Post by Arjun Chand

Recently I’ve been working on a research project to develop a machine-learning based model for carbon capture prediction. As the oil and gas industry pushes towards industrial decarbonization, optimizing carbon capture facilities is critical. However, chemical absorption processes are highly dynamic and non-linear. Traditional physical modeling can be computationally heavy and slow to respond to real-time process changes. To address this, machine learning can be deployed as an intelligent, real-time software sensor to predict CO2 capture rates instantaneously. ⬇️Here are some notable data science takeaways from the project so far⬇️    🔴 Inertia must be accounted for Standard machine learning models have a habit of looking at data points in isolation. In a real-world carbon capture facility, temperatures, pressures, flow rates, and concentrations do not spike instantly. This is due to inertia: a physical delay. By implementing lag features, the model is given a short-term memory of the plant's recent measurements, allowing an accurate portrayal of this inertia, which massively boosts the model’s predictive accuracy.  🔴 The roles of linear and nonlinear models A standard multiple linear regression model with lag features predicted captured carbon very well, achieving an R2 value of ~ 0.97. In other words, the AI successfully captures 97% of the actual process dynamics, attributing the remainder of the variance to random high-frequency sensor noise. Chemical processes don't operate in a perfectly linear manner, however. Temperatures pressures, flow rates, and concentrations changing over time in smooth, continuous curves. By upgrading to an artificial neural network and using Bayesian Optimization to find the optimal mathematical activation function, a slightly improved R2 value of ~ 0.98 was achieved. 💭 While this project is actively developing, it is already illustrating the potential of integrating fundamental chemical engineering and data analysis principles with machine learning models to resolve the oil and gas industry’s most critical environmental challenges.