论文部分内容阅读
针对多向主元分析(MPCA)不能提取复杂的非线性系统变量间的非线性特性以及T2统计量置信限的确定是以主元得分呈正态分布为假设前提的情况,提出了一种基于自组织神经网络与核密度估计的非线性MPCA在线故障监测方法.该方法用自组织神经网络去提取变量间的非线性特征信息;用核概率密度函数去估计非线性主元的置信限.将该方法应用到β-甘露聚糖酶补料分批发酵过程的在线故障监测中,应用效果表明用非线性主元比用同样数目的线性主元能够获取更多的变量信息,并且用核密度估计置信限的方法比用参数估计的方法能更准确地对故障进行监测.
For MPCA can not extract the nonlinear characteristics of complex nonlinear system variables and the determination of T2 statistic confidence limits is based on the assumption that the principal component score is normal distribution as a prerequisite, Nonlinear MPCA online fault monitoring method based on self-organizing neural network and kernel density estimation This method uses self-organizing neural network to extract the nonlinear characteristic information between variables and uses kernel probability density function to estimate the confidence limits of nonlinear principal components. The method is applied to the on-line fault monitoring of fed-batch fermentation of β-mannanase. The application results show that more variable information can be obtained by using nonlinear principal components than by the same number of linear principal components, The method of estimating the confidence limit can more accurately monitor the fault than the method of parameter estimation.