fluctuations.