Constrained Factorized Dilated Temporal Convolutional Networks for Process Gases Management in Steel Manufacturing

Abstract

Process gases play a crucial role in steelmaking, but their efficiency is limited by various production factors. Accurate and real-time forecasting of process gas flow is essential for optimising low-caloric gas as a recycled energy source, reducing waste, and improving energy efficiency. However, traditional forecasting models, such as recurrent statistical approaches, fail to capture the intricate spatio-temporal dependencies of process gases, leading to inaccurate predictions and excessive gas flaring. To address this challenge, we propose a constrained intelligent forecasting system that delivers real-time, high-precision gas flow predictions while being generalisable to various process gases stages in compliance with gases constraints. Our method introduces a novel Factorized Dilated Temporal Convolutional Network (FD-TCN). While classical TCNs leverage dilated convolution layers for capturing spatial-temporal information, they suffer from a fixed, task-independent dilation factor which skips important local dynamics of steelmaking processes. FD-TCN mitigates this issue by introducing a factorised dilation mechanism, enabling the efficient expression of diverse temporal patterns. Additionally, existing intelligent gas flow forecasting models often produce unrealistic predictions by failing to enforce system constraints on numerical flow rates. Our approach integrates domain-specific constraints into the training process, eliminating out-of-range forecasts and improving reliability. We validated FD-TCN across single-variate and multi-variate gas flow forecasting tasks and generalised the method into gas production, gas in manufacturing consumption, and mixed gas recycling. The optimised model via hyperparameter tuning achieved the lowest RMSE, MAPE, MAE and prediction time among tested architectures. Compared to state-of-the-art deep learning models, including long short term-memory (LSTM) and standard TCN, our constrained FD-TCN demonstrated up to 39.7% higher prediction accuracy and a 7.5× faster prediction time. Constrained FD-TCN presents an efficient, accurate and generalisable gas forecasting solution for energy-efficient steelmaking.

Keywords: Process gases, Steel plant, Temporal convolutional networks, Factorized convolution, Constrained learning