近年來，環保意識的抬頭以及石化能源的逐漸枯竭下，再生能源成為全世界發展的目標。然而再生能源的發電量極不穩定，因此需要將電儲存起來，並在適當的時間輸出，所以本研究將以此為出發點，利用集成學習來預測用電需求和太陽能發電，並利用預測的結果，來探討微電網使用預測數據的電力調度(power dispatch)。
微電網(microgrid)是由分散式電源、儲能設備、負載和調度監控及保護裝置彙集而成的小型電網系統，除了可以與主電網進行並網(grid-conncted)運作模式，也可以獨立操作進行孤島(island)運作模式，所以微電網能夠在離島或是偏遠地區獨立運作。
本研究的微電網分別使用柴油發電機、燃氣發電機和太陽能發電等分散式發電機，以及蓄電池裝置來建構。同時根據用電形式的不同，將微電網分為住宅和商業區域，並且兩個區域間可以互相傳輸電力。電力調度的過程就是主電網、分散式發電、區域間輸電、蓄電池和用電需求間的電力平衡。
為了預測用電需求、太陽能發電以及即時邊際電價，使用Matlab來進行集成學習。首先，分析相關性以決定要使用的影響因子，接著使用分別使用線性回歸(linear regression)、邏輯回歸(logistic regression)、類神經網路(neural network)、高斯過程回歸(Gaussian process regression)、支援向量機(support vector regression)、隨機森林(random forest)和梯度提升樹(gradient tree boosting)等機器學習演算法，對各個影響因子和預測目標進行建模，再來比較集成模型與單一模型的效果，結果顯示，集成模型較單一模型的效果來得好。此外，為了應用在後續的電力調度，在預測上將分為一小時(hour-ahead)預測和一日(day-ahead)預測，以此來貼近電力調度的實務面。
對於預測的不確定性，本研究對於各個小時會設定一個備轉容量(operating reserve)來滿足在用電需求和太陽能發電上的誤差，同時在電力調度上，提出使用一小時預測再加上一日預測的方法，並用蓄電池來調節預測的誤差。最後，使用GAMS分別對微電網在並網模式和孤島模式下，對經濟、環境等目標進行優化，並探討結果。

Power dispatch is the process of finding a balance between main grid, distributed generators, power flow, storage battery and electricity demand. Distributed generators, including gas engines (GE), diesel engines (DE) and photovoltaic arrays (PV), and storage battery (SB) are used to construct microgrid in this research. Moreover, two different types of electricity usage, which are residential and commercial, are considered in the research as well.
In order to predict the electricity demand, solar power and electricity price, Matlab is used to perform ensemble learning. To do so, correlation is analyzed to determine impact factor which will be made use of. Afterwards, prediction models are built by means of seven machine learning algorism, including linear regression, logistic regression, neural network, Gaussian process regression, support vector regression, random forest, gradient tree boosting; ensemble learning model is compared with each model. The result shows that ensemble learning model is superior to all of the others. In addition, in practical perspective, prediction is divided to hour-ahead and day-ahead prediction to close to real power dispatch.
Due to uncertainty of the prediction, operating reserve is set on an hourly basis to compensate the offset of power demand and solar power; storage battery is also used to balance the error happening during the prediction. Eventually, power dispatch of the microgrid is optimized by GAMS in terms of economic and environment in both grid-connected mode and island mode.

[1] Ahn, S. J., Nam, S. R., Choi, J. H., & Moon, S. I. (2013). Power scheduling of distributed generators for economic and stable operation of a microgrid. IEEE Transactions on Smart Grid, 4(1), 398-405.
[2] Jiayi, H., Chuanwen, J., & Rong, X. (2008). A review on distributed energy resources and MicroGrid. Renewable and Sustainable Energy Reviews, 12(9), 2472-2483.
[3] Lasseter, R. H., & Piagi, P. (2004, June). Microgrid: A conceptual solution. In IEEE Power Electronics Specialists Conference (Vol. 6, pp. 4285-4291).
[4] Lasseter, R. H., & Piagi, P., "Control and Design of Microgrid Components, Final project report," PSERC publication 06-03
[5] Zia, M. F., Elbouchikhi, E., & Benbouzid, M. (2018). Microgrids energy management systems: A critical review on methods, solutions, and prospects. Applied Energy, 222, 1033-1055.
[6] Opitz, D., & Maclin, R. (1999). Popular ensemble methods: An empirical study. Journal of Artificial Intelligence Research, 11, 169-198.
[7] Polikar, R. (2006). Ensemble based systems in decision making. IEEE Circuits and Systems Magazine, 6(3), 21-45.
[8] Rokach, L. (2010). Ensemble-based classifiers. Artificial Intelligence Review, 33(1-2), 1-39.
[9] Zhou, Z. H. (2015). Ensemble learning. Encyclopedia of Biometrics, 411-416.
[10] Sill, J., Takács, G., Mackey, L., & Lin, D. (2009). Feature-weighted linear stacking. arXiv Preprint arXiv:0911.0460.
[11] Williams, C. K., & Rasmussen, C. E. (2006). Gaussian Processes for Machine Learning (Vol. 2, No. 3, p. 4). Cambridge, MA: MIT press.
[12] Smola, A. J., & Schölkopf, B. (2004). A tutorial on support vector regression. Statistics and Computing, 14(3), 199-222.
[13] Breiman, L. (2001). Random forests. Machine learning, 45(1), 5-32.
[14] James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An Introduction to Statistical Learning (Vol. 112, p. 18). New York: springer.
[15] Chen, T., & Guestrin, C. (2016, August). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd International Conference on Knowledge Discovery and Data Mining (pp. 785-794). ACM.
[16] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735-1780.
[17] Altman, N. S. (1992). An introduction to kernel and nearest-neighbor nonparametric regression. The American Statistician, 46(3), 175-185.
[18] MacKay, D. J., & Mac Kay, D. J. (2003). Information Theory, Inference and Learning Algorithms. Cambridge university press.
[19] Maaten, L. V. D., & Hinton, G. (2008). Visualizing data using t-SNE. Journal of Machine Learning Research, 9(Nov), 2579-2605.
[20] Mockus, J. (2012). Bayesian Approach to Global Optimization: Theory and Applications (Vol. 37). Springer Science & Business Media.
[21] Frazier, P. I. (2018). A tutorial on bayesian optimization. arXiv Preprint arXiv:1807.02811.
[22] Wang, J., Wang, X., & Wu, Y. (2005). Operating reserve model in the power market. IEEE Transactions on Power systems, 20(1), 223-229.
[23] Hyndman, R. J., & Koehler, A. B. (2006). Another look at measures of forecast accuracy. International Journal of Forecasting, 22(4), 679-688.
[24] Van Der Maaten, L. (2009, April). Learning a parametric embedding by preserving local structure. In Artificial Intelligence and Statistics (pp. 384-391).
[25] Matos, M. A., & Bessa, R. J. (2010). Setting the operating reserve using probabilistic wind power forecasts. IEEE Transactions on Power Systems, 26(2), 594-603.
[26] Yan, X., Francois, B., & Abbes, D. (2015, June). Operating power reserve quantification through PV generation uncertainty analysis of a microgrid. In 2015 IEEE Sudhoven PowerTech (pp. 1-6). IEEE.
[27] Yan, X., Abbes, D., & Francois, B. (2017). Uncertainty analysis for day ahead power reserve quantification in an urban microgrid including PV generators. Renewable Energy, 106, 288-297.
[28] Mason, K., Duggan, J., & Howley, E. (2017, July). Evolving multi-objective neural networks using differential evolution for dynamic economic emission dispatch. In Proceedings of the Genetic and Evolutionary Computation Conference Companion (pp. 1287-1294). ACM.