針對季節性需求商品製造業者面臨的生產波動問題，本研究提出一「基於季節性高峰需求預測之多層次滾動式平準化生產模式」，得以最小化生產過程中的波動，避免庫存過剩或短缺，同時提升整體效率、降低成本、提高產品品質。
本方法首先使用CNN-LSTM深度學習模式，進行下年度與該年季節性高峰之銷售需求預測，再使用CNN-LSTM深度學習模式，搭配多步時間序列預測方法，進行滾動式、階層式之年與月的即期需求預測，動態地掌握未來各年與各月的銷售需求。這預測方法能夠捕捉到長期趨勢和短期波動，為後續的平準化生產規劃提供可靠的基礎。接著，應用移動式窗平準法，對每年、每月和每週的生產量進行平準處理。最後，結合多目標函數和非凌越排序遺傳演算法來優化生產策略，確保生產計劃在滿足各項約束條件的同時，達到最佳的平準化效果，以全面、深入地掌握市場需求與生產資源之關係。
本研究以某食品製造公司為例，針對所提「多層次滾動式平準化生產模式」進行驗證與評量。實驗結果顯示，本方法在每月和每週尺度上均展現出顯著的效果，成功地減少了生產波動。此外，在庫存管理方面，本方法也表現出良好的調節作用，儘管在短期內會出現庫存水平的暫時上升，但從長期來看，能夠更好地滿足市場需求，並提升生產效率。

To address the issue of production fluctuations faced by manufacturers of seasonal demand products, this study proposes a "Multi-level Rolling Smoothing Production Model Based on Seasonal Peak Demand Forecasting." This model aims to minimize fluctuations during the production process, thereby avoiding excess or shortage in inventory, while simultaneously improving overall efficiency, reducing costs, and enhancing product quality.
The proposed method first employs a CNN-LSTM deep learning model to forecast sales demand for the upcoming year and the seasonal peaks within that year. Subsequently, a CNN-LSTM model, combined with a multi-step time series forecasting approach, is used to perform rolling and hierarchical demand forecasting for the immediate future on both annual and monthly bases. This dynamic approach allows for an accurate understanding of both long-term trends and short-term fluctuations, providing a reliable foundation for subsequent smoothing production planning. Next, a moving window smoothing method is applied to smooth production volumes on an annual, monthly, and weekly basis. Finally, multi-objective functions and a non-dominated sorting genetic algorithm (NSGA-II) are employed to optimize the production strategy, ensuring that the production plan meets all constraints while achieving the best possible smoothing effect. This comprehensive approach facilitates a deep understanding of the relationship between market demand and production resources.
The proposed "Multi-level Rolling Smoothing Production Model" was validated and evaluated using a case study of a food manufacturing company. The experimental results demonstrate that this method effectively reduces production fluctuations on both monthly and weekly scales. Additionally, in terms of inventory management, the method shows strong regulatory effects. Although there may be a temporary increase in inventory levels in the short term, the method is capable of better satisfying market demand and improving production efficiency in the long term.

中文參考文獻
林柏呈. (2023). 基於機器學習之多步銷售預測方法研究. 國立成功大學製造資訊與系統研究所學位論文
英文參考文獻
Alessandro Crivellari, Euro Beinat, Sandor Caetano, Arnaud Seydoux, Thiago Cardoso. (2022). Multi-target CNN-LSTM regressor for predicting urban distribution of short-term food delivery demand. Journal of Business Research 144 (2022) 844–853.
Asiye Kaymaz Özcanlı, Mustafa Baysal, Asst.Prof. (2022). A novel Multi-LSTM based deep learning method for islanding detection in the microgrid. Electric Power Systems Research 202 (2022) 107574.
Charles C. Holt. (2004). Forecasting seasonals and trends by exponentially weighted moving averages. International Journal of Forecasting 20 (2004) 5–10.
Christopher Carlström & Peder Emond. (2011). Improving the flow of materials and information from a Lean perspective - A study performed as a part of a project at Faiveley Transport Nordic AB. Engineering logistics, Lund University.
Changrui Deng, Yanmei Huang, Najmul Hasan, Yukun Bao. (2022). Multi-step-ahead stock price index forecasting using long short-term memory model with multivariate empirical mode decomposition. Information Sciences, 607, 297-321.
Felix Kamhubera, Thomas Sobottkaa, Bernhard Heinzl, Jan Henjes, Wilfried Sihna. (2020). An efficient hybrid multi-criteria optimization approach for rolling production smoothing of a European food manufacturer. Computers & Industrial Engineering Volume 147, September 2020, 106620.
Gorman, M. F., & Brannon, J. I. (2000). Seasonality and the production-smoothing model. International Journal of Production Economics, 65(2), 173–178.
Hanxiao Shi, Anlei Wei, Xiaozhen Xu, Yaqi Zhu, Hao Hu, Songjun Tang. (2024). A CNN-LSTM based deep learning model with high accuracy and robustness. Journal of Environmental Management 352 (2024) 120131.
for carbon price forecasting: A case of Shenzhen’s carbon market in China
He, Q.-Q., Wu, C., & Si, Y.-W. (2022). LSTM with particle swarm optimization for sales forecasting. Electronic Commerce Research and Applications, 51, 101118.
He-Xuan Hu, Liang-Huan Shao, Qiang Hu, Ye Zhang, Zhen-Yun Hu. (2021). Multi-objective Reservoir Optimal Operation Based on GCN and NSGA-II Algorithm. 2021 4th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE). IEEE.
Jimmy Ming-Tai Wu, Zhongcui Li, Norbert Herencsar, Bay Vo, Jerry Chun-Wei Lin. (2021). A graph-based CNN-LSTM stock price prediction algorithm with leading indicators, Multimed Syst, vol. 22, no. 2, pp. 1-20, 2021.
Jun Li, Guofang Wu, Yongpeng Zhangc, Wenhui Shi. (2024). Optimizing flood predictions by integrating LSTM and physical-based models with mixed historical and simulated data. Heliyon 10 (2024) e33669.
Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, T. Meyarivan. (2022). A fast and elitist multiobjective genetic algorithm: NSGA-II. In 2022 IEEE Transactions on Evolutionary Computation (Volume: 6, Issue: 2, April 2002). IEEE.
Lewis, C. D. (1982). Industrial and business forecasting methods. London: Butterworths.
Liang Guo, Weiguo Fanga, Qiuhong Zhao, Xu Wang. (2021). The hybrid PROPHET-SVR approach for forecasting product time series demand with seasonality. Computers & Industrial Engineering 161 (2021) 107598.
Lina Jianga, Xiaofeng Sun, Cuicui Ji, Stefane Mostefa Kabene, Mohammed Yousuf Abo Keir. (2021). PDCA cycle theory based avoidance of nursing staff intravenous drug bacterial infection using degree quantitative evaluation model. Results in Physics 26 (2021) 104377.
Longfei Zhou, Lin Zhang, Berthold K.P. Horn. (2020). Deep reinforcement learning-based dynamic scheduling in smart manufacturing. Procedia CIRP Volume 93, 2020, Pages 383-388.
Mohaiad Elbasheer, Francesco Longo, Letizia Nicoletti, Antonio Padovano,Vittorio Solina, Marco Vetrano. (2022). Applications of ML/AI for Decision-Intensive Tasks in Production Planning and Control. Procedia Computer Science Volume 200, 2022, Pages 1903-1912.
Mohammad Ehteram, Mahdie Afshari Nia, Fatemeh Panahi, Alireza Farrokhi. (2024). Read-First LSTM model: A new variant of long short term memory neural network for predicting solar radiation data. Energy Conversion and Management 305 (2024) 118267.
Olumide Emmanuel Oluyisola, Fabio Sgarbossa, Jan Ola Strandhagen. (2020). Smart Production Planning and Control: Concept, Use-Cases and Sustainability Implications. Sustainability, 12 (9) (2020), p. 3791
Qiang Li, Xinjia Guan, Jinpeng Liu. (2023). A CNN-LSTM framework for flight delay prediction. Expert Systems With Applications 227 (2023) 120287.
Rewers R. (2019). Planning the inflow of products for production levelling. Machines. Technologies. Materials. Vol. 13 (2019), Issue 10, pg(s) 439-442.
Shengluo Yang, Zhigang Xu. (2021). Intelligent Scheduling for Permutation Flow Shop with Dynamic Job Arrival via Deep Reinforcement Learning. In 2021 IEEE 5th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC). IEEE.
Shu Luo. (2020). Dynamic scheduling for flexible job shop with new job insertions by deep reinforcement learning. Applied Soft Computing Journal 91 (2020) 106208.
Sushil Punia, Sonali Shankar. (2022). Predictive analytics for demand forecasting: A deep learning-based decision support system. Knowledge-Based Systems 258 (2022) 109956.
Tareq M. Shami, Ayman A. El-Saleh, Mohammed Alswaitti, Qasem Al-Tashi, Mhd Amen Summakieh, Seyedali Mirjalili. (2022). Particle Swarm Optimization: A Comprehensive Survey. In 2022 IEEE Access (Volume: 10) (pp. 10031-10061). IEEE.
Tianqi Chen, Tong He, Michael Benesty, Vadim Khotilovich, Yuan Tang, Hyunsu Cho, Kailong Chen, Rory Mitchell, Ignacio Cano, Tianyi Zhou. (2015). Xgboost: Extreme gradient boosting. R Package Version 0.4-2, 1 (4) (2015), pp. 1-4
Weijie Zhou, Yuke Cheng, Song Ding, Li Chen, Ruojin Li. (2021). A grey seasonal least square support vector regression model for time series forecasting. ISA Transactions. Volume 114, August 2021, Pages 82-98.
Yaqing Lei, Ying Wan, Junjie Fu, Bei Hu. (2022). NSGA-II based Electric Vehicle Charging Scheduling for Daily Commuting. In 2022 34th Chinese Control and Decision Conference (CCDC). IEEE.
Yasin Tadayonrad, Alassane Balle Ndiaye. (2023). A new key performance indicator model for demand forecasting in inventory management considering supply chain reliability and seasonality. Supply Chain Analytics 3 (2023) 100026.
Yejlan Zhao, Yanhong Wang, Yuanyuan Tan, Jun Zhang, Hongxia Yu. (2021). Dynamic Job-shop Scheduling Algorithm Based on Deep Q Network. In 2021 IEEE Access (Volume: 9). (pp. 122995-123011). IEEE.
Yukun Bao, Tao Xiong, Zhongyi Hu. (2014). Multi-step-ahead time series prediction using multiple-output support vector regression. Neurocomputing, Volume 129, 10 April 2014, Pages 482-493.
Zhen Cui, Yanlai Zhou, Shenglian Guo, Jun Wang, Chong-Yu Xu. (2022). Effective improvement of multi-step-ahead flood forecasting accuracy through encoder-decoder with an exogenous input structure. Journal of Hydrology, 609, 127764.
Žliobaitė, I., Bakker, J., & Pechenizkiy, M. (2012). Beating the baseline prediction in food sales: How intelligent an intelligent predictor is? Expert Systems with Applications, 39(1), 806-815.