摘要
针对烟气含氧量测量成本高、测量不稳定等问题,依据深度学习理论,采用非线性组合深度置信网络(nonlinear combined deep belief network, NCDBN)方法建立烟气含氧量模型。在该方法中,将输入变量分为控制变量和状态变量。对原始数据进行归一化预处理之后,采用lasso算法选取相关性强的变量作为预测模型输入参数。然后,采用DBN算法分别建立控制变量预测模型和状态变量预测模型。最后,将两个预测模型进行非线性组合,获得烟气含氧量的最终预测模型。根据实际生产数据进行实验,结果表明4种对比算法的平均绝对误差分别为1.319%,2.5103%,1.9586%,5.4634%,2.5350%,而NCDBN方法的平均绝对误差为1.2428%,说明NCDBN方法能够准确地预测烟气含氧量。
To solve the problem of high cost and instability of the oxygen content measurement,the nonlinear combined deep belief network(NCDBN) method was used to establish the oxygen content of flue gas model based on the deep learning theory. In this method, input variables are divided into control variables and state variables. The original data were normalized, The variables which have significant correlation with the oxygen content of flue gas were selected as inputs of the prediction model by a lasso algorithm. Then,the deep belief network(DBN) was addressed to established predicted models using control variables and state variables separately. Finally,two models were nonlinearly combined into the final predicted model. The practical data obtained from actual production were utilized in the experiments. The experimental results illustrate that the mean absolute percent error of four common-used algorithms were 1.319%,2.5103%,1.9586%,5.4634%,2.5350%, while the mean absolute percent error of NCDBN method was 1.2428%. The result show that NCDBN method can accurately predict oxygen content of flue gas.
作者
唐振浩
李艳艳
曹生现
TANG Zhen-hao;LI Yan-yan;CAO Sheng-xian(School of Automation Engineering,Northeast Electric Power University,Jilin 132012,China)
出处
《哈尔滨理工大学学报》
CAS
北大核心
2020年第5期127-135,共9页
Journal of Harbin University of Science and Technology
基金
国家自然科学基金(61503072)
吉林市科技创新基金(20166009)。
