Application of Adaptive Neuro-Fuzzy Inference System to Construct and Analyze Defroster Model for Refrigerator

[[abstract]]我們都知道冷凍庫在商業上的運用非常廣泛，只要開機後都是24小時運轉，在運轉上也是相當的耗費成本及能源，其中運轉過程中除霜所使用的電熱器耗電也是相當的大，因此如何合理的縮短除霜的時間，以便節能但又要能兼顧除霜完全，便是值得探索的問題。 在論文中採用實驗的方法進行冷凍庫除霜預測模型的研究，主要蒐集的實驗參數是以冷凍庫的除霜頻率、運轉溫度、除霜時間三個因子作為輸入因子，而除霜排出的水量則是輸出因子。本文首先以田口實驗的方法找出三個因子中，影響預測模型較為顯著的輸入因子並增加其水準數，以提升模型預測準確率，然後可將建構預測模型的實驗分為，訓練實驗及驗證實驗，可得到的參數分別為訓練參數及驗證參數兩種。之後使用訓練參數輸入適應性類神經模糊推論系統，來建構冷凍庫除霜預測模型，在訓練時選擇不同的歸屬函數數目及不同種類的歸屬函數，藉以找出對預測模型準確度最佳的組合，再使用驗證參數來檢驗預測模型的準確度。在這個實驗中，透過比較後發現以3x5x3歸屬函數數目搭配三角形歸屬函數可以得較佳的準確度。便以此設定繪出除霜排出水量分佈圖，並探討除霜時間的效率，以提出合理且較為節省能源的除霜時間設定，並再實驗36組數據與預測出的除霜排出水量做比較，結果顯示建構出的冷凍庫除霜預測模型，可作為省能設定的參考。[[abstract]]We all know the refrigerator which is very extensive application in commercially At after starting all refrigerator run in twenty four hours Refrigerator need much cost and energy during running which the heater of defroster used energy too very large Therefore it is worth exploring the issue how to reasonably reduce defroster time to save-energy and completely defrost In the thesis we adopt the experimental way to investigate of the defroster model for refrigerator In this experiment we use the defroster frequency the operating temperature and the defroster time as input factor and the defroster discharge water as the output factor We firstly use the Taguchi method Then we can find out the significant factor for predicting model Following that we increase the levels of the significant factor to improve the learning effect of the predicting model Then we can plan the training experiments and the examining experiments for the predicting model respectively Therefore we get the training data and the examining data respectively The training data are applied to construct the defroster model for refrigerator by the adaptive neuro-fuzzy inference system（ANFIS） In the training to select different quantity of membership function and different types of membership function to find out the best combination for accuracy of prediction model In addition the examining data were employed to examine the accuracy for the predicting model In the experimented we find that the 3x5x3 quantity of membership function and triangle membership function which have the best of accuracy Then we can plot defroster discharge water distribution diagram which explore the efficiency of defroster time We propose reasonable and save-energy of defroster time setting In addition we compare thirty six sets by way of experimental with predicting defroster discharge water The results reveal that the defroster model can be used save-energy setting referenc