Enhanced Bidirectional LSTM (BiLSTM) Model with an Attention Mechanism for Demand Forecasting in Manufacturing Systems: A Case Study of the Polymer Industry

Mehdi  Sadeghi Moein; Mahmoud Mohammadi; Mehrdad  Allahgholizadeh Azari

Authors

Mehdi Sadeghi Moein Department of Industrial Management, CT.C., Islamic Azad University, Tehran, Iran Author https://orcid.org/0009-0003-5351-4515
Mahmoud Mohammadi Department of Industrial Management, CT.C., Islamic Azad University, Tehran, Iran Author https://orcid.org/0000-0001-5064-4419
Mehrdad Allahgholizadeh Azari Department of Industrial Management, CT.C., Islamic Azad University, Tehran, Iran Author https://orcid.org/0009-0008-8671-0615

Keywords:

Demand forecasting, BiLSTM, Attention, Deep learning, Time, series forecasting, Make, to, stock (MTS)

Abstract

Accurate demand forecasting is a fundamental prerequisite for production planning, inventory control, and operational decision-making in manufacturing industries. In production systems, demand fluctuations can lead to reduced resource productivity, increased inventory levels, higher operating costs, and lower customer service levels. Although deep learning–based models have advanced considerably in recent years, many of them still face limitations in simultaneously capturing long-term temporal dependencies and focusing on the most informative patterns within time-series data. This study proposes a BiLSTM-Attention neural network model for demand forecasting of products in a polyester film (BOPET) manufacturing company. Historical demand data were preprocessed, normalized, and structured into temporal sequences for model training. Model performance was evaluated using a walk-forward validation scheme and the MAE, RMSE, and Safe MAPE metrics, and was compared with the forecasting method currently used in the plant. The results indicate that the proposed model effectively learns nonlinear relationships, temporal dependencies, and fluctuating demand patterns, and substantially reduces forecasting error relative to the existing method. Moreover, the model’s forecasts can serve as reliable inputs for make-to-stock (MTS) production planning and other operational decisions. The findings demonstrate that the BiLSTM-Attention architecture not only improves demand forecasting accuracy but also offers practical applicability in real manufacturing environments and potential generalizability to other industries with time-series data.

Downloads

Download data is not yet available.

References

[1] R. Shirazizadeh, "An Intelligent Demand Forecasting Model in the Supply Chain Using Machine Learning Approaches," Yazd University, 2022.

[2] M. Soltani, S. M. Khatami, M. Amiri, and M. Hajian Heidari, "A Hybrid Approach to Integrated Omnichannel Demand Forecasting Using Machine Learning: Time-Series Clustering with the Dynamic Time Warping Algorithm and Artificial Neural Networks," Journal of Production Management, University of Isfahan, vol. 14, no. 32, pp. 121-140, 2023.

[3] A. Alfialali, H. Ben Lamar, S. Alfialali, and M. Kasbuya, "Applications of Machine Learning in Supply Chain Management: An Optimized LSTM Model for Demand Forecasting," International Journal of Intelligent Engineering and Systems, vol. 15, no. 2, 2022.

[4] S. S. Moradi Sanjed Bouri, "A Hybrid Meta-Learning and Generative Adversarial Network Approach for Dairy Product Demand Forecasting," Mazandaran University of Science and Technology, 2022.

[5] C. C. Wang, C. H. Qin, and W. J. C. Therapy, "Application of ARIMA and LSTM for Order Demand Forecasting Based on Short Lead Time and On-Time Delivery Requirements," Processes, vol. 9, no. 7, p. 1157, 2021, doi: 10.3390/pr9071157.

[6] H. D. Nguyen, S. T. Tran, A. Tomasi, and M. Hamad, "Forecasting and Anomaly Detection Approaches Using LSTM Autoencoder Techniques with Applications in Supply Chain Management," International Journal of Information Management, vol. 57, p. 102282, 2021, doi: 10.1016/j.ijinfomgt.2020.102282.

[7] S. H. Rafie, N. Almasud, S. Rahman, and K. Hossain, "A Short-Term Load Forecasting Approach Using an Integrated CNN and LSTM Network," IEEE Access, vol. 10, 2021, doi: 10.1109/ACCESS.2021.3060654.

[8] M. S. Kim, J. Y. Lee, C. Lee, and J. Jeong, "A KDE and LSTM-Based Two-Dimensional Forecasting Framework for Cost-Effective Inventory Management in Smart Manufacturing," Applied Sciences, vol. 12, no. 5, p. 2380, 2022, doi: 10.3390/app12052380.

[9] S. K. A. Al-Bayati, "Electric Load Forecasting Using an LSTM Neural Network and the Grey Wolf Algorithm," Imam Reza International University, 2022.

[10] A. M. M. Al-Zaidawi, "Oil Price Forecasting Using an LSTM Network Optimized by the Grey Wolf Algorithm," Imam Reza International University, 2023.

[11] R. V. Joseph, A. Mohanty, S. Tyagi, S. Mishra, S. K. Satpati, and S. N. Mohanty, "A Hybrid Deep Learning Framework with CNN and Bidirectional LSTM for Store Item Demand Forecasting," Computers & Electrical Engineering, vol. 103, p. 108358, 2022, doi: 10.1016/j.compeleceng.2022.108358.

[12] X. Hao, Y. Liu, L. Pei, W. Li, and Y. Du, "Atmospheric Temperature Prediction Based on a BiLSTM-Attention Model," Symmetry, vol. 14, no. 11, p. 2470, 2022, doi: 10.3390/sym14112470.

[13] Z. Shahim Garmi, "Optimization of LSTM-CNN Hybridization for Time-Series Forecasting," Islamic Azad University, Safadasht Branch, 2022.

[14] J. Zhang, L. Ye, and Y. Lai, "Stock Price Prediction Using CNN-BiLSTM-Attention Model," Mathematics, vol. 11, no. 9, p. 1985, 2023, doi: 10.3390/math11091985.

[15] L. Cui, Y. Chen, J. Deng, and Z. Han, "A Novel AttLSTM Framework Combining Bidirectional LSTM and Self-Attention Mechanism for Demand Forecasting," Expert Systems with Applications, vol. 254, p. 124409, 2024, doi: 10.1016/j.eswa.2024.124409.

[16] A. Ghazouani, A. Masmoudi, A. Mejri, and S. B. Laib, "A Hybrid Deep Learning Model CNN-LSTM for Detergent Product Demand Forecasting," International Journal of Supply and Operations Management, vol. 11, no. 4, pp. 417-429, 2024.

[17] D. Kim, Y. Lee, S. K. Jo, and Y. M. Wai, "Mid-Term Minimum Demand Forecasting Based on Parallel LSTM-MLP Model," IEEE Access, vol. 12, 2024, doi: 10.1109/ACCESS.2024.3520986.

[18] S. Hamian, Y. Kano, H. Khalifi, and M. Telmam, "Comparative Analysis of LSTM, ARIMA, and Hybrid Models for GDP Forecasting," International Journal of Safety and Security Engineering, vol. 29, no. 3, pp. 853-861, 2024, doi: 10.18280/isi.290306.

[19] Y. Jung, C. X. Chen, B. C. Jung, C. H. Tsai, and S. Y. Yang, "Optimizing LSTM Parameters for Flash Flood Forecasting Using Genetic Algorithms," Water Resources Management, vol. 38, no. 1, pp. 1141-1164, 2024, doi: 10.1007/s11269-023-03713-8.

[20] O. Bouakline, Y. El Marabet, A. El Idrissi, K. Khamsi, and R. Laghreib, "A Hybrid Deep Learning Model Integrating LSTM and Modified Genetic Algorithm for Air Quality Applications," Environmental Monitoring and Assessment, vol. 196, p. 1264, 2024, doi: 10.1007/s10661-024-13447-8.

[21] J. Zhang, S. Ko, Z. Zhang, and S. Cheng, "Improved Genetic Algorithm and Optimized LSTM Model for Short-Term Traffic Flow Forecasting," PeerJ Computer Science, vol. 8, p. e1048, 2022, doi: 10.7717/peerj-cs.1048.

[22] S. Nikhil, R. K. Sah, S. K. Parki, and S. Reddy, "Stock Market Prediction Using a Hybrid LSTM-CNN Model with Genetic Algorithm," 2023: IEEE.

[23] Y. Huang, C. Chen, and X. Dai, "A CNN-BiLSTM-Attention-Based Deep Learning Approach for Predicting Asphalt Pavement Performance," Buildings, vol. 16, no. 6, p. 1150, 2026, doi: 10.3390/buildings16061150.

[24] S. Yu, Y. Tang, X. Xu, and T. Liu, "A Hybrid CNN-BiLSTM-Attention Model with Feature Engineering for Accurate Carbon Emission Forecasting," Air Quality, Atmosphere & Health, vol. 19, p. 56, 2026, doi: 10.1007/s11869-026-01954-1.

[25] Y. Wang, R. Tian, X. Liu, and R. Tan, "TOC Logging Prediction Method Based on CNN-BiLSTM-Attention," Journal of Geophysical Prospecting, vol. 2, no. 3, pp. 207-217, 2026, doi: 10.1016/j.jgp.2026.05.003.

[26] Y. Gao, T. Fu, M. He, F. Yang, and X. Li, "Heuristic Algorithm Optimization of CNN-BiLSTM-Attention for Reference Crop Evapotranspiration Forecasting under Limited Meteorological Data Availability," Atmosphere, vol. 17, no. 4, p. 382, 2026, doi: 10.3390/atmos17040382.

[27] N. Wu, D. Chen, R. Gu, C. Wen, W. Tian, and J. Lu, "A Study on the MSC-BiLSTM Ship Track Prediction Model Incorporating an Adaptive Attention Mechanism," Journal of Marine Science and Engineering, vol. 14, no. 10, p. 924, 2026, doi: 10.3390/jmse14100924.

[28] H. Zheng et al., "Accurate Time-Series Forecasting of Floating Platform Motion via a Reinforced Fusion CNN-BiLSTM-Attention Model," PLOS ONE, 2026, doi: 10.1371/journal.pone.0342081.

[29] S. E. Sorour and A. E. Amin, "EvoDeep-Quality: A Closed-Loop Hybrid Framework Integrating CNN-LSTM and NSGA-III for Adaptive Quality Optimization in Smart Manufacturing," Sustainability, vol. 18, no. 8, p. 3679, 2026, doi: 10.3390/su18083679.

[30] J. Feng, "Multi-Node Supply Chain Demand Forecasting and Fluctuation Modeling Based on Transformer-LSTM Hybrid Model," Journal of Engineering, Project & Production Management, vol. 16, no. 5, p. 1, 2026, doi: 10.32738/JEPPM-2025-366.

[31] H. Hu, J. Cai, and C. Xu, "A Mathematical Framework for E-Commerce Sales Prediction Using Attention-Enhanced BiLSTM and Bayesian Optimization," Mathematics, vol. 31, no. 1, p. 17, 2026, doi: 10.3390/mca31010017.

[32] S. Debnath, M. U. Mia, M. Abubakkar, M. R. Islam, M. S. I. Mridul, and A. K. Biswas, "Hybrid Multi-Scale Deep Learning Enhanced Electricity Load Forecasting Using Attention-Based Convolutional Neural Network and LSTM Model," IEEE Access, vol. 14, 2026, doi: 10.1109/ACCESS.2026.3656545.