DT-MNLR: a novel hybrid machine learning framework for precise coke strength and reactivity prediction

Abstract

Accurately predicting coke strength after reaction (CSR) and coke reactivity index (CRI) is important for optimising coke quality in metallurgical industry, thereby minimising production costs and maximising resource utilisation. This study introduces a novel machine-learning model, the decision tree multi-output non-linear regression (DT-MNLR) model, for accurately predicting both CSR and CRI. The DT-MNLR model leverages the strengths of multiple algorithms: decision trees for efficient coal blend classification, multi-output regression for handling the interrelated nature of CSR and CRI, and a backpropagation neural network for capturing complex non-linear relationships within the data. Recognising the intricate interactions among coal properties that significantly impact coke quality, the model incorporates high-level polynomial features and additional coal property variables, enhancing its predictive accuracy. Rigorous validation using diverse testing samples demonstrates the DT-MNLR model's superior performance across a wide range of CSR and CRI values. Comparative analysis against other machine-learning methods showcases the DT-MNLR model's advantages, including lower prediction errors, improved generalisation to unseen data and enhanced robustness in handling outliers. This research significantly advances the field of coke quality prediction by establishing the DT-MNLR model as a powerful tool for coal blend analysis and quality control. The model's effectiveness paves the way for significant advancements in intelligent systems for industrial applications, promoting optimal resource utilisation and process efficiency.