基于平稳子空间分析的特征融合及非平稳过程监测OA

Chemical processes often exhibit significant non-stationary characteristics due to their complex internal mechanisms,which poses substantial challenges to multivariate statistical monitoring methods.Stationary subspace analysis(SSA),a technique that constructs monitoring models by extracting stationary components from signals,has been widely used in non-stationary process monitoring.However,traditional SSA methods mainly focus on constructing statistics within the stationary subspace,often neglecting the fault features contained in the non-stationary subspace,which can result in the loss of certain fault information.To address this limitation,an SSA-stacked autoencoder(SAE)-support vector data description(SVDD)integrated monitoring framework was proposed.First,the process data was mapped into stationary and non-stationary subspaces via SSA.Subsequently,monitoring statistics were constructed directly in stationary subspace,while SAE-based reconstruction errors were established in non-stationary subspace.Finally,comprehensive monitoring was achieved by applying SVDD to dual-space statistical indicators.The proposed method was validated using a real industrial process and compared with existing non-stationary monitoring methods,demonstrating superior performance.