Dynamic control of thermal hydrolysis to maximize net energy recovery from sewage sludge based on machine learning

Thermal hydrolysis pretreatment coupled with anaerobic digestion (THP-AD) substantially improves the energy recovery from sludge; however, its high thermal energy input often undermines overall system efficiency. This study developed a machine-learning-driven optimisation framework. The results indicated that, compared to the other three models, extreme gradient boosting achieved the highest predictive performance (R² > 0.90) for both methane yield and net energy output. Model interpretability revealed distinct composition-specific responses of sludge to THP, highlighting that sludge physicochemical properties and THP parameters were the primary drivers of system behaviour. A field-scale application using real operational data from a wastewater treatment plants, confirmed that an optimal adaptive THP temperature strategy improved the net energy output by up to 39 % compared with fixed-temperature operation. These findings redefine process control in sludge management, marking a fundamental shift from "one-size-fits-all thermal hydrolysis" to "on-demand thermal hydrolysis"-laying the groundwork for more intelligent, energy-resilient wastewater treatment systems.