Electric Load Forecasting Using Support Vector Machines Optimized by Genetic Algorithm

Abstract

Electric load forecasting has become an important research area for secure operation and management of the modern power systems. In this paper we have proposed a seven support vector machines model for daily peak load demand long range forecasting. One support vector machine for each day of the week is trained on the past data and then used for the forecasting. In tuning process of support vector machines there are few parameters to optimize. We have used genetic algorithm for optimization of these parameters. The proposed model is evaluated on the electric load data used in EUNITE load competition in 2001 arranged by East-Slovakia Power Distribution Company. A better result is found as compare to best result found in the competition.

Midterm Demand Prediction of Electrical Power Systems Using a New Hybrid Forecast Technique

“…The first MTLF method of Table III (SVM) was the winning entry of the EUNITE competition [8]. The second method is a combination of SVM and GA wherein the GA is used to optimize the SVM parameters [23]. In [24], 12 forecast methods have been tested on the EUNITE competition data, wherein the "Extended Bayesian Training" produces superior results, reported as the third method in Table III.…”

Short-term load forecasting based on support vector regression and load profiling

“…The kernel transformation allows the handling of nonlinear relationships in the data as a higher dimensional space is obtained in which a linear regression model can be built. The kernel function typically used for electric load forecasting purposes is the radial basis function, due to its ability to deal with nonlinearities . The radial basis function is defined in Equation (6), where γ is viewed as another predefined constant that represents the width of the basis function.…”

“…A machine‐learning method based on support vector machines (SVMs) is also being evaluated. This method has, according to several authors , more advantages than using neural network models, not only because it implements a structural risk minimization principle rather than an empirical risk minimization principle (this conclusion will be further discussed in Section 2.2) but also because its training derives from quadratic programming, resulting in a global optimal solution. Following this trend, this paper proposes a methodology to forecast the electric load for the 24 h of the following day based on the support vector regression methodology, which can also be used to analyze the potential benefits of load profiling information before and after a feature selection stage.…”

Short-term load forecasting using information obtained from low voltage load profiles

“…Recent approaches are also exploited to ascertain the most influent variables, e.g. using principal component analysis [5] to reduce the original input space to several characteristic variables or using support vector machines, an alternative to neural networks, that implements a structural risk minimization principle in regression thematic [6].…”