基于集成学习的支持向量回归预测模型

Support vector regression model with ensemble learning

作为统计机器学习中最为流行的算法之一,支持向量回归(SVR)在小样本、非线性、高维数据预测中有着许多优越的性质和实验表现。然而,SVR的复杂度直接由训练样本的尺寸n决定(其时间和空间复杂度分别为O(n2)、O(n3)),为此提出了一种基于集成的SVR预测模型。该模型将训练样本多次随机地分割为代表数据子集和验证数据子集,从而建立多个简化的SVR子模型及其评价,再利用组合法形成最终的集成预测器。最后,江西省某县的天气、日尖峰负荷数据用以检验该模型的适用性。

As one of the most popular algorithms in statistical machine learning,support vector regression( SVR) has many advantages in small samples,nonlinear and high dimensional data prediction. However,the complexity of SVR is directly determined by the size n of the training samples( its time and space complexity are O( n2),O( n3),respectively). Therefore,this paper proposes a SVR model with ensemble learning. The proposed model randomly splits the training sample into a representative subset and a validation subset,then establishes a simplified sub-SVR model and gives its evaluation,so as to form the final ensemble predictor by combining all the sub-SVR models. Finally,the weather on the weather and peak load dataset of Jiangxi Province is used to test the applicability of the proposed model.