基于多维特征和GBDT模型的输电线路线损率预测

PREDICTION OF LINE LOSS RATE OF TRANSMISSION LINES BASED ON MULTI-DIMENSIONAL FEATURES AND GBDT MODEL

传统输电线损率计算方法在数据特征方面的研究较少,模型预测能力有限。对此提出一种基于多维特征和梯度提升树(Gradient Boosting Decision Tree,GBDT)模型的输电线路线损率预测方法。该方法由多维信息特征工程和拟合建模两部分组成,其中多维信息特征工程对线损率信息、时序数据及线路本体信息进行特征构建,形成多维特征。在拟合建模中,将三类特征进行拼接作为输入,通过GBDT模型拟合,构建输电线路线损率预测模型。对某省500 kV输电线路实际数据进行测试,结果表明该模型的线损率平均误差为0.703%,相比于支持向量机模型,均方根误差下降84.5%;相较于无多维特征的GBDT模型,误差减少下降15.1%,验证了该方法的有效性。

There are few researches on the data mining by traditional transmission line loss rate computing methods,and the model prediction ability is limited.For the above reasons,a transmission line loss rate prediction method multi-dimensional features GBDT model was proposed.The method consists of multi-dimensional information feature engineering and fitting modeling.The multi-dimensional information feature engineering constructed the high-information feature of line loss rate information,sequence information and line ontology information.In the fitting modeling,three types of features were splicing as input,and the GBDT model was matched to form a line loss rate prediction model for transmission lines.The actual data of 500 kV transmission line in a province was tested.The results show that the mean error of line loss rate of the proposed model is 0.703%,and the root mean square error decreases by 84.5%compared with the SVM model.Compared with the GBDT model without multi-dimensional features,the error reduction decreased by 15.1%.The results verify the effectiveness of the proposed method.