Load prediction in short-term implementing the multivariate quantile regression

• Published in: Energy (Oxford) 2020-04, Vol.196, p.117035, Article 117035
• Main Authors: Xing, Yazhou, Zhang, Su, Wen, Peng, Shao, Limin, Rouyendegh, Babak Daneshvar
• Format: Article
• Language: English
• Subjects: Benchmarks; Hybrid prediction method; Interim probability-based prediction; Multi-variable regression; Predictions; Quantile regression forest; Real time; Statistical analysis
• ISSN: 0360-5442; 1873-6785
• DOI: 10.1016/j.energy.2020.117035

Probability-based interim demand prediction plays and important role in managing the grid and optimizing the transmitted power through lines. Improved prediction techniques able to offer precise forecasting are supposed to be compatible with their own implementational situations in interim operation and must be highly efficient and fast. A lot of prediction techniques based on data are excessively verbose and not very suitable. The mentioned challenge emerges when the numerous demands are supposed to be forecasted at the same time, for example assessing and optimizing the energy delivery network. Here, a novel hybrid prediction framework is suggested, which improves the probability-based prediction of each load in real-time. The improvement approach uses the multi-variable quantile regression that is implemented on each prediction in real-time when a new observational data is inputted to the system. The proposed procedure is assessed using the demand data released by the Independent System Operator-new England for eight areas, which is composed of six states of the U.S. The performance of the probability-based prediction are compared to that of three other benchmarks with respect to reliability and accuracy. The suggested approach shows better accuracy compared to the highest-ranked benchmark. •Proposal of new cooperative prediction structure.•Extracting the time-dependent arrangement in the consumption curve of each load.•Application of new version of QR which uses bias correction in QRF.•Contrasting the location-based and random categorization techniques.