基于非负矩阵分解的相关向量机短期负荷预测模型

A Short-term Load Forecasting Model Based on Relevance Vector Machine with Nonnegative Matrix Factorization

针对传统特征提取方法只能抽取样本的代数特征而无法顾及问题实际意义的缺点,提出一种基于非负矩阵分解的相关向量机短期负荷预测模型。通过非负矩阵分解算法对输入样本进行分解,得到非负的低维映射矩阵,将该低维矩阵输入相关向量机进行训练预测。由于此低维矩阵具有非负性质,因而该模型在消除冗余数据、降低维数的同时,保留了原始问题的实际意义。实验结果表明,所提出的方法能有效降低输入变量的维数,预测精度也得到了提高。

In view of the limitation of traditional feature extracting algorithm in extracting only the algebraic features of samples to the neglect of the practical significance of the original problem, a short-term load forecasting model based on the relevance vector machine （RVM） is proposed. By using the nonnegative matrix factorization （NMF） algorithm, the dimension of input variables is reduced, then a short-term load forecasting model based on the RVM is proposed. The input data is decomposed using the NMF algorithm, where the nonnegative lower-dimension mapping matrix derived is taken as the input of RVM for training and prediction. Owing to the nonnegative property of the lower-dimension matrix, it retains the practical significance of the original problem while eliminating the redundant data and reducing dimensions. Simulation results show that the dimensions of the input variables can he effectively rerhlced and the accuracy greatly improved.