- 學位論文
冰水機性能經驗模式適用性之研究
Study of the Suitability of Empirically Based Performance Models for Liquid Chillers
摘要
本研究的主要目的是評估文獻中較具代表性之冰水機性能經驗模式在不同運 轉變數下的適用性。所挑選較具代表性之性能經驗模式，包括六個黑盒模式與五 個灰盒模式；用於經驗模式訓練用的測試數據包含: 定速及變頻定流量、定速及變 頻變冰水量、以及第三公證等三類型。最佳性能經驗模式之評估，係依據模式預 測準確性、數據訓練需求(量測點數與筆數)、校正或訓練所需功夫、模式一般性、 計算需求、以及模式係數之物理釋義能力等項目進行。研究結果發現: 所有性能經 驗式中，以黑盒模式表現普遍較佳，灰盒模式次之；黑盒模式Model 2、 Model 3、 Model 4 與Model 6 的預測準確度較佳，平均CVRMSE 為0.63 %、0.64 %、0.74 ％及0.72 %.，而灰盒模式則以Model 8 預測準確度較佳，其平均CVRMSE 為 1.22 %。Model 6 雖然預測準確度佳但是數據分析上較為煩瑣，先要將測試數據分為全 載及部分負載且須要足夠之全載數據。至於ASHRAE Guideline 14(2002) 所建議的 冰水機預測模型Model 10 在所有數據的驗證結果下的預測準確性為最低，其所有 數據之平均CVRMSE =5%。所有模式中，以Model-2、Model-3 與Model-4 模式適 用性整體表現較佳。本研究所得結果， 除了適用於能源效率量測與驗證 (Measurement and Verification, M&V)之外，也可以作為自動偵測與診斷(FDD)、監 控以及建築節能分析之參考。
並列摘要
This paper evaluates the suitability of empirically based performance models for liquid chillers under different parameters. The typical empirical models proposed in the open literature including six black-box models and five gray-box models are employed in this study. The test data using in the empirical model training include constant-speed and variable-frequency driving liquid chillers with constant flow, constant-speed and variable-frequency driving liquid chillers with primary variable flow, and constant-speed driving liquid chillers with constant flow from the witness of the third party in USA. The suitability of empirically based performance models is performed on the basis of following considerations: accuracy of model prediction, requirements of data training (number of measuring points), efforts of calibration or training, generality of the model, computational requirements, and ability to physically interpret coefficient of the model. It is found that the black-box models are better than the gray-box models among all empirical performance models. Among six black-box models, Models 2, 3, 4 and 6 have better prediction accuracy, and their average CVRMSE are 0.63%, 0.64%, 0.74％ and 0.72% respectively. Among five gray-box models, Model 8 has the best prediction accuracy and its average CVRMSE is 1.22%. Model 6 has better prediction accuracy but more complicate, dividing data into full-load and part-load and requiring enough full-load data. Model 10, recommended by ASHRAE Guideline 14[3], has the worst prediction accuracy and its average CVRMSE is 5.00%. Models 2, 3 and 4 have better suitability of empirically based performance models among all models. Results can be as references of not only measurement and verification (M&V) but also iii supervisory control, automated fault detection and diagnosis (FDD) and analysis of energy saving for buildings.