[关键词]
[摘要]
基于集成学习中的随机森林(Random forest,RF)算法,利用某燃煤机组SCR脱硝系统历史运行数据,通过计算各特征变量重要性评分选取出最优输入变量集,建立脱硝系统出口NOx浓度最优的RF预测模型。与常见建模方法进行对比,结果表明:基于RF算法的模型预测精度明显高于BP神经网络和SVM等建模方法,且经最优变量选择后,具有更好的泛化能力和更短的建模时间,能够有效应用于电厂烟气脱硝系统出口NOx浓度预测中。
[Key word]
[Abstract]
Based on random forest (RF) algorithm in ensemble learning,utilizing actual operating data of thermal power plant denitration system and calculating the importance index of each characteristic variable,the optimal input variable set is selected to establish the RF prediction model of outlet NOx concentration in denitration system.The prediction results comparing with other common modeling methods show that the model prediction accuracy based on RF algorithm is obviously higher than the modeling approaches of BP neural network and SVM.The model has better generalization ability and shorter modeling time.It can be effectively applied to the prediction of outlet NOx concentration for power plant flue gas denitration system.
[中图分类号]
TK39
[基金项目]
上海市电站自动化技术重点实验室(13DZ2273800);上海市科委地方能力建设项目(18020500900);上海市自然科学基金(19ZR1420700)