冰水主機系統之預測模型及實證研究

Prediction Model of Chilled Water System and an Empirical Study

潘佳欣(Chia-Hsin Pan)

指導教授 ： 張國浩

國立清華大學/工學院/工業工程與工程管理學系/碩士(2020年)

摘要

在冰水主機系統中，外氣焓值是一個舉足輕重的外部參數，雖然無法控制但卻影響了系統中許多的相關因子，其中我們發現，以焓值50千焦耳/公斤(kJ/kg)為分界，小於50 kJ/kg與大於50 kJ/kg兩者的系統有著不同的表現。當焓值小於50 kJ/kg時，冰水系統的各項參數分布會呈現較為不規則的分布，因此傳統的線性迴歸較難以準確預測。此外，冰水主機的耗電預測系統一般皆是由廠務人員依經驗輸入相關參數，但這有可能會造成具有潛在影響力的因子被遺漏，進而影響模型的預測表現；且因系統內部因子總共多達17種，如冷卻水進水溫度、冰水出水溫度、冷凝器趨近溫度等等，難以確定哪些是最重要的因子。因此，本研究乃是針對焓值小於50 kJ/kg的冰水主機系統，先使用主成分分析(PCA)揀選最具影響力的主成分作為模型輸入變數、並藉此找出可能造成機差的因子，再使用倒傳遞類神經網路(BPN)進行建模與預測分析，期望能建立一個準確的耗電預測系統以供未來使用。

關鍵字

焓值 ； 冰水主機系統 ； 主成分分析 ； 倒傳遞類神經網路

並列摘要

In the chilled water system, the enthalpy is an important external parameter; although it is uncontrollable, it affects many related factors in the system. We found that the demarcation line of chiller water systems is an enthalpy value of 50 kJ/kg. When enthalpy is less than 50 kJ/kg, the data distribution of chilled water system’s parameters are more irregular, estimating the performance through traditional linear regression thus becomes unfeasible. Furthermore, the input parameters of prediction system of chilled water system are usually determined by experience, but this may make some potentially influential factors are ignored, having an effect on the performance of prediction system. Furthermore, the related parameters in the system have about 17 types, such as the temperature of chilled water out of chiller, the temperature of cooling water into chiller, and the approach temperature of condenser, so it is hard to determine which factor is the most important. As such, this study focuses on system under enthalpy value below 50 kJ/kg, choosing the most influential components by Principal component analysis (PCA) as input parameters, finding out factors that results in the machine difference, and further utilizing Back Propagation Neural Networks (BPN) for modeling and predictive analysis to build an accurate prediction system of chilled water system.

並列關鍵字

Enthalpy ； Chilled Water System ； Principal Component Analysis ； Back Propagation Neural Network

參考文獻

Abdi, H., & Williams, L. J. (2010). Overview principal component analysis. Wiley interdisciplinary reviews: computational statistics, 2(4), 433-459.

Alpaydin, E. (2014). Introduction to machine learning. Cambridge, England: The MIT Press.

Chang Y. C. (2005). Verification for the promotion of chiller performance audits. Journal of refrigeration and air conditioning professional engineer, 2, 61-67.

Chen, M., Challita, U., Saad, W., Yin, C., & Debbah, M. (2019). Artificial neural networks-based machine learning for wireless networks: a tutorial. IEEE communications surveys & tutorials, 21(4), 3039-3071.

Chien, C. F., Chen, Y. J., Han, Y. T., Hsieh, M. K., Lee, C. M., Shih, T., Wu, M.Y. & Yang, W. W. (2018). AI and big data analytics for wafer fab energy saving and chiller optimization to empower intelligent manufacturing. 2018 e-Manufacturing & design collaboration symposium (eMDC), Taiwan.