模糊進化規劃法於空調負載預測之應用

Application of Fuzzy Evolutionary Programming for Load Forecasting on HVAC

李信蔚(Shin-Way Lee)

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

在中央空調系統中，預測空調負載對於節能是非常重要的一環，若是能預先預測出空調負載，則對於空調主機調度、協調、排程和系統安全性的後續規劃即有明確的目標值可以依循設計規劃。應用人工智慧系統的方法是為了能更有效地得到短期負載預測的結果。人工智慧應用在短期負載預測包含有類神經網路、專家系統及模糊理論。本論文使用模糊理論預測負載，因為模糊理論使用在控制方面，並不需要了解精確的數學模型，它主要是模仿人的經驗控制而不是以控制模型為對象，並可解決數學模式無法精確掌控的系統，但若藉由人工模糊歸屬度的定義，則會造成不夠精準的輸出結果。
為了改善此問題，本論文提出進化規劃法應用於搜尋最佳的歸屬函數，根據過去的天氣狀況及負載，使用最少輸入變數的模式預測空調負載。進化規劃是一種利用多點搜尋的最佳化演算法，所以較能避免落入局部最佳解中，其不但省去了基因演算法的編碼程序，並只藉突變搜尋全域最佳解，因此能比基因演算法計算時間更為縮短，以更快速度獲得全域最佳解。為了驗證其有效性，本論文使用的測試資料是以新竹某半導體廠實際資料作為應用實例，並和傳統模糊預測作比較。預測所得的結果顯示本法均比傳統模糊預測更為精準，非常適合空調系統實際操作上的應用。

關鍵字

冰水主機 ； 進化規劃法 ； 模糊理論 ； 負載預測

並列摘要

In the air-conditioning system, Load Forecasting on HVAC is a key factor for energy saving. If the Load Forecasting on HVAC is possible, there will be a distinct target value as ground for further programming on Dispatch, Coordination, Schedule and System Security of chiller. The reason why applying artificial intelligence system is to get short-term load forecasting result more effectively. Artificial intelligence systems which can be applied on short-term load forecasting includes neural network, expert systems and fuzzy theory. Fuzzy theory is adopted in this study because it’s not necessary for researchers to understand accurate mathematical models. Fuzzy theory mainly simulates human control experience rather than control models. It can also deal with systems which can not be precisely controlled by mathematical models. However, defining fuzzy theory through artificial fuzzy membership functions results in inaccurate outputs.
In order to solve the problem, Evolutionary Programming (EP) is brought up in this study for searching optimal membership functions, using the model with fewest variables input to forecast load of HVAC, based on past weather statuses and loads. EP is a multi-point researching optimal algorithm, freer from local optimal. Leaving out the encoding process of Genetic algorithm and searching global optimal through mutations only, EP calculates faster than Genetic algorithm and gets global optimal in a shorter time. To examine the effectiveness, real data from a semiconductor plant in Hsin-chu is used as test material and studying case in this study, and is comparing it to Fuzzy load forecasting method. The forecasting result shows that this method is more accurate than Fuzzy load forecasting method, and suitable for applying on real-world operation of air-conditioning

並列關鍵字

Chiller ； Evolutionary Programming ； fuzzy theory ； load forecasting

參考文獻

[10]莊虔毅，「以基因演算法求解含凹凸特性曲線之冰水主機最佳負載分配」，台北科技大學碩士論文，2005。

[19]林豐澤，演化式計算上篇：演化式演算法的三種理論模式，智慧科技與應用統計學報，3卷1期，2005。

[21]楊宏澤、陳世杰、林金山，「自我組織式模糊系統作短期電力負載預測」，中原學報，Vol.28 No.1，頁73-86，2000。

[4]S. J. Kiartzis, A. G. Bakirtzis, V. Petridis “Short-term load forecasting using neural networks,”Electric Power Systems Research, Volume 33, Issue 1, April 1995, Pages 1-6

[5]林良興，應用適應性類神經模糊推論系統於冰水主機耗電量之預測，碩士論文，台北科技大學，2007。