應用預測資料與不確定性之冰水主機群負載分配最佳化

Abstract

由於台灣位於亞熱帶的地區，屬於炎熱且潮濕的氣候，中央空調是大多數建築維持室內熱舒適度的必要工具。在一般使用下，空調耗電量會占整棟建築的40%，在尖峰時刻更會達到50% 。然而，當全球暖化效應下，全球平均溫度不斷上升，空調耗電量將會佔整棟建築的更大比例，減少空調的耗電量不僅能減緩溫室效應、降低發電廠負擔，更能減少電費支出。前人研究指出，不當操作冰水主機將會使整體建築能源消耗大幅提升，因此本研究期望透過基因演算法解決冰水主機負載最佳化(optimal chiller loading, OCL)與冰水主機時序控制最佳化 (optimal chiller sequencing, OCS)，同時符合現實操作的合理性，以達到比傳統人工操作更省電的結果。

本研究選取辦公大樓做為研究對象，主要針對改善負載分配效率、開關機次數與限制最大連續運轉時數提出新的方法。第一步取得歷史資料做為訓練值對未來空調需求量做預測，第二步將預測值設定為基因演算法的目標值，取得最佳的冰水主機時序控制(Sequencing Control):是指在多個主機間決定開關機，第三步將前一步的結果做為限制條件，以實際的歷史資料做為實際值設定為基因演算法的目標值，模擬實際發生的情形作為驗證。

研究結果顯示與傳統人工操作相比，採用基因演算法搭配預測的空調需求量做控制並考慮不確定性，能在年耗電量降低的情況，不額外增加切換次數與機械損耗，且易於應用於實際案例。與歷史資料相比，年耗電量降低約15%、切換次數減少約16%、COP提升約14%。

Due to its location in the subtropical region, Taiwan experiences a hot and humid climate, making central air conditioning a necessary tool for maintaining indoor thermal comfort in most buildings. Under normal usage, air conditioning accounts for 40% of the total electricity consumption of a building, reaching up to 50% during peak hours. However, with the ongoing global warming effect and the continuous increase in global average temperatures, air conditioning's electricity consumption will occupy a larger proportion of the building's total energy usage. Reducing air conditioning's power consumption not only helps mitigate the greenhouse effect and lessen the burden on power plants but also leads to reduced electricity expenses. Previous studies have indicated that improper operation of chiller plants can significantly increase the overall energy consumption of a building. Therefore, this study endeavors to utilize genetic algorithm to tackle the complex optimization challenges associated with optimal chiller loading (OCL) and optimal chiller sequencing (OCS) in order to achieve both practicality in real-world operations and greater energy efficiency compared to traditional manual operations.

Within this study, office building serve as the focal point for research, representing a crucial domain where efficient energy utilization is paramount. It primarily focuses on improving load distribution efficiency, reducing the number of on-off cycles, and limiting the maximum continuous operating hours by proposing a new method. The first step involves obtaining historical data as training values to predict future air conditioning demand. The second step sets the predicted values as the target values for the genetic algorithm, aiming to achieve the optimal sequencing control of the chiller plants, which refers to determining the on-off status among multiple units. The third step uses the results from the previous step as constraints and sets the actual historical data as the target values for the genetic algorithm, simulating real-life scenarios for validation.

The research results demonstrate that compared to traditional manual operation, using genetic algorithms in combination with predicted air conditioning demand for control while considering uncertainty can reduce annual electricity consumption without increasing additional switching times and mechanical losses. Moreover, it is easily applicable to practical cases. Compared to historical data, the annual electricity consumption is reduced by approximately 15%, the number of switching times is reduced by around 16%, and the Coefficient of Performance(COP) is increased by approximately 14%.