改进的非线性PCA方法及其在过程监控中的应用

针对化工聚合反应过程的特点，结合小波分解多分辨率特性和独立元分析(ICA)提取个数较少的相互独立信号的优点，改进了基于自相关神经元网络的非线性主元分析(NLPCA)方法。在传统的非线性PCA方法中引入了独立元分析模块，不仅解决了自相关神经元网络中确定各层神经元个数的问题，而且以最少的独立元个数捕捉数据的非线性特征。多尺度监控可以识别各种幅值的故障，提高了监控效果。在此基础上，计算I²、I²_e和SPE统计量用于故障检测。贡献图法用于识别故障变量。在聚酯生产过程上的仿真结果表明，改进后的方法比传统的非线性PCA方法更及时地检测到过程故障，运用贡献图可以有效地实现故障变量分离。

An improved nonlinear principal component analysis (PCA) strategy and its application in chemical process monitoring

An improved nonlinear principal component analysis(NLPCA) strategy,based on an auto-associative neural network and combining the merits of wavelet analysis and independent component analysis(ICA),has been derived in order to study chemical polymerization processes.An ICA procedure was incorporated in order to capture the nonlinear characteristics with a minimum number of independent components.It was shown that multi-scale monitoring is effective in detecting different faults and leads to an increase in the reliability of process monitoring;I2,I2e and SPE statistics were subsequently constructed for fault detection.A contribution plot method was used to identify the faulty variables.The simulation results for a polyester process showed that the improved method can detect process faults earlier than traditional NLPCA,and that faulty variables can be indicated exactly by the contribution plot.