PREDICTION OF PRODUCTION AMOUNT IN STEEL MANUFACTURING CONSIDERING TRIBOLOGICAL EFFECTS USING LSTM, GRU, BILSTM, AND HYBRID MODWT–LSTM MODELS

Precise prediction of the amount of steel that can be produced within a month (tons) is important to improve the efficiency of planning and also ensure that there are no delays in making the delivery. The paper compares the capacity of four neural network models, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional LSTM (BiLSTM), and a Multiresolution Wavelet Decomposition Hybrid (MODWT-LSTM), to predict production amount in 42 months. R², MSE, MAPE, and RMSE were used to test model performance. GRU and optimised BiLSTM models provided the highest predictive accuracy with MREs of 0.64 % and 0.71 %, with R² greater than 0.997. The LSTM model also performed well (MRE 1.33 %, R² 0.9893), and the MODWT-LSTM model delivered more errors. Even though the optimized BiLSTM can capture in terms of the temporal patterns, the performance of the GRU was slightly higher.