Enabling industrial decarbonization through data-metrics guided interpretable machine learning

Achieving industrial decarbonization requires accurate system-level modeling for the analysis of industrial systems. First-principle models, based on assumptions, are unable to forecast the actual condition of large-scale industrial systems, whereas pure data-centric machine learning (ML) algorithms do not possess an interpretable concept. This highlights the significance of addressing this issue to fully harness AI's potential for industrial decarbonization. The interpretable ML models can be trained by integrating the system's information, as quantified by data metrics, into the algorithmic design that offers improved interpretability performance for industrial system design, operation, and control. In this article, we offer a framework on how an interpretable ML-powered lever can significantly enhance energy efficiency, reduce resource consumption, and reduce emissions through smart operation management of industrial systems, including hard-to-decarbonize industrial sectors, thereby facilitating industrial climate action.