基于正则贯序极端学习机的换热设备污垢预测  

Fouling Prediction of Heat Exchanger based on Sequential Regularized Extreme Learning Machine

作　　者：孙灵芳[1] 陶苗苗[1] 朴亨[1] 

机构地区：[1]东北电力大学自动化工程学院,吉林吉林1320122

出　　处：《东北电力大学学报》2015年第4期84-90,共7页Journal of Northeast Electric Power University

摘　　要：对换热设备积聚的污垢快速、准确的预测,可以为换热设备污垢的监测和解决对策提供指导和依据,进而避免污垢对换热设备安全经济运行带来的不利影响。通过在线贯序极端学习机理论建立换热设备污垢预测模型,利用初始样本建立初始模型,随着工况变化,不断加入新的样本集,更新预测模型,实现换热设备污垢的在线预测。并通过引入正则参数权衡经验风险和结构风险,提高模型的泛化性能和预测精度。与传统神经网络建立的预测模型相比,基于在线贯序极端学习机的换热设备污垢预测模型训练速度更快,精度更高,泛化能力更好。为此,进一步基于MATLAB和LABVIEW混合编程设计了正则贯序极端学习机污垢预测系统,结果表明,该污垢预测系统可以实现换热设备污垢的在线实时预测,并且预测精度较高,预测效果较好。A quick and accurate prediction of heat exchanger fouling can provide support and direction for fouling monitoring and countermeasures, and can avoid the adverse effects of safe opperation on heat exchanger. The prediction model of heat exchanger fouling is built in this paper, according to the Online Sequential Regularized Extreme Learning Machine （SRELM） , using the initial sample to establish the initial model, the model was updated on the basis of input new samples with the changes of working conditions, accomplished online prediction of heat exchanger fouling. The generalization performance and prediction accuracy of the model was improved by the means introducing a regularization parameter to balance the empirical risk and structural risk. Simulation results show that the prediction model based on SRELM has a faster training speed and better generalization performance and higher prediction accuracy compared with the prediction model based on traditional neural network. The SRELM fouling prediction system was designed based on MATLAB and LABVIEW hybrid programming. The results show that the system can online in real time to predict with higer predict accuracy.

关 键 词：污垢 在线预测 极端学习机 污垢预测系统 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]