利用支持向量機建構太陽能發電量預測模型-以臺灣為例

Abstract

隨著經濟與文明的進步，地球環境保護以及能源缺乏兩大問題日益嚴重。近年來國際非常重視全球暖化與環保議題，故各國紛紛開發太陽能、風力、生物質能等低汙染等再生能源(Renewable Energy)。太陽能發電系統相較於其他再生能源發電系統較不易受到地形限制，體型輕、薄，沒有轉動組件的噪音，應用上安全性高且易與環境建物結合。臺灣地狹人稠，安裝的太陽能設置容量因而受到限制，在高昂的設置成本下，若無法增加太陽能發電量系統的轉換效率以增加發電量，太陽能系統設置成本將成為廠商使用太陽能發電的一大阻礙。因此，在推廣太陽能發電系統時，需先了解在裝置太陽系統時可能影響太陽能發電效率之重要因子，進而探究如何利用這些重要因子增加太陽能發電系統輸出的發電量。支持向量機(SVM)與傳統的統計預測模型不同之處是SVM具有不斷重複學習系統的能力之外，且是遵循結構風險最小化原則的新型學習機器。因此，本研究針對實際運作的太陽能發電量系統資料進行分析，分別利用SVM和整合逐步迴歸分析與SVM之兩階段預測方法來建構太陽能發電量預測模型，以提高發電量預測模型之準確性。本研究利用臺灣某單位目前已架設之太陽能發電量系統之效能參數與臺灣各地區太陽輻射量等變數，建立一個有效的太陽能發電量預測模型，能精準預測太陽能系統發之發電量，可用作日後政府或廠商投資太陽能發電前成本效益分析的參考依據。

Accompanying with the progress in economy and civilization, two major problems, globe environmental protection and the deficiency of power, are getting worse day by day. International nations have stressed on global warming and other environmental issues in recent years, hence, each country started to develop renewable energy such as solar energy, wind power, and biomass. Comparing with other renewable energy electric generating system, solar electric generating system is unlikely to be confined by landform, moreover, it has lighter body type, soundless operation, higher safety in application and can easily combined with buildings in surroundings. Because of the cramped but highly populated land in Taiwan, the capacity of solar electric generating system installed in Taiwan is limited. At such a high cost of installation, if the solar electric generating system cannot be increased to raise up power generation, the heavy installation cost of solar electric generating system will be one significant barrier for promoting the solar power system. Consequently, when promoting the solar electric generating system, it is necessary to find the important factors that may affect the efficiency of generating the solar power and then explore how to make good use of the important factors to increase the output of solar electric generating system. Support vector machine (SVM) is a brand-new learning machine which can repeat the learning system to minimize the prediction hazard. The objective of this study is to analyze the actual data from the solar power generation system, using SVM and integrated stepwise regression analysis and two-stage prediction method (Stepwise and SVM)to improve the accuracy of power generation forecasting model. Finally, this study utilized solar energy generation system performance parameters and solar radiation variables that have a set of real data provided by an organization in Taiwan to build an effective solar power generation forecasting model which can accurately predict the power generation of solar energy system. Also, it can be used as the cost - benefit analysis for investment in solar power generating system.