摘要
基于层叠泛化策略SG(stacked generalization)提出一种新的母线负荷预测方法。该方法包含两级学习层,第1层针对原始母线负荷样本空间,对一组支持向量机SVM(support vector machine)进行交互验证式训练,训练完成后得到新的特征空间,该特征空间由这些支持向量机的输出和对应的真实值组成;第2层对输出进行线性组合,将新特征空间中的输出序列作为观测,对应的输出权值作为状态,使用卡尔曼滤波对权值进行递推估计。实例仿真证明,采用所提方法模型的泛化能力得到改善,从而提高母线负荷的预测精度。
A novel method for bus load forecasting was proposed based on stacked generalization. The proposed ap- proach includes two learning level spaces. The first one is for the original bus load data space, after the cross-valida- tion training and testing on a set of SVMs,a new space,composing of the output of the SVMs and the corresponding o- riginal data, is obtained and named as "level 1 space". Then, in the "level 2 space", the original output series and corresponding output weights are taken as the observations and states of Kalman filter, respectively. Finally, simulation results demonstrate that higher generalization accuracy can be obtained by using the proposed hybrid method, thus the forecasting accuracy can be improved greatly.
出处
《电力系统及其自动化学报》
CSCD
北大核心
2013年第3期8-12,55,共6页
Proceedings of the CSU-EPSA
基金
国家自然科学基金项目(50877024
51107032
61104045)
关键词
层叠泛化算法
支持向量机
卡尔曼滤波
母线负荷预测
stacked generalization
support vector machine (SVM)
Kalman filter
bus load forecasting