Domain consistent industrial decarbonisation of global coal power plants

Abstract Machine learning and optimisation techniques offer promise for decarbonising thermal power plants by reducing process-related emissions. However, their adoption in industrial settings often falters due to inadequate domain-consistency. Here, we introduce a human-in-the-loop constraint-based optimisation framework that integrates domain expertise with data-driven methods to deliver implementable solutions. We showcase its effectiveness at improving thermal efficiency and reducing turbine heat rate in a 660 megawatts coal-fired power plant operating at supercritical pressure. Empirical tests reveal a mean efficiency gain of 0.64% and a mean turbine heat rate reduction of 93 kilojoules per kilowatt-hour. When extended to 56 coal plants of similar capacity and fuel type worldwide, our model projects a cumulative lifetime mitigation of 60.2 million tons of carbon emissions. These results emphasise the transformative potential of machine learning and optimisation techniques with human expertise input in delivering domain-compliant, deployable solutions for large-scale industrial decarbonisation.