Your Feedback

Energy News Monitoring

Deep Learning Tool for Better CO2 Storage

Aenert news. Environment & Climate Change
Climate change is one of the most pressing issues of our time which requires immediate action to avert its harmful effects. One instrument to counteract this development is an efficient CO2 capture technology.  This requires complex processes and digital systems to optimise big data prediction as well as to reduce production time. A mathematical and statistical approach such as machine learning plays an important role in solving research problems. This approach provides fast results in predicting big data and cost-efficient tools.

There, numerical simulations analysing subsurface flow and transport behaviour are of paramount importance to overcome the challenges due to the multiphysical nature, highly non-linear governing equations, inherent parameter uncertainties, and the need for high spatial resolutions to capture multi-scale heterogeneity.

Now (2023), scientists at the National Energy Technology Laboratory concluded a project in the course of which a deep learning tool for subsurface monitoring was developed that could help ensure safe storage of carbon dioxide (CO2) at geologic sites. A new type of fiber-optic sensing technology called distributed acoustic sensing (DAS) has shown promise as a seismic monitoring tool, but current data management and processing methods used with DAS are not sufficient to fully realise the technology’s potential.

The tool was trained on DAS measurements using convolutional neural networks. Through this deep-learning approach the scientists were able to develop a data filter with high computational efficiency and an improved signal-to-noise ratio relative to other commonly applied filters. The deep-learning tool included faster and more accurate real-time monitoring, identification of more seismic events and improved subsurface imaging.

In view of the climate crisis, scientists are looking into new methods to improve carbon storage. In 2023, a machine learning (ML) system was developed to predict the potential health issues of solvents through uncovering the hidden relationship between substances and toxicity. Prolonged exposure to solvents had been proved to pose significant risks to human health. Therefore, a predictive model for health performance was developed by identifying the contributing factors to solvent toxicity. In the study, Rough Set Machine Learning (RSML) was chosen for this work due to its interpretable nature of the generated models.  Based on the data collection, several models on the toxicity of various organic solvents, the construction of predictive models with decision rules, and model verification were developed. The results showed that there were correlations between solvent toxicity and the Balaban index, valence connectivity index, Wiener index, and boiling points. The generated predictive model using RSML was able to give insightful observations about the correlation between human toxicity and molecular attributes.

Image: Simplified information table of organic solvent toxicity

Source: Wey Ying Hoo, Jecksin Ooi, Nishanth Gopalakrishnan Chemmangattuvalappil, Chong Jia Wen/ An Interpretable Predictive Model for Health Aspects of Solvents via Rough Set Theory/ Processes 11(8):2293, July 2023/ DOI:10.3390/pr11082293/ Open Access This is an Open Access article is distributed under the terms of the
Creative Commons Attribution 4.0 International (CC BY 4.0)

In 2023, an integrated decision-making method for solving the site selection problem for CO2 geological storage was developed. To this end, a multi-objective optimisation was employed including ratio analysis as well as the full multiplicative form (MULTIMOORA) method and prioritized aggregation operators in Pythagorean fuzzy environment. The academic contributions of this study included some Pythagorean fuzzy Schweizer–Sklar prioritized aggregation (PFSSPA), which accommodated the priority levels of criteria and the risk preferences of decision makers. The study also expanded the classical MULTIMOORA method based on the developed aggregation operators (named PFSSPA-MULTIMOORA). Also, the calculation process of this method was described in detail. Later, the PFSSPAMULTIMOORA method was used for ranking the alternatives. Finally, the PFSSPA-MULTIMOORA method was used to solve the site selection problem of CO2 geological storage. A comparative analysis of existing methods was able to prove the effectiveness and robustness of the proposed method.

Image: Distribution of the four sites

Source: Yang Yang, Zhang Chao/ MULTIMOORA Method‑Based Schweizer–Sklar Operations for CO2 Geological Storage Site Selection Under Pythagorean Fuzzy Environment/ International Journal of Computational Intelligence Systems 16(1), March 2023/ DOI:10.1007/s44196-023-00201-0/ Open Access This is an Open Access article is distributed under the terms of the Creative Commons Attribution 4.0 International (CC BY 4.0)

There are several advantages to using deep-learning for CO2 storage: In this system, various system flaws could be mitigated by adapting and optimising novel deep-learning techniques to improve the detectability of certain signals of interest, including local, regional and distant earthquakes. Deep learning which uses artificial neural networks helped the scientists detect changes underground by monitoring the subsurface with sensors. In the case of this project, any form of seismic activity was monitored.

The next step will be to develop the technology into a real-time monitoring tool.  Having received a SBIR Phase II award, this will be realised in the course of large study where researchers will develop a seismic monitoring workflow that could be deployed at carbon storage sites.

By the Editorial Board