Temperature Forecasting in a Wire Rod Production Line Using Real‐Time Data

In iron and steel production processes, monitoring the cooling status is important to improve the microstructure and strength properties of the steel. The geometric shape of the wire rods makes it difficult to achieve homogeneous cooling. The inhomogeneity of the distribution requires the correct realization of the air cooling conditions. In order to control the air cooling process, the performances of different models are investigated using real‐time temperature data taken at regular intervals in a specific region of the conveyor line and other process parameters affecting the cooling. Recurrent neural network (RNN), long short‐term memory (LSTM), and random forest (RF) models are used to capture time dependencies and to learn the relationships between independent variables. The performance of these models was tested with error measures, such as mean square error (MSE) and R‐squared ( R 2 ). The results indicate that the LSTM model provides more robust performance in modeling temporal relationships compared to the RNN and RF models, while the RF model demonstrates strong capability in learning complex relationships among the input variables.