Extended modeling procedure based on the projected sample for forecasting short-term electricity consumption. Issue 2 (April 2016)

• Record Type: Journal Article
• Authors: Chang, Che-Jung; Lin, Jan-Yan; Chang, Meng-Jen
• Is Part Of: Advanced engineering informatics. Volume 30:Issue 2(2016:Apr.)
• Page Start: 211
• Page End: 217
• Publication Date: 2016-04
• Subjects: Forecasting -- Small data set -- Latent information -- Electricity consumption; Computer-aided engineering -- Periodicals; Engineering -- Data processing -- Periodicals; 620.00285
• DOI: 10.1016/j.aei.2016.03.003
• Languages: English
• ISSNs: 1474-0346

Abstract

Highlights: Forecasting electricity consumption plays a vital role for policy makers. Short-term predictions using new limited data for managers are important. The proposed modeling procedure can extract hidden information for knowledge learning. The proposed method is an appropriate tool for forecasting short-term consumption. Abstract: Effectively forecasting the overall electricity consumption is vital for policy makers in rapidly developing countries. It can provide guidelines for planning electricity systems. However, common forecasting techniques based on large historical data sets are not applicable to these countries because their economic growth is high and unsteady; therefore, an accurate forecasting technique using limited samples is crucial. To solve this problem, this study proposes a novel modeling procedure. First, the latent information function is adopted to analyze data features and acquire hidden information from collected observations. Next, the projected sample generation is developed to extend the original data set for improving the forecasting performance of back propagation neural networks. The effectiveness of the proposed approach is estimated using three cases. The experimental results show that the proposed modeling procedure can provide valuable information for constructing a robust model, which yields precise predictions with the limited time series data. The proposed modeling procedure is useful for small time series forecasting.