結合類神經網路與進化策略演算法於空調冰水系統最佳化運轉之研究

Application of Neural Network and Evolution Strategy on Optimal Chilled Water System Operation of HVAC

陳證中(Cheng-Cjung Chen)

指導教授 ： 張永宗

國立臺北科技大學/機電學院/能源與冷凍空調工程系碩士班/碩士(2010年)

摘要

在國內使用空調的比例是日益漸增、是無可取代的必需品，尤其以半導體廠或辦公大樓的耗能情況更加嚴重；生活機能的便利，對於家家戶戶具備冷氣的供需是必要的，在文明的進步下伴隨而來是暖化議題的重視性與衝擊性。以中央空調系統來說，冰水主機是整個系統最耗能的部分，其出水溫度設定會直接影響主機本身消耗功率及其性能特性，談到節能措施對於空調系統整體性的優先考量以主機的耗能為重點。在過去所知道的觀念中，傳統控制手法以固定冰水出水溫來提供室內負載的需求，這將會造成無謂的耗電量浪費，而本文運用類神經-進化策略最佳化控制進行節能控制並與傳統控制手法的耗電量相互比較。
本研究以實驗室設備為主，利用冰水溫度設定來針對負載的變化做調度並運用類神經網路對於空調冰水系統的冰水主機、空調箱、送風機、區域泵等耗電量元件所收集到的數據進行建模訓練，與實際耗電量相比，所得到的預測結果相當準確。以冰水主機耗電量、送風機耗電量和冰水泵浦耗電量加總值作為一目標參考值，在滿足相關條件的負載需求下，結合進化策略方法求取三項參數的最佳解，來控制冰水主機的參數值設定以達到主機最佳化運轉，以減少主機多餘的能源消耗，並達到節能效益的成效。

關鍵字

類神經網路 ； 進化策略演算法 ； 冰水主機 ； 最佳化運轉

並列摘要

The use of air conditioning systems in Taiwan has grown extensively in recent years, and has become an irreplaceable necessity for many activities However, increased usage of air conditioners also means elevated power consumption, and the problem is particularly serious at establishments such as semi-conductor foundries and office buildings. The convenience and comfort that characterize modern living render air conditioning systems an inseparable appliance for countless households. On the other hand, the advancement of civilization comes with the tradeoff of problems and impact of global warming. Taking a central air conditioning system for example; the chiller is the key component of the system that demands the most power, and the configuration of chilled water temperature directly affects the power consumption of the chiller and its performance. When it comes to energy saving measures for air conditioning systems as a whole, the priority lies in achieving optimal efficiency for the chiller. Conventional concepts of air conditioning systems involve the application of traditional control techniques to achieve the required indoor cooling load by ensuring constant water supply temperature. However, such techniques have resulted in unnecessary power consumption. In this study, the author has proposed an optimized control technique derived from neural network – evolution strategy to achieve energy saving control. In addition, the research will also present a comparison of the featured technique against traditional control techniques in terms of power consumption performance.
For the purpose of this study, the author has chosen to focus on laboratory equipment by manipulating chilled water temperature settings to moderate the cooling load while utilizing a neural network for model training from data collected from various power consuming components such as the chiller, air handling unit, fan, secondary pump in a chilled water system. Compared to the actual power consumption, the predictions made from the data turned out to be fairly accurate. The power consumption of the chiller, fan and chilled water pump have been added up as the objective function, and under the premise of satisfying all constraints to achieve the required cooling load, the author will attempt to derive the best solution for the three parameters using the evolution strategy to control the parameter settings for the chiller to achieve optimal system operation. This minimizes unnecessary power consumption by the chiller and accomplishes the goal of energy conservation

並列關鍵字

Neural Network ； Evolution Strategy ； Chiller ； Optimal Operation

參考文獻

[6]D. E. Rice, “A suggested energy-savings evalution method for AC djustable speed drive applications”, IEEE Transactions. on Industry Applications Vol. 24, pp. 1107-1117, 1988。

[12]Y.W. Wang, W.J Cai, Y.C Soh, S.J. Li, Lu Lu, Lihua Xie, “A simplified modeling of cooling coils for control and optimization of HVAC systems”, Energy Conversion and Management, Vol. 45, pp. 2915-2930, 2004。

[13]Lu Lu, W.J Cai, Lihua Xie, Shujiang Li, Y.C Soh, “HVAC system optimization-in building section”, Energy and Buildings, Vol. 37, pp. 11-22, 2005。

[15]陳建進，應用進化策略演算法於冰水主機負載分配之最佳化，碩士論文，台北科技大學，2006。

[16]Yung-Chung Chang,"Sequencing of chillers by estimating chiller power consumption using artificial neural networks," Building and Environment 42:180–188, 2007。