We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
Adaptive Least Squares Support Vector Machines Method for Short-Term Load Forecasting Based on Mutual Information for Inputs Selection.
- Authors
Stojanovic, Miloš B.; Božic, Miloš M.; Stankovic, Milena M.; Stajic, Zoran P.
- Abstract
The accuracy of load forecasts has a significant effect on many aspects of power systems nowadays, such as power systems operations, economy management and control of power systems. Therefore, the electric utility corporation's savings highly depend on their ability to provide precise load forecasts. In the past few years least squares support vector machines (LS-SVMs) have been successfully employed in electric load forecasting. However, the accuracy of LS-SVMs is highly related to the selection of inputs which forms the training data set. Recently, mutual information (MI) has been employed as a criterion in the regression tasks, mostly for feature selection and for the identification of real inputs in noisy data sets. This paper proposes a new methodology for adaptive training set inputs selection, based on MI computation between training set inputs and current forecasting input, aiming to, primarily, improve the quality of load forecasting. In this way for each prediction step a new training set is formed, one that fits the current forecasting scenario better. Accordingly, several LS-SVM forecasting models have been built, based on inputs selection according to the MI criterion, for hourly forecasting of electric load for one day ahead. As the experimental results show, this approach can lead to significant improvements in the accuracy of load forecasts being achieved, as well as a great reduction in the number of inputs in the training set.
- Subjects
LEAST squares; SUPPORT vector machines; ELECTRIC power systems; ADAPTIVE control systems; ELECTRIC utilities; REGRESSION analysis; MATHEMATICAL models
- Publication
International Review of Electrical Engineering, 2012, Vol 7, Issue 1, p3574
- ISSN
1827-6660
- Publication type
Article