.png){kind=icon}
.png){kind=icon}
Return
Day-ahead load forecast using random forest and expert input selection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.enconman.2015.07.041.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The electrical load forecast is getting more and more important in recent years due to the electricity market deregulation and integration of renewable resources. To overcome the incoming challenges and ensure accurate power prediction for different time horizons, sophisticated intelligent methods are elaborated. Utilization of intelligent forecast algorithms is among main characteristics of smart grids, and is an efficient tool to face uncertainty. Several crucial tasks of power operators such as load dispatch rely on the short term forecast, thus it should be as accurate as possible. To this end, this paper proposes a short term load predictor, able to forecast the next 24 h of load. Using random forest, characterized by immunity to parameter variations and internal cross validation, the model is constructed following an online learning process. The inputs are refined by expert feature selection using a set of if-then rules, in order to include the own user specifications about the country weather or market, and to generalize the forecast ability. The proposed approach is tested through a real historical set from the Tunisian Power Company, and the simulation shows accurate and satisfactory results for one day in advance, with an average error exceeding rarely 2.3%. The model is validated for regular working days and weekends, and special attention is paid to moving holidays, following non Gregorian calendar. (C) 2015 Elsevier Ltd. All rights reserved.
Keywords:
Short term load forecast
Random forest
Expert input selection
Online learning
Variable importance
Prediction of holidays
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
10.9
Papers:
2.0W
Citations:
11.3W