基于混合模型和Stacking框架的循环水出口温度预测

Prediction of Outlet Temperature of Circulating Water Based on Hybrid Model and Stacking Framework

针对单一机理模型与数据驱动模型泛化能力弱、预测效果不理想的现状,以某电厂1000MW的凝汽器管侧模型为例,提出了一种基于混合模型和Stacking框架的参数预测方法。首先将原始数据集按人工经验分为温度数据与压力数据,使用5折交叉验证训练SVM温度模型与AdaBoost压力模型,其次以管侧机理模型的输出值与实际值为输入训练GDBT误差补偿模型,最后基于Stacking框架进行多模型融合,既弥补了机理模型在复杂工况下的模型参数失配问题,又可以对时间序列数据进行深层次的挖掘。实验结果表明,Stacking框架下的混合模型在测试集上均方误差比GRNN模型降低了0.012,判定系数相比于数据驱动组合模型提高了4.53%,提高了模型的整体预测精度。

In view of the weak generalization ability and unsatisfactory prediction effect of single mechanism model and data-driven model,a parameter prediction method based on hybrid model and Stacking framework was proposed by taking a 1000MW condenser tube side model of a power plant as an example.Firstly,the original data set was divided into temperature data and pressure data according to manual experience,and 5-fold cross-validation was used to train the SVM temperature model and the AdaBoost pressure model.Secondly,the output value and actu-al value of the tube-side mechanism model were used as input to train the GDBT error compensation model.Finally,multi-model fusion was carried out based on the Stacking framework,which not only compensates for the problem of model parameters mismatch in complex working conditions,but also enables deep mining of time series data.The ex-perimental results show that the mean square error of the hybrid model under the framework of Stacking is 0.012 low-er than that of the GRNN model,and the determination coefficient is 4.53%higher than that of the data-driven com-bination model,which improves the overall prediction accuracy of the model.