不同於以往傳統機械式電表，僅以抄表的方式紀錄所消耗的累積總電量，智慧電表可以每隔一小時紀錄並回傳資料，讓電力公司更能即時掌握終端用戶當下的用電資訊，藉此進行整體電力負載的預測與調整。

本研究將運用支持向量迴歸技術建構電力負載之預測模型，研究中除了考慮歷史用電量外，並納入天氣方面之影響因子。此外，為提高模型之預測準確率且避免過度擬合的風險，本研究運用粒子群演算法，以找出支持向量迴歸的最佳化參數值，並以此參數值建構支持向量迴歸模型。

為驗證模型的準確性，本研究利用北部台灣每小時的平均負載資料進行實證分析。在數值範例中，本研究針對預測一小時後及24小時候的兩種模型，分別挑選出合適的預測因子，並利用所提出的PSO-SVR方法最佳化支持向量迴歸的參數值。相較於過去的研究，本研究以PSO演算法提升決定參數值之效率，也提升了模型的準確度，其預測誤差分別為0.70%及2.55%。結果顯示即使短期電力負載預測變動較大，本研究仍能得到相當好的結果。

本研究的應用不僅提供電力負載預測使電力公司能進行平衡管控調配，而不會輸出多餘電力，造成能源的損失；並配合不同用電時段的供電價格，協助消費者了解其用電習慣並調整最佳用電量，以達到節省能源之目標。

Smart grid has recently become a common system in many countries, and smart meter is one of the essential components of this system. Compared with traditional meters read on a monthly basis, smart meters record consumption of electric energy in intervals of an hour or less. The consumption data gathered from smart meters allow electricity companies to better understand electricity usage in the future and monitor electricity supply more efficiently.

An electricity consumption forecasting model established on the framework of support vector regression (SVR) was proposed in this study. Given that various factors affected consumption patterns, apart from historical data, weather variations and features of a particular time were also considered in this study. Based on this information, the stepwise regression analysis was applied for feature selection. However, the accuracy of the SVR model is largely dependent on the selection of the model parameters. The particle swarm optimization (PSO) algorithm was proposed to determine the optimal values of parameter that improves the accuracy and efficiency of the SVR model.

The proposed procedure was demonstrated step by step to forecast the load in the north of Taiwan from November 8, 2014, to November 30, 2014 (552 hours in total). Experimental results show that this method can provide electricity forecasts with 0.70% and 2.55% of mean absolute percentage error (MAPE) for 1 and 24 hours ahead, respectively. This result indicates promising performance in forecasting accuracy. This model is also computationally efficient and can be applied to make predictions within 1 second.

[1] 台灣電力公司，台灣電力公司全球資訊網站，http://www.taipower.com.tw
[2] 交通部氣象局，中央氣象局全球資訊網，http://www.cwb.gov.tw/V7/index.htm
[3] TOKYO ELECTRIC POWER COMPANY，http://www.tepco.co.jp/en/index-e.html
[4] Andrew C. C., Comparing Neural Networks and Regression Models for Ozone Forecasting, Journal of the Air & Waste Management Association, 47:6, 653-663, DOI: 10.1080/10473289.1997.10463925,1997.
[5] Basak, D., Pal, S. and Patranabis, D.C., Support Vector Regression, Neural Information Processing – Letters and Reviews, Vol. 11, No. 10, 2007.
[6] Carlisle, A. and Dozier, G., An off-the-shelf PSO, Proceedings of the workshop on particle swarm optimization, 2001.
[7] Cherkassky, V. and Ma, Y., Practical Selection of SVM Parameters and Noise Estimation for SVM Regression, Neural Networks, Volume 17, Issue 1, Pages 113–126, 2004.
[8] Clerc, M., The swarm and the queen: towards a deterministic and adaptive particle swarm optimization, Proceedings of the 1999 Congress on Evolutionary Computation, Volume 3, Pages 1951-1957, 1999.
[9] Duan, K., Keerthi, S. and Poo, A., Evaluation of simple performance measures for tuning SVM hyperparameters, Technical Report, Department of Mechanical Engineering, National University of Singapore, 2001.
[10] Eberhart, R. C. and Shi,Y., Tracking and Optimizing Dynamic Systems with Particle Swarms, Proceedings of the 2001 Congress on Evolutionary Computation, Volume 1, Pages 94-100, 2001.
[11] Fan, S., Chen, L., Lee, W. J., Machine learning based switching model for electricity load forecasting, Energy Conversion and Management, Volume 49, Pages 1331–1344, 2008.
[12] Fok, C.Y. and Vai, M.I., Very Short Term Load Forecasting for Macau Power System, Intelligent Computing Technology, Volume 7389, Pages 538-546, 2012.
[13] Hassan, R., Cohanim, B., de Weck, O., A Comparison of Particle Swarm Optimization and the Genetic Algorithm, 2004.
[14] Humeau, S., Wijaya, T.K. Vasirani, M. and Aberer, K., Electricity Load Forecasting for Residential Customers: Exploiting Aggregation and Correlation between Households, Sustainable Internet and ICT for Sustainability (SustainIT), 2013.
[15] Kandil, N., Wamkeue, R., Saad, M. and Georges, S., An efficient approach for short term load forecasting using artificial neural networks, Electrical Power and Energy Systems, Volume 28, Pages 525–530, 2006.
[16] Liang, Y.Z., Xu, Q.S., Li, H.D. and Cao, D.S., Support Vector Machines and Their Application in Chemistry and Biotechnology, CRC Press, 2011.
[17] Liu, N., Babushkin, V. and Afshari, A., Short-Term Forecasting of Temperature Driven Electricity Load Using Time Series and Neural Network Model, Journal of Clean Energy Technologies, Vol. 2, No. 4, 2014.
[18] Meng, M., Shang, W. and Niu, D., Monthly Electric Energy Consumption Forecasting Using Multiwindow Moving Average and Hybrid Growth Models, Journal of Applied Mathematics, Volume 2014, Article ID 243171, 7 pages, 2014.
[19] Shi, Y. and Eberhart,R. C., Empirical Study of Particle Swarm Optimization, Proceedings of the 1999 Congress on Evolutionary Computation,Volume 3, 1999.
[20] Sood, R. Koprinska, I. and Agelidis, V.G., Electricity Load Forecasting Based on Autocorrelation Analysis,The 2010 International Joint Conference on Neural Networks (IJCNN), 2010.
[21] Vapnik, V., Cortes, C., Support-vector networks, Machine Learning 20 (3):273, 1995.
[22] Zhang, G. P., Time series forecasting using a hybrid ARIMA and neural network model, Neurocomputing, Volume 50, Pages 159–175, 2003.