Search

Many industrial processes are nonlinear distributed parametersystems(DPSs). Data-based spatiotemporal modeling is required for analysisand control when the first-principles model is unknown. Because aDPS is infinite-dimensional and time–space coupled, a low-ordermodel is necessary for prediction and control in practice. For low-ordermodeling, traditional principal component analysis (PCA) is oftenused for dimension reduction and time–space separation. However,it is a linear method and leads to only one set of fixed spatial basisfunctions. Therefore, it might not be always effective for nonlinearsystems. In this study, a spatiotemporal multimodeling approach isproposed for unknown nonlinear DPSs. First, multimodel decompositionis performed, where probabilistic PCA (PPCA) is used to obtain multiplesets of spatial basis functions from the experimental data by maximizinga likelihood function. Using these multiple sets of PCA spatial basesfor time–space separation, the high-dimensionality spatiotemporaldata can be reduced to multiple sets of low-dimensionality temporalseries. Then, multiple low-order neural models can be easily establishedto model these local dynamics. Finally, the original spatiotemporaldynamics can be reconstructed by multimodel synthesis. Because theproposed spatiotemporal modeling approach involves a multimodelingmechanism, it can achieve better performance than the traditionalPCA-based single-modeling for nonlinear DPSs, which is demonstratedby numerical simulations. [ABSTRACT FROM AUTHOR]