Harnessing the power of machine learning for carbon capture, utilisation, and storage (CCUS) – a state-of-the-art review

Abstract

A review of the state-of-the-art applications of machine learning for CO2 capture, transport, storage, and utilisation.

Research progress in metal–organic frameworks (MOFs) in CO2capture from post-combustion coal-fired flue gas: characteristics, preparation, modification and applications

“…However, in the face of such a huge search space, it is actually impossible to screen the MOFs by only using single-point DFT calculations and GCMC simulations. 170,171 In recent years, machine learning (ML) has gradually become a powerful means of high-throughput screening, which is a collective term for advanced statistical tools and algorithms used to classify, predict, optimize, and cluster data. The ML is mainly composed of three parts, namely collecting data sets, preparing and analyzing descriptors as input variables, and training algorithms (such as decision trees, support vector machines, neural networks, random forests, etc.…”

“…As an emerging field in MOF screening, ML still has some deficiencies and points worthy of breakthrough. 170–172 The ML is a data-driven technology, and its performance usually depends on the size and quality of the database. It is necessary to establish an up-to-date, accurate, consistent and comprehensive database, and combine and examine MOFs from different databases in the ML studies to promote the discovery of common laws.…”

Advances, challenges, and perspectives for CCUS source-sink matching models under carbon neutrality target

“…Several machine learning algorithms have been used to predict and monitor CO 2 leaking to ensure the safe and long-term storage of injected CO 2 and create surrogate models for optimizing the EOR process and uncertainty analysis. Applying machine learning in CCUS shows great potential in identifying links between data or results that are not readily identifiable, and it also provides alternative lower computing cost pathways [36].…”

Physics-Based Neural Networks for Simulation and Synthesis of Cyclic Adsorption Processes

“…To address the computational challenges posed by adsorption process design and optimization, the use of machine learning techniques such as artificial neural networks (ANNs) have emerged as alternatives to rigorous mathematical models. , To this end, Sant Anna et al developed three-layer feed-forward ANN (input layer, one hidden layer, and one output layer) models for the separation of methane and nitrogen using PSA. Using these models in the optimization, the authors show that the computational times significantly reduced from 15.7 h to 50 s. Subraveti et al constructed three-layer feed-forward ANN models initially within an optimization framework and subsequently used them to determine the Pareto solutions of multiobjective maximization of CO 2 purity and recovery for a complex eight-step PSA cycle designed for precombustion CO 2 capture.…”