A New Soft Sensor Based on Dynamic Principal Component Analysis and on-line Potential Fuzzy Clustering

In this paper, a new soft sensor methodology is proposed for estimation of product concentration in a chemical process. The new soft sensor utilizes dynamic principle component analysis (DPCA) method to select the optimum reduced process variables as the appropriate inputs. DPCA eliminates the high correlations among the process variable measurements, leading to a lower dimensional uncorrelated principle components of the process measurements. The DPCA transformed measurements are then used to train a global nonlinear fuzzy model, based on an on-line potential fuzzy clustering approach, for unmeasurable variable estimation. The developed soft sensor performance is demonstrated on a simulated distillation column benchmark problem.