本研究主要提出使用雙向長短記憶（Bi-directional Long Short-Term Memory, Bi-LSTM）神經網路建立一個的冷凍噸預測模型，使工程師在計劃的契約容量內，做出最適合的冰水主機調配，以達到降低能源消耗的效果，進一步提高能源利用效率。
本研究蒐集了某醫療院區的冰水主機監控資料及氣候資訊進行模型訓練和建立，在模型訓練之前運用皮爾森積動差相關係數方法，選擇了運轉台數、冰水回水溫度、用電需量、氣壓、海平面氣壓、氣溫、露點溫度、地溫共十一項與冷凍噸高相關的參數，並與冷凍噸結合作為訓練數據，建立一個具有高度準確性的模型。在模型比較的實驗中，本研究所建立的Bi-LSTM模型在訓練、驗證均方誤差及平均絕對誤差方面均優於其他模型，進一步證實了該模型在冷凍噸數預測方面的有效性。最後本文以實際蒐集的冷凍噸數與本研究所建立的預測模型進行實際測試，由結果得知二月份預測準確率達到95.17%，三月份預測準確率達到98.53%，四月份預測準確率達到92.95%，再次驗證所建立的預測模型具準確性與有效性，藉此結果應用於日後預測在現今與未來可能之溫度變化下所需之冷凍噸數。

This study proposes using a Bi-directional Long Short-Term Memory (BiLSTM) neural network to establish a predictive model for refrigeration capacity, enabling engineers to allocate chillers optimistically within the planned electricity consumption, aiming to reduce energy consumption and further enhance energy utilization efficiency.
This study collected monitoring data from chiller units and climate information from a medical facility to train and establish the model. Before preparing the model, Pearson's product-moment correlation coefficient method was used. Eleven parameters with high correlation to the refrigeration capacity were selected, including the number of operating units, chilled water return temperature, power consumption, air pressure, sea-level pressure, temperature, dew point temperature, and ground temperature. These parameters were combined with the refrigeration capacity as training data to establish a highly accurate model. In the experimental comparison of models, the Bi-LSTM model developed in this study outperformed other models in training, validation mean squared error, and mean absolute error. This further confirms the effectiveness of the Bi-LSTM model in predicting refrigeration capacity. Finally, the refrigeration capacity data collected was compared with the prediction model established in this study for practical testing. According to the results, the prediction accuracy for February was 95.17%, for March, it was 98.53%, and for April, it was 92.95%. These findings further validate the precision and effectiveness of the predictive model developed in this study. This result can be applied to future predictions of refrigeration capacity requirements under current and potential temperature variations.

摘要 i
ABSTRACT ii
誌謝 iv
目錄 v
圖目錄 vii
表目錄 x
第一章 緒論 1
1.1 研究背景與動機 1
1.2 文獻回顧 2
1.3 論文架構 5
第二章 冰水主機系統架構與設備維護 6
2.1 理想蒸氣壓縮冷凍循環 6
2.2 理想蒸氣壓縮冷凍循環與實際循環差異 8
2.3 冰水主機系統架構 10
2.4 系統設備維護與檢查 19
第三章 系統配置與預測模型 23
3.1 系統配置與架構 23
3.2 研究方法及流程 34
3.3 預測模型 37
3.4 超參數優化 44
第四章 冷凍噸預測模型建置與實證 46
4.1 蒐集整理數據資料 46
4.2 深度學習模型建立與訓練 58
4.3 神經網路模型比較 66
4.4 神經網路模型預測 71
第五章 結論與未來研究方向 78
5.1 結論 78
5.2 未來研究方向 79
參考文獻 80

[1] 內政部建築研究所，「綠空調實踐與應用」，p.1，2008。
[2] 楊昇達，空調冰水主機系統節能運轉之研究，國立臺灣科技大學，碩士論文，2015年。
[3] 吳書凡，冰水主機節能效益分析，國立高雄科技大學，碩士論文，2022年。
[4] 陳彥勳，蒸發冷卻模組應用R-410A分離式空調機之空氣側性能分析，國立高雄科技大學，碩士論文，2022年。
[5] 黃文男，醫學中心綜合實驗大樓空調能源管理系統，國立高雄科技大學，碩士論文，2015年。
[6] 林鈺倯，基於Modbus之SCADA系統，國立雲林科技大學，碩士論文，2021年。
[7] 邱景鴻，基於BiLSTM模型的音樂類別分析，逢甲大學，碩士論文，2022年。
[8] S. Wang, Y. Sun, J. Wang, D. Hou, L. Zhang, Y. Zhou, “Very Short-term Prediction for Wind Power Based on BiLSTM-Attention, ” 2021 IEEE Sustainable Power and Energy Conference (iSPEC), Nanjing, China, pp. 292-296, 2021.
[9] S. Siami-Namini, N. Tavakoli, A. S. Namin, “The Performance of LSTM and BiLSTM in Forecasting Time Series, ” 2019 IEEE International Conference on Big Data (Big Data), Los Angeles, CA, USA, pp. 3285-3292, 2019.
[10] Hamidreza Jahangir, Hanif Tayarani, Saleh Sadeghi Gougheri, Masoud Aliakbar Golkar, Ali Ahmadian, Ali Elkamel, “Deep Learning-Based Forecasting Approach in Smart Grids With Microclustering and Bidirectional LSTM Network,” in IEEE Transactions on Industrial Electronics, vol. 68, pp. 8298-8309, 2021.
[11] 周勁言，水冷式冰水主機維護保養週期評估之研究，國立臺北科技大學，碩士論文，2022年。
[12] 陳志鵬，以熱力學第二定律效率探討R-134a冷媒之運轉特性，國立屏東科技大學，碩士論文，2001年。
[13] 蔡奇芝，以大數據分析達成冰水主機系統節能目的，國立清華大學，碩士論文，2020年。
[14] 壓縮機工作原理動態圖，2017年每日頭條(https://kknews.cc/zhtw/news/nm8xg68.html)
[15] 台灣日立江森自控股份有限公司(https://www.jci-hitachi.tw/products/products_level3.aspx?545A5F548C5AD91F)
[16] TRANE Taiwan (https://www.trane.com/commercial/asia-pacific/tw/zh_ tw.html)
[17] 金日實業股份有限公司 (https://www.kingsunct.com.tw/tw)
[18] 洪國安，水冷式冰水主機汰換成分散式氣冷冰水主機之耗能分析，國立勤益科技大學，碩士論文，2015年。
[19] 日立離心式冰水主機 (https://www.everbrite.com.tw/catalog-detail/chiller_2/)
[20] 林鈺倯，基於Modbus之SCADA系統，國立雲林科技大學，碩士論文，2021年。
[21] 張光裕，整合Modbus與Websocket協定之聯網醫療資料採集嵌入式系統研製，國立中央大學，碩士論文，2020年。
[22] 李明鴻，具BACnet通訊協定之DDC空調終端控制器，國立臺灣大學，碩士論文，2000年。
[23] EasyIO-30P使用說明書(https://www.controlstraders.com/assets/brochure s/30P-SF45.pdf)
[24] ChatGPT (https://openai.com/blog/chatgpt)
[25] Y. J. Chew, S. Y. Ooi, K.-S. Wong, Y. H. Pang and N. Lee, “Adoption of IP truncation in a privacy-based decision tree pruning design: A case study in network intrusion detection system, ” Electronics, vol. 11, no. 5, pp. 805, Mar. 2022.
[26] R. Jin, L. Lu, J. Lee, and A. Usman, “Multi-representational convolutional neural networks for text classification,” Comput Intell, vol. 35, no. 3, 2019.
[27] A. Ponmalar and V. Dhanakoti, “An intrusion detection approach using ensemble support vector machine based chaos game optimization algorithm in big data platform”, Appl. Soft Comput., vol. 116, Feb. 2022.
[28] J. Du, K. Yang, Y. Hu and L. Jiang, “NIDS-CNNLSTM: Network Intrusion Detection Classification Model Based on Deep Learning, ” in IEEE Access, vol. 11, pp. 24808-24821, 2023.
[29] Z. Guo and M. Yang, “TFFC-RNN:A New RNN Based Approach for Bearing and Misalignment Compound Fault, ” 2022 International Power Electronics Conference (IPEC-Himeji 2022- ECCE Asia), Himeji, Japan, pp. 2504-2509, 2022.
[30] S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Comput, 1997.
[31] K. Moharm, M. Eltahan and E. Elsaadany, “Wind Speed Forecast using LSTM and Bi-LSTM Algorithms over Gabal El-Zayt Wind Farm, ” 2020 International Conference on Smart Grids and Energy Systems (SGES), Perth, Australia, pp. 922-927, 2020.
[32] Y. Zhou, S. Chen and D. Xiao, “Study on Natural Gas Price Forecasting Based on Prophet-GRU Nonlinear Combination, ” 2022 7th International Conference on Computational Intelligence and Applications (ICCIA), Nanjing, China, pp. 118-122, 2022.
[33] P. Bahad, P. Saxena, and R. Kamal, “Fake News Detection using Bi-directional LSTM-Recurrent Neural Network,” 2020.
[34] B. J. T and D. S. Misbha, “Detection of Attacks using Attention-based Conv-LSTM and Bi-LSTM in Industrial Internet of Things, ” 2022 International Conference on Automation, Computing and Renewable Systems (ICACRS), Pudukkottai, India, pp. 402-407, 2022.
[35] P. N. Thanh, M. -Y. Cho, C. -L. Chang and M. -J. Chen, “Short-Term Three-Phase Load Prediction With Advanced Metering Infrastructure Data in Smart Solar Microgrid Based Convolution Neural Network Bidirectional Gated Recurrent Unit, ” in IEEE Access, vol. 10, pp. 68686-68699, 2022.
[36] J. Bergstra and Y. Bengio, “Random search for hyper-parameter optimization”, J. Mach. Learn. Res., vol. 13, no. 2, pp. 281-305, 2012.
[37] F. Hutter, H. H. Hoos and K. Leyton-Brown, “Sequential model-based optimization for general algorithm configuration, ” Proc. Int. Conf. Learn. Intell. Optim., pp. 507-523, 2011.
[38] J. Snoek, H. Larochelle and R. P. Adams, “Practical Bayesian optimization of machine learning algorithms, ” Proc. Adv. Neural Inf. Process. Syst., vol. 25, pp. 2951-2959, 2012.
[39] 鍾松融，智慧型配電饋線絕緣礙子洩漏電流監測，國立高雄科技大學，碩士論文，2022年。
[40] 農業氣象觀測網監測系統 (https://agr.cwb.gov.tw/)
[41] J. Liu, Y. Zhang and Q. Zhao, “Adaptive ViBe Algorithm Based on Pearson Correlation Coefficient, ” 2019 Chinese Automation Congress (CAC), Hangzhou, China, pp. 4885-4889, 2019.
[42] A. I. Ivanov, S. E. Vyatchanin and P. S. Lozhnikov, “Comparable estimation of network power for chi-squared Pearson functional networks and Bayes hyperbolic functional networks while processing biometric data, ” 2017 International Siberian Conference on Control and Communications (SIBCON), Astana, Kazakhstan, pp. 1-3, 2017.
[43] X. Zhi, S. Yuexin, M. Jin, Z. Lujie and D. Zijian, “Research on the Pearson correlation coefficient evaluation method of analog signal in the process of unit peak load regulation, ” 2017 13th IEEE International Conference on Electronic Measurement & Instruments (ICEMI), Yangzhou, China, pp. 522-527, 2017.
[44] Y. Sai, R. Jinxia and L. Zhongxia, “Learning of Neural Networks Based on Weighted Mean Squares Error Function, ” 2009 Second International Symposium on Computational Intelligence and Design, Changsha, China, pp. 241-244, 2009.