預測季節性商品的遠期高峰需求量與近期變化，以有效掌握商品需求趨勢，據以進行生產規劃與生產平準化，避免產銷失調與庫存失衡，為季節性商品產銷的重要活動。
傳統線性統計模型被廣泛應用於銷售預測，但因未考慮時間序列數據的特徵與特徵關係，無法適應銷售量動態變化且差異大的銷售預測。近期以深度學習為基礎的銷售預測方法研究，多偏向於短期或下一期的銷售預測，而這些方法也往往未考慮複雜的市場現象和影響因素，以致無法因應現今因影響因素眾多而快速變化的市場環境之預測需求。
本研究設計一基於機器學習之多步銷售預測方法與開發相關技術，供企業針對季節性商品之遠期高峰點銷售量進行預測，再依近期銷售變化並考量變化影響因素與市場參考因素，動態調整高峰點銷售預測量，據以進行近期生產規劃與生產平準化。
本研究以Y企業所提供的商品銷售數據，進行預測模型之訓練、開發與評比，篩選最佳的預測模型後，再用於所設計之四個預測方法並進行效能評量。在預測模型篩選方面，CNN-LSTM模型之均方根誤差較其它模型少了約28%、19%和8%，說明該模型在本研究的商品銷售數據預測表現具有較好的準確性和預測效果。針對方法效能評量，本研究以兩項季節性商品的歷史銷售數據進行實驗，結果顯示，多輸出預測方法在各項評量指標的表現相對出色，較能準確地預測出季節性的銷售高峰。最後，透過注意力機制對預測準確度的影響實驗，本研究得出無注意力機制的模型在均方根誤差的表現比有注意力機制的模型好12.6%和7.4%，提供了啟示和未來的研究方向。

Predicting the future peak demand and near-term variations of seasonal products is a crucial activity in managing the demand trends of goods, enabling effective production planning and leveling to prevent imbalances between production and sales, as well as inventory disparities. Traditional linear statistical models have been widely employed for sales forecasting. However, due to their failure to consider the temporal characteristics and feature relationships of time series data, they struggle to adapt to the dynamic and diverse sales forecasts with significant variations. Recent sales forecasting methods based on deep learning tend to focus on short-term or next-period predictions, often neglecting complex market phenomena and influencing factors. As a result, they struggle to accommodate the rapidly changing market environment with numerous impacting factors.

This study designs a machine learning-based multi-step sales forecasting approach and develops related techniques for enterprises to predict the future peak sales volumes of seasonal products. This prediction is refined based on recent sales changes, considering variable influencing factors and market reference element. This refinement allows for dynamic adjustment of peak sales forecasts. This information is then utilized for short-term production planning and leveling.

Utilizing sales data provided by Company Y, this study conducts training, development, and evaluation of predictive models. After selecting the optimal predictive model, it is applied to the four designed forecasting methods for performance assessment. In terms of model selection, the CNN-LSTM model exhibits approximately 28%, 19%, and 8% lower root mean square error compared to other models, indicating its superior accuracy and forecasting performance in predicting sales data for this study's products. Regarding method performance evaluation, experiments are conducted using historical sales data of two seasonal products. The results demonstrate that the multi-output prediction method performs relatively well in various evaluation metrics, accurately predicting seasonal sales peaks. Finally, through an experiment analyzing the impact of attention mechanisms on predictive accuracy, this study concludes that the model without attention mechanism outperforms the model with attention mechanism by 12.6% and 7.4% in terms of root mean square error performance. Implications and future research directions are offered.

Abbasimehr, H., Shabani, M., & Yousefi, M. (2020). An optimized model using LSTM network for demand forecasting. Computers & Industrial Engineering, 143, 106435.
Acar, A. Z., Yilmaz, B., & Kocaoglu, B. (2014). Demand forecast, up-to-date models, and suggestions for improvement: An example of a business. Journal of Global Strategic Management, 1(8), 26-26.
Arulraj, T. C., & Jancy Sickory Daisy, S. (2021). Mining online reviews for predicting sales performance. Materials Today: Proceedings, 47(1), 93-99.
Arulraj, T. C., & Jancy Sickory Daisy, S. (2021). Mining online reviews for predicting sales performance. Materials Today: Proceedings, 47(1), 93-99.
Avdeeva, Z., Grebenyuk, E., & Kovriga, S. (2019). Combined approach to forecasting of manufacturing system target indicators in a changing external environment. Procedia Computer Science, 159, 943-952.
Bogaert, M., Ballings, M., Van den Poel, D., & Oztekin, A. (2021). Box office sales and social media: A cross-platform comparison of predictive ability and mechanisms. Decision Support Systems, 147, 113517.
Crivellari, A., Beinat, E., Caetano, S., Seydoux, A., & Cardoso, T. (2022). Multi-target CNN-LSTM regressor for predicting urban distribution of short-term food delivery demand. Journal of Business Research, 144, 844-853.
Cui, Z., Zhou, Y., Guo, S., Wang, J., & Xu, C.-Y. (2022). Effective improvement of multi-step-ahead flood forecasting accuracy through encoder-decoder with an exogenous input structure. Journal of Hydrology, 609, 127764.
Deng, C., Huang, Y., Hasan, N., & Bao, Y. (2022). Multi-step-ahead stock price index forecasting using long short-term memory model with multivariate empirical mode decomposition. Information Sciences, 607, 297-321.
Dong, C., Zhang, K., Wei, X., Wang, Y., & Yang, Y. (2022). Spatiotemporal Graph Attention Network modeling for multi-step passenger demand prediction at multi-zone level. Physica A: Statistical Mechanics and its Applications, 603, 127789.
Dong, L., Zheng, H., Li, L., & Hao, L. (2022). Human-machine hybrid prediction market: A promising sales forecasting solution for E-commerce enterprises. Electronic Commerce Research and Applications, 56, 101216.
Drus, Z., & Khalid, H. (2019). Sentiment Analysis in Social Media and Its Application: Systematic Literature Review. Procedia Computer Science, 161, 707-714.
Guo, C., & Berkhahn, F. (2016). Entity embeddings of categorical variables. arXiv preprint arXiv:1604.06737.1
He, Q.-Q., Wu, C., & Si, Y.-W. (2022). LSTM with particle swarm optimization for sales forecasting. Electronic Commerce Research and Applications, 51, 101118.
Ibrahim, N. F., & Wang, X. (2019). Decoding the sentiment dynamics of online retailing customers: Time series analysis of social media. Computers in Human Behavior, 96, 32-45.
In, Y., & Jung, J. Y. (2022). Simple averaging of direct and recursive forecasts via partial pooling using machine learning. International Journal of Forecasting, 38(4), 1386-1399.
Laib, O., Khadir, M. T., & Mihaylova, L. (2019). Toward efficient energy systems based on natural gas consumption prediction with LSTM recurrent neural networks. Energy, 177, 530-542.
Li, Y., Yang, Y., Zhu, K., & Zhang, J. (2021). Clothing sale forecasting by a composite GRU-Prophet model with an attention mechanism. IEEE Transactions on Industrial Informatics, 17(12), 8335-8344.
Lin, J., Ma, J., Zhu, J., & Cui, Y. (2022). Short-term load forecasting based on LSTM networks considering attention mechanism. International Journal of Electrical Power & Energy Systems, 137, 107818.
Liu, H., Zhang, A., Liu, C., Zhao, Y., Zhao, A., & Wang, D. (2021). Analysis of the time-lag effects of climate factors on grassland productivity in Inner Mongolia. Global Ecology and Conservation, 30, e01751.
Mediavilla, M. A., Dietrich, F., & Palm, D. (2022). Review and analysis of artificial intelligence methods for demand forecasting in supply chain management. Procedia CIRP, 107, 1126-1131.
Mu, S., Wang, Y., Wang, F., & Ogiela, L. (2021). Transformative computing for products sales forecast based on SCIM. Applied Soft Computing, 109, 107520.
Pan, S.-Y., Liao, Q., & Liang, Y.-T. (2022). Multivariable sales prediction for filling stations via GA improved BiLSTM. Petroleum Science, 19(5), 2483-2496.
Pathak, A. R., Pandey, M., & Rautaray, S. (2021). Topic-level sentiment analysis of social media data using deep learning. Applied Soft Computing, 108, 107440.
Peng, S., Han, W., & Jia, G. (2022). Pearson correlation and transfer entropy in the Chinese stock market with time delay. Data Science and Management, 5(3), 117-123.
Punia, S., & Shankar, S. (2022). Predictive analytics for demand forecasting: A deep learning-based decision support system. Knowledge-Based Systems, 258, 109956.
Ramos, P., Santos, N., & Rebelo, R. (2015). Performance of state space and ARIMA models for consumer retail sales forecasting. Robotics and Computer-Integrated Manufacturing, 34, 151-163.
Saha, P., Gudheniya, N., Mitra, R., Das, D., Narayana, S., & Tiwari, M. K. (2022). Demand Forecasting of a Multinational Retail Company using Deep Learning Frameworks. IFAC-PapersOnLine, 55(10), 395-399.
Salman, A. G., Heryadi, Y., Abdurahman, E., & Suparta, W. (2018). Single Layer & Multi-layer Long Short-Term Memory (LSTM) Model with Intermediate Variables for Weather Forecasting. Procedia Computer Science, 135, 89-98.
Sharma, R., & Singhal, P. (2019). Demand Forecasting Of Engine Oil For Automotive And Industrial Lubricant Manufacturing Company Using Neural Network. Materials Today Proceedings, 18(7), 2308-2314.
Shrestha, S. M., & Shakya, A. (2022). A Customer Churn Prediction Model using XGBoost for the Telecommunication Industry in Nepal. Procedia Computer Science, 215, 652-661.
Sousa, M., Tomé, A. M., & Moreira, J. (2022). Long-term forecasting of hourly retail customer flow on intermittent time series with multiple seasonality. Data Science and Management, 5(3), 137-148.
Thakkar, A., & Chaudhari, K. (2020). Predicting stock trend using an integrated term frequency–inverse document frequency-based feature weight matrix with neural networks. Applied Soft Computing, 96, 106684.
Wang, X., Xiang, Z., Xu, W., & Yuan, P. (2022). The causal relationship between social media sentiment and stock return: Experimental evidence from an online message forum. Economics Letters, 216, 110598.
Xue, P., Jiang, Y., Zhou, Z., Chen, X., Fang, X., & Liu, J. (2019). Multi-step ahead forecasting of heat load in district heating systems using machine learning algorithms. Energy, 188, 116085.
Zha, W., Liu, Y., Wan, Y., Luo, R., Li, D., Yang, S., & Xu, Y. (2022). Forecasting monthly gas field production based on the CNN-LSTM model. Energy, 260, 124889.
Zhang, C., Tian, Y.-X., & Fan, Z.-P. (2022). Forecasting sales using online review and search engine data: A method based on PCA–DSFOA–BPNN. International Journal of Forecasting, 38(3), 1005-1024.
Zhu, K., Li, Y., Mao, W., Li, F., & Yan, J. (2022). LSTM enhanced by dual-attention-based encoder-decoder for daily peak load forecasting. Electric Power Systems Research, 208, 107860.
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.
陳建安,(2021).以影響力為基之市場區隔趨勢預測方法與技術研發，國立成功大學製造資訊與系統研究所，碩士論文
徐浩翔,(2022). 基於重要影響因素之銷售預測方法與技術開發，國立成功大學製造資訊與系統研究所，碩士論文