Hybrid prediction model of oxygen consumption in converter steelmaking for small-sample data

Abstract

Oxygen blowing critically affects metallurgical efficiency and molten steel quality, making accurate oxygen consumption prediction essential for optimising converter smelting. Growing multi-variety, small-lot production and new steel grade development make small-sample prediction a major challenge in the steel industry. This article proposes a hybrid oxygen consumption prediction model based on parameter optimisation and the oxygen balance mechanism. First, the influence of assumed parameter fluctuations in the mechanism model on oxygen consumption is analysed. Then, oxygen consumption is divided into non-optimised and optimisation-required parts. Finally, the differential evolution algorithm is adopted to optimise parameters and improve prediction accuracy. The dataset is split into development and independent test sets, with 5-fold cross-validation employed to ensure reliable parameter optimisation and model evaluation. The proposed model outperforms unoptimised mechanism, BPNN and SVR models in accuracy and adaptability. It achieves 1.97% MRPE and 255.20 m³ RMSE, with hit rates of 39.39%, 78.79% and 96.97% within ±1%, ±3% and ±5% errors. The model enables accurate small-sample prediction, reducing production costs and improving efficiency.