Abstract:
Temperature Swing Adsorption (TSA) is a widely used cyclic adsorption process that leverages the temperature sensitivity of adsorbent materials to selectively capture and release target gas species. In TSA systems, contaminants such as water vapor or CO2 are adsorbed at low temperatures and then desorbed during a high-temperature regeneration step, enabling repeated cycles of purification. Compared to Pressure Swing Adsorption (PSA), TSA is particularly well-suited for removing strongly adsorbed species in low-concentration streams or in applications requiring deep drying.
TSA is extensively applied in natural gas processing, where water must be removed to prevent hydrate formation and corrosion during subsequent low-temperature steps such as liquefied natural gas (LNG) production [1]. It is also used as a pretreatment step in the sulphonation process for manufacturing surfactants, where dry air is required to avoid sulfuric acid formation that would otherwise degrade product quality and corrode downstream equipment.
Despite being highly energy-intensive, TSA systems are often operated inefficiently due to reliance on conservative heuristics or rule-based logic, which fail to exploit the full potential of process dynamics and energy-saving opportunities [2]. In this work, we develop and apply a hybrid economic model predictive control (EMPC) framework for the real-time optimization of a twin-bed industrial silica gel drying system.
The drying and regeneration operations are modeled using first-principles mass and energy balances, including water adsorption and desorption kinetics. These mechanistic models are augmented with a gated recurrent unit (GRU) neural network to form a hybrid model that balances physical interpretability with data-driven flexibility [3].
Real-time estimation of internal process states is achieved using moving horizon estimators (MHE) for both drying and regeneration cycles, enabling inference of key unmeasured variables such as internal moisture content. The estimated states are used to predict the end time of the ongoing regeneration cycle via open-loop simulation. This prediction is provided to the EMPC, which minimizes energy consumption and maximizes throughput in the drying cycle, subject to strict moisture constraints for downstream processing. The EMPC uses a variable horizon formulation, continuously adapting horizon length based on the predicted end time to handle the long-time scales of TSA operation.
Simulations of the twin-bed system show that the proposed hybrid EMPC strategy achieves up to a 22% reduction in energy usage compared to baseline control with fixed cycle durations, while maintaining product quality and operational safety. The hybrid model enhances predictive accuracy, and the control strategy demonstrates robustness to disturbances and model uncertainty.
References
Berg, F., Pasel, C., Eckardt, T., & Bathen, D. (2019). Temperature swing adsorption in natural gas processing: A concise overview. ChemBioEng Reviews, 6(2), 71–83.
Sadighi, S., Asgari, M., Mohammadi, H., & Noorbakhsh, F. (2016). Increasing the efficiency of a temperature swing adsorption (TSA) natural gas dehydration plant. Petroleum and Coal, 58(3), 317–324.
Chai, S., Kong, X., & Mercangöz, M. (2025). State estimation for industrial desiccant air dryers using hybrid mechanistic and machine learning models. Computers in Industry, 168, 104274.