太陽光電系統短期發電量預測技術

Research on the Short-Term Photovoltaic Power Forecasting

葉志明(Chih-Ming Yeh)

指導教授 ： 陳昭榮

國立臺北科技大學/電資學院/電機工程系研究所/碩士(2011年)

摘要

自從京都議定書的簽訂與各國對二氧化碳訂定排放量限制之後，已促使綠能產業的蓬勃發展。台灣在太陽能產業技術領域可說是相當成熟，其主要特色包括可結合建築、建置容易、低污染、以及減少化石燃料消耗等。太陽能是藉著太陽照度轉成電能，其照度會隨著季節、時間、雲量、天氣等因素而改變，因此太陽能發電量存在相當高的不確定性，併入系統後也將造成電力系統可靠度的不穩定性，進而衍生出太陽能發電量預測的重要性。
本論文使用的預測方法包括時間序列法、倒傳遞類神經網路、以及適應性網路模糊推論系統: 時間序列法主要特色是能觀察出資料的相關性以及資料特性；倒傳遞類神經網路能針對非線性數據做有效的預測;適應性網路模糊推論系統最主要精髓是使用模糊理論的解模糊化以及類神經網路的學習特色進而提升預測的準確度。
本論文的預測資料是以台中火力電廠PV場、澎湖科技大學PV場、馬來西亞綠能展示館PV場三個地區歷史資料為依據，其中台中數據又可分兩個時段的案例來分析，分別為4~9月和10~3月，澎湖也可分成兩個案例資料，包括為5~8月和9~12月，三處PV場地的容量分別為72kW、70kW、45.36kW，其材料分別為多晶、多晶、單晶。預測結果顯示適應性網路模糊推論系統(ANFIS)預測誤差較為準確，因為它能有效將每個輸入參數做模糊分類，然後藉由類神經網路的學習，這樣不僅具有模糊的優點也具備類神經網路的優點，並且可強化整體預測結構。如此預測精準度也相對會提升許多，有了此預測架構，當太陽光電裝置容量增大時能更精確的預測其發電量。模擬結果顯示在這五個案例中，採用ANFIS預測模式的預測誤差較為準確，約為3.8%，較精確的預測不僅提供給業者參考，以便朝更大的容量發展，也能將此資料提供台電公司作為經濟調度的參考。

關鍵字

太陽能發電量預測 ； 時間序列 ； 倒傳遞類神經網路 ； 適應性網路模糊推論系統 ； 自迴歸整合移動平均模式。

並列摘要

Since the the signing of the Kyoto Protocol and the global efforts in reducing carbon emission, the green energy industry has been developing with great vitality in recent years. Taiwan in particular boasts a well-established solar energy industry. Characterized by advantages like easy installation and integration into buildings, low pollution, and the capability of lowering fossil fuel consumption, Solar energy relies on capturing and converting solar radiation into electricity. However, subject to the changes in season, time, weather, cloud amount and other external factors, solar radiation is marked with uncertainty as it is difficult to predict the energy output in the even the next hour. This inherent instability renders the prediction of energy output an especially crucial issue in the effective operation of solar power systems.
This paper uses prediction methods including Time series analysis aims at measuring the correlation between data and identifying the special features of data to facilitate prediction. Back-propagation neural network is capable of performing effective prediction by analyzing nonlinear statistical data; The main essence of the Adaptive Neuro-Fuzzy Inference Systems solution is the use of fuzzy theory and neural network learning characteristics and thus enhance the prediction accuracy.
The forecast data are historical data in Taichung, Penghu and Malaysia, and the solar energy capacities are respectively 72kW, 70kW and 45.36kW. The predicted results show that Adaptive Neuro-Fuzzy Inference Systems prediction error and low frequency high, because it can effectively be done for each input variable fuzzy classification, and learning by neural networks, fuzzy features that not only has the characteristics of neural networks, and strengthen the overall predicted structure. This will increase the forecast accuracy is relatively many, with the predicted structure, when the capacity increases to more accurately predict when the document generation. Simulation results show that in these five cases, ANFIS is more accurate the prediction error is about 3.8% accurate forecasts for the industry not only provides reference for the development towards a greater capacity, can also provide this information as an economic Taipower scheduling.

並列關鍵字

solar Power forecast ； Time series ； Back Propagation Network ； Adaptive Neuro-Fuzzy Inference Systems ； Autoregressive Integrated Moving Average Models.

參考文獻

[16] 許登芳，基礎管理數學，臺北：藍海文化事業股份有限公司，2010。

[1] W. Cai, L. Zhang, X. Yu and D. Wang, "A study on solar energy heating for residential buildings in small village," Asia-Pacific Conference on Power and Energy Engineering, Ningbo, China, March, 2010, pp.1-4.

[5] I. Kaizuka, T. Ohigashi, H. Matsukawa and O. Ikki, "PV trends in Japan: New framework for introduction of PV system," IEEE Photovoltaic Specialists Conference, Tokyo Japan, June 2009, pp. 712-716.

[6] Seong-Ho Lee, "The Role of New and Renewable Energy Center(KEMCO) in the Expansion of PV Market in Korea," IEEE 4th World Conference on Photovoltaic Energy Conversion, Gyeonggi-Do Korea, May 2006, pp. 12-14.

[7] M.H. Rahman, J. Nakayama, K. Nakamura and S. Yamashiro, "A viable grid-connected PV-ECS system with load leveling function using a day-ahead weather forecast," International Conference on Power System Technology, Japan, Nov. 2004, pp. 311– 316.