飲水機已成為基本必需品，在台灣的私人物業和公共場所已成為常見的景象。台灣典型的飲水機提供3種水溫：熱水，暖水和冷水。儘管發現它對用戶來說很方便，但是根據需要提供各種溫度的水，使飲用水分配器成為台灣第五大耗電的家用電器。高能耗來自加熱和冷卻過程，以保持水箱中的水溫。當在某些時間段沒有飲用水需求時，重複的冷卻和加熱過程可能產生能量浪費。
本研究提出了飲水機的數據驅動控制策略。本研究中討論的控制策略是基於從安裝在分配器上的傳感器和智能儀表收集的可用數據以及相關的外部數據構建的。根據現有數據，建立了五種預測模型：電力使用，熱耗狀態，冷消耗狀態，熱水溫度和冷水溫度。利用多層感知器（MLP）方法建立電力使用和水溫預測模型，採用隨機森林技術建立用水狀態預測模型。這五種預測模型用於構建優化模型，以通過調度加熱和冷卻過程考慮水溫和水消耗限制來最小化電力使用。使用粒子群優化（PSO）方法求解優化模型。
已經進行了一個案例研究來評估所提出的模型的性能。 2019年1月14日關於案例研究中使用的高峰時段的數據。結果表明，在保持冷水滿意度​​為100％，熱水滿意度為95.83％的同時，可以實現6.3％的節能效果。然而，由於需要改進預測模型，因此該結果可能不准確。

Drinking water dispenser has become a basic necessity and has been a common sight in private properties and public places in Taiwan. Typical drinking water dispenser in Taiwan provides 3 levels of water temperature: hot, warm, and cold. Even though it is found to be convenient for user, providing various temperature of water that can be readily available on demand makes drinking water dispenser the 5th most electricity-consuming household appliances in Taiwan. High energy consumption comes from the process of heating and cooling to maintain water temperature in the tank. Repeated processes of cooling and heating may generate energy waste when there is no demand of drinking water at some periods of time.
This study proposes data-driven control strategies for drinking water dispenser. Control strategies discussed in this study are constructed based on available data collected from sensors and smart meter installed on the dispenser along with related external data. Based on available data, five predictions model are built: electricity power usage, hot consumption status, cold consumption status, hot water temperature, and cold water temperature. Electricity power usage and water temperature prediction models are constructed using Multi-Layer Perceptron (MLP) method while water consumption status prediction model is built using random forest technique. These five prediction models are used to construct an optimization model to minimize electricity power usage with considering water temperature and water consumption constraints by scheduling heating and cooling processes. The optimization model is solved using Particle Swarm Optimization (PSO) method.
A case study has been conducted to evaluate the performance of the proposed model. Data on 14 January 2019 on peak hours used in the case study. The results show 6.3% energy-savings could be achieved while still maintaining satisfaction level of cold water at 100% and hot water satisfaction level at 95.83%. However, this result may not be accurate since prediction models need to be improved.

1. Chen, Y.T., 2017, The Factors Affecting Electricity Consumption and Consumption Characteristics in The Residential Sector-A Case Example in Taiwan, Sustainability, 1484-1499.
2. Naug, A, and Biswas, G., 2018, Data Driven Methods for Energy Reduction in Large Buildings, IEEE International Conference on Smart Computing, 133-138
3. Kusiak, A, Xu, G., and Zhang, Z., 2014, Minimization of Energy Consumption in HVAC Systems with Data-driven Models and an Interior-point Method, Energy Conversion and Management, 85, 146-153.
4. He, X., Zhang, Z., and Kusiak, A., 2014, Performance optimization of HVAC Systems with Computational Intelligence Algorithms, Energy and Buildings, 81, 371-380.
5. Choubin, B., Khlaighi-Sigaroodi, S., Malekian, A., and Kisi, O., 2016, Multiple linear regression, multi-layer perceptron network and adaptive neuro-fuzzy inference system for forecasting precipitation based on large-scale climate signals, Hydrological Sciences Journal, 61 (6), 1001-1009.
6. Ak, R., Fink, O., and Zio, E., 2016, Two Machine Learning Approaches for Short-Term Wind Speed Time-Series Prediction, IEEE Transactions on Neural Networks and Learning Systems, 27 (8), 1734-1747.
7. Fan, X., Wang, L., and Li, S., 2016, Predicting chaotic coal prices using a multi-layer perceptron network model, Resources Policy, 50, 86-92.
8. Dudek, G., 2016, Multilayer perceptron for GEFCom2014 probabilistic electricity price forecasting, International Journal of Forecasting, 32, 1057-1060
9. Azimi, R., Ghayekhloo, M., and Ghofrani, M., 2016, A hybrid method based on a new clustering technique and multilayer perceptron neural networks for hourly solar radiation forecasting, Energy Conversion and Management, 118, 331-344.
10. Nguyen, C., Wang, Y., and Nguyen, H.N, 2013, Random Forest Classifier Combined with Feature Selection for Breast Cancer Diagnosis and Prognostic, Journal of Biomedical Science and Engineering, 551-560.
11. Azar, A.T., Elshazly, H.I., Hassanien, A.E., Elkorany, A.M., 2014, A Random Forest Classifier for Lymph Diseases, Computer Methods and Programs in Biomedicine, 465-473.
12. Bosch, A., Zisserman, A., and Munoz, X., 2007, Image Classification using Random Forests and Ferns, 2007 IEEE 11th International Conference on Computer Vision.
13. Khalilia, M., Chakraborty, S., and Popescu, M., 2011, Predicting Disease Risks from Highly Imbalanced Data using Random Forest, BMC Medical Informatics and Decision Making, 11(51).
14. Guo, L. Ma, Y., Cukic, B., and Singh, H., 2004, Robust Prediction of Fault-proneness by Random Forests, 15th International Symposium on Software Reliability Engineering.
15. Shi, Y., and Eberhart, R.C., 1999, Empirical Study of Particle Swarm Optimization, Proceedings of the 1999 Congress on Evolutionary Computations, 1945-1950
16. Eberhart, R.C., Shi, Y., 2001, Particle Swarm Optimization: Developments, Applications, and Resources, Proceedings of the 2001 Congress on Evolutionary Computations, 81-86
17. Delgram, N., Sajadi, B., Kowsary, F., and Delgarm, S., 2016, Multi-objective optimization of the building energy performance: A simulation-based approach by means of particle swarm optimization (PSO), Applied Energy, 170, 292-303
18. Chen, Z., Xiong, R., Wang, K., and Jiao, B., 2015, Optimal Energy Management Strategy of a Plug-in Hybrid Electric Vehicle Based on Particle Swarm Optimization Algorithm, Energies, 3661-3678
19. Scikit-learn Developers, 2019, sklearn.feature_selection.f_regression, https://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.f_regression.html (Online accessed on 10 May 2019)
20. Scikit-learn Developers, 2019, sklearn.prepocessing.MinMaxScaler, https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.MinMaxScaler.html (Online accessed on 10 May 2019)
21. Wilson, D., 1972, Asymptotic Properties of Nearest Neighbor Rules Using Edited Data, IEEE Transactions on Systems, Man, and Cybernetics, 2 (3),408-421.
22. Lemaitre, G., Nogueira, F., Oliveira, D., and Aridas, C., 2016, imblearn.under_sampling.EditedNearestNeighbours, https://imbalanced-learn.readthedocs.io/en/stable/generated/imblearn.under_sampling.EditedNearestNeighbours.html#imblearn.under_sampling.EditedNearestNeighbours (Online accessed on 10 May 2019)
23. Scikit-learn Developers, 2019, sklearn.metrics.f1_score, https://scikit-learn.org/stable/modules/generated/sklearn.metrics.f1_score.html. (Online accessed on 10 May 2019)
24. Davis, J, and Goadrich, M., 2006, The Relationship Between Precision-Recall and ROC Curves, Proceedings of The 23rd International Conference on Machine Learning, 233-240.
25. Scikit-learn Developers, 2019, sklearn.metrics.roc_auc_score, https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_auc_score.html. (Online accessed on 10 May 2019)
26. Stoppato, A., Cavazzini, G., Ardizzon, G., and Rossetti, A., 2014, A PSO (Particle Swarm Optimization)-based Model for The Optimal Management of A Small PV(Photovoltaic)-pump Hydro Energy Storage in A Rural Dry Area