基于迁移学习的湿法烟气脱硫系统出口SO_(2)浓度预测研究  

Research on the Prediction of SO_(2) Concentration at the Outlet of Wet Flue Gas Desulfurization System Based on Transfer Learning

作　　者：王力光 贠勇博 朱保宇 司风琪[2] Wang Liguang;Yun Yongbo;Zhu Baoyu;Si Fengqi(Franchise Branch of Datang Environmental Industry Group Co., Ltd., Nanjing 211100,China;Key Laboratory of Energy Thermal Conversion and Control ofMinistry of Education, Southeast University, Nanjing 210096, China)

机构地区：[1]大唐环境产业集团股份有限公司特许经营分公司,南京211100 [2]东南大学能源热转换及其过程测控教育部重点实验室,南京210096

出　　处：《发电设备》2021年第6期420-425,共6页Power Equipment

摘　　要：以某600 MW机组脱硫系统为对象,提出了一种基于迁移学习的湿法烟气脱硫系统出口SO_(2)浓度预测模型。针对不同浆液循环泵组合下模型特性参数的边缘分布发生变化的场景,建立了基于核均值匹配(KMM)的样本加权最小二乘支持向量机(LSSVM)模型,并与实际运行数据进行了对比验证。结果表明:融合样本迁移权重的LSSVM模型可以有效提升不同浆液循环泵组合运行方式下脱硫系统出口SO_(2)浓度预测模型的精度,并增强了模型的泛化能力。Taking the desulfurization system of a 600 MW unit as an object,a prediction model of SO_(2) concentration at the outlet of wet flue gas desulfurization system based on transfer learning was proposed.In view of the situation that the edge distribution of the characteristic parameters of the model varies under different combinations of slurry circulating pumps,a sample-weighted least squares support vector machine(LSSVM)model based on kernel mean matching(KMM)was established,and a comparison for verification with the actual running data was also conducted.Results show that the prediction accuracy of SO_(2) concentration at the outlet of the desulfurization system under different combination operation modes of slurry circulating pumps can be effectively improved when using the LSSVM model with sample transfer weight,and the generalization ability of the model can also be enhanced.

关 键 词：湿法脱硫 迁移学习 核均值匹配 最小二乘支持向量机 

分 类 号：X701.3[环境科学与工程—环境工程]