摘要
针对过程工业中难以直接测量的变量建立其软测量模型,对于实现关键指标的在线监测和实时控制具有十分重要的意义。变量的选择直接关系到神经网络软测量模型的预测性能,针对现有输入变量和网络结构选择方法在工业应用中的不足,提出了一种基于敏感度分析的方法来确定网络输入变量集和前馈神经网络隐含层节点个数,并建立了高密度聚乙烯(HDPE)产品质量指标熔融指数(M I)软测量模型,以实际工业应用验证了该方法的有效性。
Since some important process variables are difficult to be measured in real time for industrial processes, the establishment of soft sensing model plays an important role in on-line monitoring and real-time controlling of the key indicators. The variables selection is relevant to the prediction ability of neural network soft-sensing model. A method based on sensitivity analysis is prolided, aiming to confirm the input variable set and the number of hidden layer nodes more accurately against the deficiency of existing methods in industry application. And the soft-sensing model of Melt Index, which is the indicator of high-density polyethylene product quality is trained and tested by BP Neural Network. The industry example indicates that the method is effective.
出处
《控制工程》
CSCD
北大核心
2011年第3期388-392,共5页
Control Engineering of China
基金
国家自然科学基金项目(60774079)
关键词
软测量
敏感度分析
输入变量选择
隐含层节点
soft sensing
sensitivity analysis
input variable selection
hidden layer nodes
