Gaussian process surrogate modelling and multiobjective optimization for energy-efficient CO2 capture using functionalized oil palm ash

Abstract

Agricultural waste-derived adsorbents offer a sustainable approach for carbon capture while supporting biomassvalorization. However, identifying operating conditions that simultaneously maximize CO2 adsorption and minimize energy consumption remains challenging because adsorption performance depends on multiple interacting variables. This study presents a machine learning-assisted optimization framework for CO2 capture using KOH-functionalized oil palm ash (OPA). A dataset comprising 346 fixed-bed adsorption experiments was used to develop Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Regression Tree models, employing the OPA-to-KOH ratio, gas flow rate, adsorption temperature, inlet CO2 concentration, and contact time as input variables. Among the models evaluated, GPR achieved the highest predictive performance, with an R2 of 0.9984, a root mean square error of 0.0262 mmol g−1, and a mean absolute error of 0.0119 mmol g−1. The validated GPR model was integrated with Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) to maximize adsorption capacity, and with Multi-Objective Particle Swarm Optimization (MOPSO) and Non-dominated Sorting Genetic Algorithm II (NSGA-II) to optimize both adsorption capacity and electrical energy consumption. Both GA and PSO predicted a maximum adsorption capacity of 2.954 mmol g−1, while the multi-objective algorithms identified Pareto-optimal operating conditions that balanced adsorption performance with energy demand. The proposed framework demonstrates the potential of integrating machine learning with multi-objective optimization to support the design of energy-efficient and sustainable CO2 adsorption processes.