基于PIRNN的燃煤电厂锅炉末级过热器壁温预测方法  

A Wall Temperature Prediction Method of Final Superheater of Coal-fired Power Plant Boiler Based on PIRNN

作　　者：张瑞琦 茅大钧 ZHANG Ruiqi;MAO Dajun(School of Automation Engineering,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]上海电力大学自动化工程学院,上海200090

出　　处：《电力科学与工程》2024年第12期63-72,共10页Electric Power Science and Engineering

基　　金：上海市“科技创新行动计划”地方院校能源建设专项项目(19020500700);中国华能集团有限公司2022年度科技项目(HNKJ22-HF22)。

摘　　要：为了解决锅炉末级过热器管壁超温的预测问题,提出基于物理信息递归神经网络(physical informed recurrent neural network,PIRNN)的末级过热器壁温预测方法。首先,通过互信息筛选出系统中的关键特征变量,同时选取以锅炉最大连续蒸发量为指标的4种工况为模型,提供基本运行参数;然后,将传统的数据损失函数结合基于一维热传导、热辐射、热对流的偏微分方程作为物理损失函数,以强化PIRNN网络对物理信息的处理能力;最后,采用随机搜索进行超参数寻优。以国内某在役的660 MW超超临界机组末级过热器为对象进行验证。PIRNN算法验证集表现出均方根差为0.807,相较于其他传统深度学习方法更优,符合物理一致性,更能适应锅炉的变负荷工况。In order to solve the prediction problem of the over-temperature of the tube wall of the final superheater,a method of predicting the wall temperature of the final superheater based on the physical information recurrent neural network(PIRNN)is proposed.Firstly,the key characteristic variables in the system are screened by mutual information,and four operating conditions with the maximum continuous evaporation of the boiler as the indexes are selected to provide basic operating parameters for the model.Secondly,the traditional data loss function combined with the partial differential equation which based on one-dimensional heat conduction,heat radiation and heat convection is used as the physical loss function to strengthen the PIRNN network’s processing capability of physical information.Finally,random search is used for hyperparameter optimization.Taking the final superheater of a 660 MW ultra-supercritical unit in service in China as the object of verification.The verification set of PIRNN algorithm shows a root mean square error of 0.807,which is better than other traditional deep learning methods,conforms to the physical consistency and can better adapt to the variable load conditions of boilers.

关 键 词：一维热传导方程 热辐射 热对流 PIRNN 末级过热器 燃煤锅炉 

分 类 号：TK223.3[动力工程及工程热物理—动力机械及工程] TK39