摘要
针对某超超临界660 MW机组锅炉,建立了基于随机森林(RF)和梯度提升决策树(GBDT)算法的氮氧化物(NO_x)排放预测模型。从电厂SIS系统筛选得到历史运行数据中的稳态工况点,利用RF模型对数据特征进行筛选,并以选中特征作为输入变量建立基于GBDT的NO_x排放预测模型。与支持向量机(SVM)、RF等模型的对比表明基于RF的特征选择能提升模型性能;较于其他模型,RF-GBDT具有最高的NO_x排放预测精度。
A nitric oxides(NO_x) emissions model was established by combining random forest(RF) algorithm and gradient boost decision tree(GBDT) algorithm for a 660 MW ultra-supercritical boiler. Data sieving was employed on operational records provided by Supervisory Information System(SIS) and steady-state records were get. The records, whose dimension was reduced with the feature ordering function of RF model then, were used as inputs to build the NO_xemissions prediction model with GBDT algorithm. Several other models were built to make a comparison. The test result proves that feature selection basing on RF algorithm can improve predictive accuracy of models. The RF-GBDT model is more outstanding in prediction effect comparing with other models.
出处
《电站系统工程》
2017年第2期5-8,共4页
Power System Engineering