SVM model of system marginal price based on developed independent component analysis

Abstract

A support vector machine forecast model of system marginal price (SMP) is presented, which integrates with developed independent component analysis. First, this paper designs the feature extraction model of SMP influencing factors based on independent component analysis, which is developed by kernel density estimate. Then the feature extraction of SMP influencing factors is realized, which produces the independent the features. After the training of regress SVM with the obtained sample set, SMP forecast model is built. According to this model, SMP forecast accuracy is enhanced, with the generalization ability of support vector machine and the feature extraction ability of independent component analysis, which not only improves the representation of SMP input samples, but also decreases the maximum of absolute value of relative error, representation of SMP input samples, but also decreases the maximum of absolute value of relative error. Finally, SMP real-word data of spot market in California is employed to demonstrate the validity of the proposed approach.