相关向量机和特征选取技术在短期负荷预测中的应用

Application of Relevance Vector Machine and Feature Selection Technology in Short-term Electric Load Forecasting

通过改进传统的Relief算法,提出一种短期负荷预测特征输入量的选取方法,并使用相关性分析法来消除冗余特征。在所选特征和气温数据的基础上,应用相关相量机来建立预测模型。以美国德州电力市场某东部城市的真实负荷数据来进行仿真分析,结果表明本文的特征选取方法能够很好的提取负荷的短期趋势特征和周期性特征,而相关相量机也获得了比支持向量机和BP神经网络要好的预测结果,体现了本文方法的优越性。

In this paper, the traditional Relief Algorithm was improved to form a feature selection method which has been applied in short-term load forecasting, and the correlation analysis was used to eliminate the redundant features. Based on the selected features and temperature data, the forecasting model is established by using the Relevance Vector Machine. The real power load data and temperature data from Texas electricity market was used for simulation analysis. The results show that the proposed feature selection method can extract load short-term trend features and cycle features, and the Relevance Vector Machine also obtained better forecast results than support vector machine （SVM） and BP neural network did.