隨著萬物聯網技術發展，逐漸增大運算需求，由許多伺服器組成的資料中心運作時產生的散熱問題逐漸受到重視，一方面是因為需要節能，一方面是日漸被重視的碳中和之要求。目前，先進的資料中心多採用高架式伺服器機房的配置，並外加循環風扇，提供伺服器內部風扇之外所需的冷卻能量，避免因冷卻流量不足所產生的熱點現象，影響資料中心的散熱。基於此，本研究分為三個階段進行，首先透過建制伺服器機櫃散熱模擬環境，開發自動化資料蒐集系統，並以此系統進行數據的蒐集。再者，利用人工智慧的深度學習，透過資料的訓練所獲得的經驗模式，此經驗模式可以根據過去的歷史資料，預測未來伺服器的溫度變化趨勢。最後，基於前述訓練完之預測模型，設計適合的機櫃散熱控制器，並在建制完成的機櫃散熱模擬系統中進行驗證。實驗結果顯示在機櫃散熱模擬環境中，所研發的無線傳輸蒐集系統，可長時間自動蒐集伺服器溫度、功率、風扇轉速等資料，提供後續的建模訓練、驗證及測試，所建立的溫度預測模型，可依據輸入的特徵預測未來的溫度，其均方根誤差在0.29℃，而基於預測模型所設計的控制器，可將機櫃環境溫度收斂制目標值。

With the rapid development of Internet of Thing technology, the demand of computing power has been largely increased. As a result, the heat dissipation problem, generated by the operation of data centers consisting of many servers, is gradually gaining attention because of the need for energy saving and the increasingly important requirement of carbon neutrality. Currently, most advanced data centers use a high-bay server room configuration with additional circulating fans to provide the cooling air needed, in addition to the internal fans of the servers, to avoid the hot spot phenomenon caused by insufficient cooling air flow. This study is divided into three stages. First, an automated data collection system is developed, by establishing a heat dissipation simulation environment for server cabinets, to sequentially collect the necessary data. Furthermore, through data training, a deep learning of artificial intelligence is utilized to obtain an empirical model, which can predict the future inlet temperature for each server, resulting from the past historical data. Finally, based on the trained prediction model, an effective controller of cabinet cooling is designed and validated in the established simulation system of cabinet cooling. The experimental results show that the wireless transmission and data collection system can automatically collect the server temperatures, server powers, fan speeds, and other data during the period of time, and the collected data can provide subsequently modeling training, validation, and testing. Based on the prediction model, the designed controller can be used to control the ambient temperature converged to the target value.

[1] S. Liu. (2020). Edge computing device and infrastructure capital expenditure (CAPEX) worldwide from 2019 to 2028. September 13,2021.[Online].Available:https://www.statista.com/statistics/1104591/worldwide-edge-computing-capex/
[2] T. Alsop. (2019). Global server unit shipments 2016-2022, by technology.September 13, 2021 [Online].Available:https://www.statista.com/statistics/934508/server-unit-shipments-by-technology-worldwide/
[3] J. Cho and Y. Kim, "Improving energy efficiency of dedicated cooling system and its contribution towards meeting an energy -optimized data center ", 2016.
[4] ASHRAE, "Thermal Guidelines for Data Processing Environments – Expanded Data Center Classes and Usage Guidance", 2011.
[5] ASHRAE, "Data Center Networking Equipment – Issues and Best Practices", ASHRAE Technical Committee (TC), 2016.
[6] R. Lloyd and M. Rebow, "Data Driven Prediction Model (DDPM) for Server Inlet Temperature Prediction in Raised-floor Data Centers", 17th IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (ITherm), 2018.
[7] S.A. Nada and K.E. Elfeky, "Experimental investigations of thermal managements solutions in data centers buildings for different arrangements of cold aisles containments", Journal of Building Engineering, pages 41–49,2016.
[8] K. Sasakura, T. Aoki, M. Komatsu and T. Watanabe, "Rack Temperature Prediction Model Using Machine Learning after Stopping Computer Room Air Conditioner in Server Room", August 2020.
[9] J. Wan, J. Zhou and X. Gui , "Intelligent Rack-Level Cooling Management in Data Centers with Active Ventilation Tiles A Deep Reinforcement Learning Approach", IEEE Intelligent Systems, January 2021.
[10] N. Ahuja, C. W Rego, S. Ahuja, S. Zhou and S. Shrivastava , "Real Time Monitoring and Availability of Server Airflow for Efficient Data Center Cooling", 29th IEEE Semiconductor Thermal Measurement and Management Symposium, March 2013.
[11] J. Athavale, Y. Joshi, M. Yoda and W. Phelps, "Impact of Active Tiles on Data Center Flow and Temperature Distribution", 15th IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (ITherm), June 2016.
[12] 林炘泓，”GRU實現伺服器PCIE溫度估測器及散熱控制系統”，國立清華大學動力機械工程學系碩士論文，2020年。
[13] ARDUINO MEGA 2560 REV3, September 13,2021. [Online].Available: https://store.arduino.cc/usa/mega-2560-r3
[14] Zig Bee Alliance, ZigBee Specification Version, September 13,2021. [Online].Available: http://www.ZigBee.org
[15] Zig Bee Alliance, IEEE 802.15.4 standard Presenter, June 2007.
[16] Texas Instruments Incorporated, LM35 Precision Centigrade Temperature Sensors datasheet (Rev. H). September 13,2021. [Online].Available: https://www.ti.com/lit/ds/symlink/lm35.pdf?ts=1631581405483&ref_url=https%253A%252F%252Fwww.ti.com%252Fproduct%252FLM35
[17] DAQ970A/DAQ973A Data Acquisition System.September 13,2021. [Online].Available: https://www.keysight.com/us/en/products/modular/data-acquisition-daq/digital-acquisition-benchtop-system/daq970a-data-acquisition-system.html
[18] W. Chu and C. Wang, "A review on airflow management in data centers", Applied Energy Volume 240, pages 84-119, April 2019.
[19] J. W. VanGilder and R. R. Schmidt, "Airflow uniformity through perforated tiles in a raised floor data center", ASME: The American Society of Mechanical Engineers, 2005.
[20] J. F. Smith, W. A. Abdelmaksoud and H. S. Erden, "Design of simulated server racks for data center research", ASME 2011 Pacific Rim Technical Conference and Exhibition on Packaging and Integration of Electronic and Photonic Systems, 2011.
[21] Perforated Plate Ratio. September 13,2021. [Online]. Available: https://www.perforated-sheet.com/calculation/how-to-calculate-open-area.html
[22] Shanghai Euchips Industrial Co. Ltd, "The difference between leading edge and trailing edge dimming of SCR dimmer", September 13,2021. [Online]. Available : http://www.triacdimmer.cn/news/industry/qb170809.html
[23] J. Liu, F. Zhao, J. O’Reilly, A. Souarez, M. Manos, C. M. Liang and A. Terzis ,"Project Genome: Wireless Sensor Network for Data Center Cooling", The Architecture Journal, 2008.
[24] K. Shahzad and B. Oelmann, "A comparative study of in-sensor processing vs. raw data transmission using ZigBee, BLE and Wi-Fi for data intensive monitoring applications", International Symposium on Wireless Communications Systems (ISWCS), 2014.
[25] L. Wang, G. von Laszewski, F. Huang, J. Dayal, T. Frulani and G. Fox, "Task scheduling with ANN－based temperature prediction in a data center: a simulation－based study", Engineering with Computers volume 27, pages 381–391 , 2011.
[26] R. Pascanu, T. Mikolov and Y. Bengio, "On the difficulty of training recurrent neural networks", Proceedings of the 30th International Conference on Machine Learning, 2013.
[27] S. Hochreiter and J. Schmidhuber, "Long Short-Term Memory", Neural Computation, vol. 9, pp. 1735-80, December 1997.
[28] K. Cho, B. van Merrienboer, D. Bahdanau and Y. Bengio, "On the Properties of Neural Machine Translation: Encoder-Decoder Approaches", Eighth Workshop on Syntax, Semantics and Structure in Statistical Translation (SSST-8), October 2014.
[29] P. Kumar, V. Sundaralingam and Y. Joshi, "Effect of Server Load Variation on Rack Air Flow Distribution in a Raised Floor Data Center", 27th Annual IEEE Semiconductor Thermal Measurement and Management Symposium, March 2011.
[30] M. I. Tradat, H. A. Alissa, K. Nemati, S. Khalili, B. G. Sammakia, M. J. Seymour and R. Tipton," Impact of Elevated Temperature on Data Center Operation Based on Internal and External IT Instrumentation", 33rd SEMI-THERM Symposium, 2017.
[31] AD5220: Increment/Decrement Digital Potentiometer Data Sheet, September13,2021.[Online]. Available:
https://www.analog.com/en/products/ad5220.html
[32] DIGI," XBee®/XBee-PRO S2C Zigbee®,User Guide",pp.80.