基于机理和数据混合模型的SD-WFGD系统SO_(2)浓度动态预测  

Dynamic Prediction for SO_(2) Concentration in SD-WFGD System Based on Mechanism-Data Combined Model

作　　者：李瑞连 曾德良[1] 张光明 谢衍 朱岩松 朱红成 LI Ruilian;ZENG Deliang;ZHANG Guangming;XIE Yan;ZHU Yansong;ZHU Hongcheng(School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China;Beijing Huairou Laboratory,Beijing 101499,China;State Power Investment Corporation Inner Mongolia Energy Co.,Ltd.,Tongliao 029000,Inner Mongolia Autonomous Region,China)

机构地区：[1]华北电力大学控制与计算机工程学院,北京102206 [2]北京怀柔实验室,北京101499 [3]国家电投集团内蒙古能源有限公司,内蒙古通辽029000

出　　处：《动力工程学报》2024年第7期1118-1128,1136,共12页Journal of Chinese Society of Power Engineering

基　　金：国家自然科学基金重点资助项目(61833011)。

摘　　要：为了保证燃煤电站脱硫系统SO_(2)排放浓度达到环保要求的同时实现经济运行,建立了单塔双循环湿法烟气脱硫(SD-WFGD)系统出口SO_(2)浓度混合动态预测模型。首先,分析吸收SO_(2)的化学反应过程,分别建立了吸收塔出口和浆液池(AFT塔)出口SO_(2)浓度机理模型;然后,利用变分模态分解(VMD)方法对历史运行数据和机理模型预测偏差数据进行分解,将分解后不同频率的模态分量进行重构,利用最小二乘支持向量机(LSSVM)算法训练并得到不同的模态分量模型,将不同模态分量的LSSVM模型进行加权叠加,建立了数据补偿模型;最后,将数据补偿模型输出和机理模型输出进行叠加,得到SD-WFGD系统SO_(2)浓度混合动态预测模型。结果表明:利用VMD算法将历史数据分解为不同模态,不同模态数据重构后可有效提高数据模型的预测精度;同时,将机理模型和数据模型结合,可提升模型预测能力。In order to achieve the economic operation and ensure that the SO_(2) emission concentration of desulfurization system meets environmental requirements in a coal-fired unit,a combined dynamic prediction model was established for SO_(2) concentration at the outlet of the single-tower double-cycle wet fuel gas desulfurization(SD-WFGD)system.Firstly,the chemical reaction process of absorbing SO_(2) was analyzed,and the SO_(2) concentration mechanism models at the outlet of the absorption tower and absorber feed tank(AFT tower)were established respectively.Secondly,the historical data and mechanism deviation data were decomposed using variational mode decomposition(VMD)method,and the decomposed arrays with different frequencies were reconstructed.After which,models with different modal components were trained and obtained based on least square support vector machine(LSSVM)algorithm,while a data compensation model was established based on adaptive weight allocation strategy by the weighted stacking of LSSVM models with different modal components.Finally,a combined dynamic prediction model for SO_(2) concentration in the SD-WFGD system was obtained through the superposition of the outputs of dynamic compensation model and mechanism model.Results show that after the decomposition of historical data into different modes by VMD algorithm,the prediction accuracy of data model can be improved effectively with the reconstruction of data in different modes.Meanwhile,the model prediction ability can be improved with the combination of mechanism model and data model.

关 键 词：燃煤电站 SD-WFGD系统 SO_(2)预测 机理建模 

分 类 号：TM621.6[电气工程—电力系统及自动化]