基於EM-GAM-AR之校園建築物之電耗預測模型｜國立臺灣科技大學博碩士論文系統

簡易檢索 / 詳目顯示

回結果列表

研究生：	何宗緯 Tsung-Wei Ho
論文名稱：	基於EM-GAM-AR之校園建築物之電耗預測模型 EM-GAM-AR based Prediction Models of Power Consumption for Campus Buildings
指導教授：	鍾聖倫 Sheng-Luen Chung
口試委員:	陸敬互 Ching-Hu Lu 郭政謙 Cheng-Chien Kuo 周瑞生 Jui-Sheng Chou 林希偉 Shi-Woei Lin
學位類別：	碩士 Master
系所名稱：	電資學院 - 電機工程系 Department of Electrical Engineering
論文出版年：	2016
畢業學年度：	104
語文別：	英文
論文頁數：	48
中文關鍵詞：	電耗估測、EM(Expectation-maximization) 、AR(Autoregressive)模型、GAM(GeneralizedAdditiveModel) 、統計自助法(Bootstrap) 、重抽樣(Resampling)
外文關鍵詞：	Expectation Maximization (EM) clustering, Autoregressive (AR), Bootstrap resampling
相關次數：	點閱：169 下載：2
分享至:	分享至facebook 分享至twitter

查詢本校圖書館目錄查詢臺灣博碩士論文知識加值系統勘誤回報

上一筆

校園建築物的電耗模型，由於學期寒暑假、期中末考試等不同的用電型態，而比商用建築物或住家的電耗模型來得複雜。精確的電耗模型可當作節能與即時異常偵測的依據。本研究的目的是由校園建築物歷史電耗資料建構能反應一年中不同月、日、小時以及不同校園作息如寒暑假與學期中特殊假期或是大考等行事曆型態的電耗預測模型。在電耗預測值的分析上，我們採用 EM-GAM-AR 的兩段迴歸分析方法：首先，Expectation-Maximization (EM) 會針對學校不同日子的電耗圖形進行分群，並取得最顯著的日子類別，接著使用 GAM (Generalized Additive Model)模型用來塑模如：月、日、時、以及不同行事曆型態的外部變數，以獲得在此外部條件下的基準預測。而對於預測範圍的決定，相較於盲目假設前述 GAM 模型所遺留誤差值為常態分佈，我們利用 AR (Autoregressive) 模型對 GAM 模型所遺留誤差值解釋其時序特性，再利用 AR 遺留的殘差值進行統計自助法 (Bootstrap resampling)，以反應歷史電耗的統計特性。針對台科大研究大樓的 2014 的歷史電耗資料，我們在 R programming上透過 Deviance Explained (DE)的驗證來決定 GAM 模型並實現上述 GAM-AR 配加 Bootstrap 統計建模的程序，得到校園電耗預測模型。在此基礎上，我們根據 2015年實際外部變數的條件，由上述模型進行預測，並且將之計算 DE、mean absolutepercentage error (MAPE)、correlation 以及 coverage absolute error (CAE) 此四項指標並進一步與其他方法進行比較來驗證我們模型的預測能力。在模型訓練品質上，我們的模型在 DE 值取得 92.8%的結果，相比於一樣在電耗預測使用 GAM 模型的方法中，我們的結果顯示對訓練資料的擬合狀況良好；此外，我們的模型在實際預測上的MAPE 值接近 10%，相較於許多現行方法而言有更好的精準度，且更進一步計算 CAE也僅有 5.74%，此兩項結果亦驗證我們模型在實際預測上有良好的預測能力。總而言之，根據歷史的電耗資料，本論文所提供的方法可以建構在不同外部變數條件下，估測建築物在不同月、日、時下的電能損耗值，可當作後續建築物能源管理方針以及即時異常偵測的依據。

For campus buildings, power consumption patterns vary significantly with respect to different periods such as summer vacation, spring break, etc. Therefore, analytic work for campus buildings is relatively complicated compared to commercial or residential buildings. This study aims to devise a generic process for constructing prediction models of power consumption for campus buildings. Our objective is to fit and then to forecast both mean values and intervals of the power consumption based on the information of time indices, day types and weather conditions. To do so, we propose an EM-GAM-AR process for the underlying modeling. ExpectationMaximization (EM) extracts the most significant day-type variables, which are then applied to a Generalized Additive Model (GAM). GAM analyzes the impact of exogenous variables of time, user behavior and weather conditions on the baseline or mean of the power consumption. Subsequently, an Autoregressive (AR) model is applied to describe the time-varying effect of the GAM residuals, which are the differences between the observed data and the values given by GAM. The sequence of the resultant AR residuals is critical for the estimation of prediction intervals. Instead of assuming these AR residuals as normal, we adopt a nonparametric bootstrapping process to preserve its statistics property for distribution characteristics. We use Research Building (RB) at NTUST as our target building in deriving power consumption patterns based on recorded data from 2014. The patterns obtained are used for data fitting of 2014 and prediction of 2015, and the results are compared to other methods using different criteria such as deviance explained (DE), mean absolute percentage error (MAPE), correlation and coverage absolute error (CAE). Briefly, our model has a rather high DE of 92.8% and performs a compelling MAPE around 10% and a CAE of 5.74% in prediction. These results prove that our model performs better in comparison with existing methods.

摘要........................................................................................................ I
Abstract....................................................................................................III
Acknowledgement ............................................................................................V
List of Figures.............................................................................................IX
List of Tables .............................................................................................X
Chapter I. Introduction.....................................................................................1
1.1. Prediction of power supply-demand .....................................................................1
1.2. Contribution...........................................................................................2
1.3. Paper Organization.....................................................................................3
Chapter II. Power Consumption Models for Campus Buildings...................................................4
2.1. Data Set from a Practical Example .....................................................................4
2.2. Criteria for Model Selection ..........................................................................6
2.3. Power Consumption Model................................................................................8
2.4. Literature Survey......................................................................................9
Chapter III. Methodology....................................................................................11
3.1. Overview of data process flow .........................................................................11
3.2. EM for exogenous variable selection....................................................................13
3.3. GAM for modeling energy consumption baseline ..........................................................16
3.4. AR-Bootstrap process for interval prediction...........................................................19
3.5. Anomaly Detection .....................................................................................24
Chapter IV. Experiment Result...............................................................................25
4.1. Visualization of GAM exogenous variables...............................................................26
4.2. Training and In-sample Test ...........................................................................28
4.3. Testing Results........................................................................................28
4.4. Online Learning Results................................................................................29
4.5. Application of Anomaly Detection ......................................................................32
Chapter V. Comparison to Related Work.......................................................................34
5.1. Comparison of Data Fit ................................................................................34
5.2. Comparison of Mean Prediction Accuracy.................................................................35
5.3. Comparison of Interval Prediction Accuracy ............................................................37
Chapter VI. Conclusion .....................................................................................39
6.1. Conclusion ............................................................................................39
6.2. Future Work ...........................................................................................40
Appendix ...................................................................................................41
Reference...................................................................................................46

                                

[1] M. H. Chung and E. K. Rhee, "Potential opportunities for energy conservation in existing buildings on university campus: A field survey in Korea," Energy and Buildings, vol. 78, pp. 176-182, 2014.
[2] H.-x. Zhao and F. Magoulès, "A review on the prediction of building energy consumption," Renewable and Sustainable Energy Reviews, vol. 16, pp.3586-3592, 2012.
[3] N. Sharma, A. Bajpai, and M. R. Litoriya, "Comparison the various clustering algorithms of weka tools," facilities, vol. 4, 2012.
[4] T. K. Moon, "The expectation-maximization algorithm," IEEE Signal processing magazine, vol. 13, pp. 47-60, 1996.
[5] P. M. Dixon, "Bootstrap resampling," Encyclopedia of environmetrics, 2002.
[6] S. Wood, Generalized additive models: an introduction with R: CRC press, 2006.
[7] D. Mayer and D. Butler, "Statistical validation," Ecological modelling, vol. 68, pp. 21-32, 1993.
[8] R. A. Johnson and D. W. Wichern, Applied multivariate statistical analysis: Pearson Education Limited Essex, 2014.
[9] T. K. Wijaya, M. Sinn, and B. Chen, "Forecasting uncertainty in electricity demand," in AAAI-15 Workshop on Computational Sustainability, 2015.
[10] A. I. Dounis, "Artificial intelligence for energy conservation in buildings," Advances in Building Energy Research, vol. 4, pp. 267-299, 2010.
[11] A. Ahmad, M. Hassan, M. Abdullah, H. Rahman, F. Hussin, H. Abdullah, et al., "A review on applications of ANN and SVM for building electrical energy consumption forecasting," Renewable and Sustainable Energy Reviews, vol. 33, pp. 102-109, 2014.
[12] H. Zhao and F. Magoulès, "Parallel support vector machines applied to the prediction of multiple buildings energy consumption," Journal of Algorithms & Computational Technology, vol. 4, pp. 231-249, 2010.
[13] K. K. Sumer, O. Goktas, and A. Hepsag, "The application of seasonal latent variable in forecasting electricity demand as an alternative method," Energy policy, vol. 37, pp. 1317-1322, 2009.
[14] Y. Wen and W. Burke, "Real-Time Dynamic House Thermal Model Identification
for Predicting HVAC Energy Consumption," Green Technologies Conference, 2013
IEEE, pp. 367-372, 2013.
[15] J.-S. Chou and A. S. Telaga, "Real-time detection of anomalous power consumption," Renewable and Sustainable Energy Reviews, vol. 33, pp. 400-411, 2014.
[16] T. J. Hastie and R. J. Tibshirani, Generalized additive models vol. 43: CRC Press, 1990.
[17] A. Ba, M. Sinn, Y. Goude, and P. Pompey, "Adaptive learning of smoothing functions: application to electricity load forecasting," Advances in neural information processing systems, pp. 2510-2518, 2012.
[18] J. Ploennigs, B. Chen, A. Schumann, and N. Brady, "Exploiting generalized additive models for diagnosing abnormal energy use in buildings," Proceedings of the 5th ACM Workshop on Embedded Systems For Energy-Efficient Buildings, pp. 1-8, 2013.
[19] B. Chen, M. Sinn, J. Ploennigs, and A. Schumann, "Statistical anomaly detection in mean and variation of energy consumption," 2014 22nd International Conference on Pattern Recognition (ICPR), pp. 3570-3575, 2014.
[20] S. Wood and M. S. Wood, "The mgcv package," www. r-project. org, 2007.
[21] P. J. Brockwell and R. A. Davis, Introduction to time series and forecasting: Springer
Science & Business Media, 2006.
[22] A. Trapletti and K. Hornik, "tseries: Time series analysis and computational finance. R package version 0.10-25," ed: URL: http://CRAN. R-project. org/package= tseries, 2011.
[23] Y. Zhu, "Introducing Google Chart Tools and Google Maps API in data visualization courses," IEEE computer graphics and applications, vol. 32, p. 6, 2012.
[24] B. Lantz, Machine learning with R: Packt Publishing Ltd, 2013.

簡易檢索 / 詳目顯示

相關論文