Development of Generalised Multi-Well CO2 Storage Proxy Models for saline aquifers and depleted gas reservoirs in CCUS Networks

Engineers are training machine learning models to simulate how carbon dioxide behaves when injected deep underground into saltwater aquifers and depleted gas fields. Carbon capture and storage (CCS) is a key technology for reducing industrial CO₂ emissions, but designing a full CCS network—linking multiple power plants or factories to multiple storage sites—requires running thousands of complex geological simulations. Each simulation can take hours or days. That computational bottleneck makes it impractical to optimise the whole network in real time. This project replaces those slow, physics-based simulations with fast, accurate proxy models built from machine learning. The models learn the essential behaviour of each storage site from a limited set of full simulations, then predict outcomes for new injection scenarios in seconds. If the approach works, network operators could rapidly test thousands of possible configurations—matching emission sources to storage sites, adjusting injection rates, and planning pipeline routes—without waiting weeks for results. The research targets two common storage types: saline aquifers and depleted gas reservoirs, so the proxy models could be applied across many real-world CCS projects. The work is applied, not fundamental: its immediate goal is to make CCS network optimisation computationally feasible.