Decarbonization of Steel Mills, how to Set-up an Effective Allocation of Energy Flows, aiming at Tracking Efficiency and Energy Savings assisted with Machine learning tools

Achieving a net zero emission steel industry in the implementation phase requires strong methodologies, along with comprehensive data collection frameworks, to enable meaningful comparisons and to monitor progress. The exploration and application of machine learning tools at process level is an aligned step toward achieving such a goal. Starting from these claims, the paper presents the application of ISO 50001 to three steel mills, different in size and location, and it supports the creation of their baselines. Finding the energy saving opportunities, which are helpful in earning carbon credits, and implementing the framework for assessing the ongoing improvements, by the means of specific performance indicators (i.e. toe/t and €/t) for its energy intensive processes, are therefore a few contributions of this work. Machine learning techniques have been applied to air compressors dataset to investigate their expected specific energy consumptions and to find potential anomalies. The comparative analysis of various performance metrics (RMSE, MSE, MAE and R2), applied to the energy performance indicators identified by the implementation of the ISO, offers a comprehensive evaluation and shows the superiority of support vector machine over other alternative algorithms. The proposed architectural framework is highly scalable, accommodating data availability ranging from yearly to monthly and daily levels also across different mills and processes. Thus, the novelty of this work is twofold. Methodologically it involves the application of ISO50001,andpractically, it applies a machine learning tool to an auxiliary service within the same mills to evaluate the robustness of the proposed architectural framework. The results show how powerful these tools are when combined and used for inter/intra assessments.