Comprehensive Literature Review of Deep Learning Models for Forecasting of Hot Metal Production in Steel Industry

A comprehensive review of literature for the years 2010 to 2025 has been accomplished, intended for the identification of the suitability of deep learning (DL) technologies for forecasting hot metal production in the steel industry. This review process has been divided into two segments, in which the first segment includes the identification of the capabilities of various machine learning techniques for chaos forecasting, and, in the second segment, their suitability for hot metal production forecasting. Since 2017, the DL techniques have been found to be more suitable than conventional machine learning processes; therefore, the review of the capabilities of LSTM and Bi-LSTM techniques by various authors has been incorporated. As far as the performance of these machine learning techniques goes, the LSTM and Bi-LSTM techniques have been found to be more suitable. However, they also have certain limitations, and their solutions have been suggested and provided through optimization of the learning process with hybridization. In 2024, the chaotic behavior of long-range rainfall was successfully identified by LSTM, Bi-LSTM, hybrid PCA-O-LSTM, and PCA-O-Bi-LSTM models, and it was observed that the PCA-O-Bi-LSTM model was found to be more accurate with 98% accuracy with 100 epochs. The entire outcome of the review is given in this article.