摘要
提出一种基于模糊聚美-最小二乘向量机-神经网络(FCM-LSSVM-ANN)的多模型融合方法,对全工况下选择性催化还原(SCR)入口温度进行提前预测。采用模糊聚类对不同工况下的锅炉系统运行数据进行分解,并建立若干个基于最小二乘支持向量机的预测模型,最后采用神经网络对预测结果进行非线性融合得到最终预测结果。多模型融合的方法可以对锅炉系统全工况的运行特性进行学习,能更准确地完成负荷大范围波动条件下SCR入口温度预测。同时本文采用某600 MW机组实际运行数据对所提方法进行对比验证,结果表明本文方法能够实现该机组30%~100%负荷范围SCR入口温度的准确预测,平均预测偏差控制在±4℃以内,相对误差大多数情况下小于1%。本模型可以为燃煤机组深度调峰下脱销系统入口温度进行提前预警。
This paper proposes a multi-model fusion method based on FCM-LSSVM-ANN and the idea of "divide and conquer" to predict SCR inlet temperature under full working conditions in advance.Firstly,fuzzy clustering is used to decompose the operation data of boiler system under different working conditions,and then several prediction models based on the least square support vector machine are established.Finally,the neural network is used to make the nonlinear fusion of the prediction results to obtain the final prediction results.The multi-model fusion method can learn the operation characteristics of the boiler system in all working conditions,so that the SCR inlet temperature can be predicted more accurately under a large range of load fluctuations.Meanwhile,the actual operation data of a 600 MW unit was used to verify the proposed method.The results showed that the proposed method could accurately predict the SCR inlet temperature under the load range of 30%~100%,and the average deviation was controlled within±4 ℃ and the relative error is mostly less than 1%.This model provides an early warning for the inlet temperature of the denitration system under deep peak load regulation of coal-fired units.
作者
王亚欧
蔡亮
胡忠旭
任少君
WANG Ya-ou;CAI Liang;HU Zhong-xu;REN Shao-jun(Jangsu Frontier Electric Technology Co.Ltd.,Nanjing,China,211102;Key Laboratory of Energy Thermal Conversion and Process Measurement and Control,Ministry of Education,Southeast University,Nanjing,China,210096)
出处
《热能动力工程》
CAS
CSCD
北大核心
2020年第8期96-103,共8页
Journal of Engineering for Thermal Energy and Power
基金
国家自然科学基金(51976031)。
关键词
深度调峰
SCR入口温度
线性支持向量机
模糊聚类
ANN
deep peak load regulation
SCR inlet temperature
least square support vector machine
fuzzy clustering
ANN
