- 學位論文
直接空調負載控制曲線自動化分類系統
Automatic Classification System of Air-Conditioner Direct Load Control Curves
摘要
隨著經濟迅速成長及人民生活水準提高，用電負載尤其是夏季尖峰時段冷氣負載乃急遽增加。為有效利用現有的發電設備及緩和電源開發壓力，評估與瞭解各項負載管理措施實施成效，即成為電力公司改善相關負載管理策略與訂定未來負載管理目標值的重要參考依據。 針對用戶直接負載控制實施成效評估，本文建構一直接空調負載控制曲線自動化分類系統，以辨識實施情況良好與不彰之負載。為達直接空調負載控制曲線分類的目標，有效的負載特徵擷取係首要工作。本研究利用自我迴歸移動平均(Autoregressive moving average, ARMA)分析之特徵擷取方法，使用時間序列之資料特性，找尋資料彼此前後關係，並利用各個誤差項建構時間序列資料之模式，進而擷取各負載曲線之重要特徵。為期獲得合適之模式架構，再根據Akaike資訊標準指標(Akaike information criterion, AIC)評估不同之自我迴歸移動平均模式，以做為負載曲線分類依據之特徵。 本文結合基因演算法與最佳化k 階均值演算法建構直接空調負載控制曲線之自動化分類系統。在方法上係利用基因演算法與k 階均值演算法將輸入資料與群集中心不斷進行演化與調整，進而產生可以充分反映輸入資料分佈型態之最佳群集中心。透過建構之自動化分類系統，對各參與用戶直接負載控制措施實施後所得負載曲線，分為符合控制類型與不符合控制類型兩類，做為進一步進行績效評估的基礎。 本研究引用民國89 年與91 年台電夏季實施中央空調直接負載控制措施資料做為建構自動化分類系統訓練與測試資料，用以驗證本文所提之可行性。其中，訓練與測試資料包括中央空調週期性暫停用電、中央空調遙控降載週期性暫停用電與儲冷式空調系統。同時，本文利用Matlab 軟體撰寫自動化分類系統之計算機程式，所得結果亦與自我組織映射網路與倒傳遞類神經網路方法進行比較，以驗證本文所提自動化分類系統於實測資料上的準確性與有效性。經由測試結果顯示不論對系統已知訓練資料或未知之測試資料，皆可準確分類由不同直接負載控制措施實際量測紀錄所得之負載曲線。
並列摘要
Owing to rapid growth of economy and living standards on Taiwan, electrical loads have increased drastically, especially attributed to the air-conditioner loads in summer. To provide solutions to efficient utilization of existing power apparatus and reduction of the pressure of installing new one, performance evaluation of demand side management (DSM) strategies would be an important approach to further improving the related strategies and determining the targets for future DSM programs. Aiming at performance evaluation of the air-conditioning DSM programs, this thesis establishes an automatic classification system of direct load control (DLC) curves. Through the system, identified is either the DLC curve that complies with the control model of a specific DLC program or the one that does not. However, to achieve purpose of precise DLC curve assification, significant features of load curves must be first extracted. In the thesis, ignificant features of the DLC curves are extracted by using the autoregressive moving average (ARMA) analysis method. The ARMA method is implemented by establishing the relation among the data for the DLC curves, regarded as time series. Then, the Akaike information criterion (AIC) is employed to determine the appropriate orders of the ARMA models. The extracted features from the ARMA models are taken as the basis of the classification system of the DLC curves. On the basis of the features extracted, the thesis presents an efficient approach to clustering the DLC curves by using the genetic k-means algorithm (GKA) in the classification system. The centroids of clusters for DLC curves are varied in the evolution process guides by the combined scheme of the genetic algorithm with k-means approach. The cluster centroid regulating based GKA creates the best cluster centroids, which may represent the distribution of the template data sets. The results of clustering centroids can then be used to classify the DLC curves into the one complying with the control model and the other one not complying. To establish and verify the classification system, employed is the practical DLC database in Years 1999 and 2001 obtained from Taiwan Power Company (TPC) as training and testing data. The database includes the data for central air conditioning duty cycling control, paging system in central air conditioner duty cycling control, and ice storage central air conditioning systems. To demonstrate the efficiency and effectiveness of the proposed approach for the practical data evaluations, the thesis compares the results obtained from the proposed approach with the those using the self-organizing maps (SOM), and the artificial neural networks methods in the classification system. The DLC classification system is implemented through Matlab programming tool in the thesis. The testing results show that a very low degree of misclassification for the TPC DLC database user has been obtained for both known training data and unknown testing data for the classification system. The practical load curves of different DLC programs can therefore, be accurately classified for further performance evaluation.