Application of Grey and Neural Network Approaches to Forecasting Solar Energy Output in Taiwan

• Main Authors: Tzu-Hui Wu, 吳姿慧
• Other Authors: Shin-Li Lu
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/65443929452551354735

碩士 === 真理大學 === 企業管理學系碩士班 === 101 === Taiwan is suitable to develop solar energy due to sufficient sun exposure, high temperature, and located in subtropical regions. Solar energy demand will be a potential orientation to study due to high cost and restriction on emission of greenhouse gases. Therefore, this study investigates solar energy output by coal, coal related product, crude, crude related product, gas, hydroelectric and nuclear power and construct the forecasting model to improve the prediction accuracy. Combing Neural network with grey system model GM (1,1) to establish the NNGM(1,1) forecast model in this study. The NNGM(1,1) model is compared to traditional ARIMA and regression model the NNGM(1,1) model in compared to. At first, using GM(1,1) model to forecaste the solar energy output, the mean absolute percentage error (MAPE) is up to 82.85%, it means the forecast is bad. Therefore, choosing four related factors to construct GM(1,4) model and improve MAPE to 4.04%. Then, only using neural network to establish the forecasting model, the MAPE is 2.47%. As the results, using the combination of neural network and grey forecast model to propose NNGM(1,4) model, which lower MAPE to 1.76%. Comparing to the traditional forecast models in this study, the traditional ARIMA forecast model can get the high prediction accuracy according to the sufficient history data.