In Process Systems Engineering, computationally-demanding models are frequent and plentiful. Handling such complexity in an optimization framework in a fast and reliable way is essential, not only for generating meaningful solutions but also for providing decision support. Indeed, optimization results need to be obtained efficiently without compromising accuracy or solution quality. Surrogate models are a cheap way of replacing complex ones, while still capturing the intrinsic features that make them unique and valuable. In this work, a methodology to build surrogate models is developed. It combines a genetic algorithm with an active learning method, harvesting the benefits of both approaches - on the one hand leveraging nature-inspired optimization procedures that explore the optimization space area, and conversely a 'smart' approach to adding meaningful sample points to the training stage. The methodology, coined GA-AL, is tested and validated for chemical absorption of CO2 in a biogas mixture inserted in a utility superstructure framework. Nine surrogate modes are tested, with Artificial Neural Networks, Random Forest and Kringing outperforming other approaches, assessed via four performance criteria.