A two-stage approach of multiplicative dimensional reduction and polynomial chaos for global sensitivity analysis and uncertainty quantification with a large number of process uncertainties
Uncertainties associated with estimates of model parameters are inevitable when simulating and modeling chemical processes and significantly affect safety, consistency, and decision making. Quantifying those uncertainties is essential for emulating the actual system behaviors because they can change the management recommendations that are drawn from the model. The use of conventional approaches for uncertainty quantification (e.g., Monte-Carlo and standard polynomial chaos methods) is computationally expensive for complex systems with a large/moderate number of uncertainties. This paper develops a two-stage approach to quantify the uncertainty of complex chemical processes with a moderate/large number of uncertainties (greater than 5). The first stage applies a multiplicative dimensional reduction method to approximate the variance-based global sensitivity measures (Sobol's method), and to simplify the model for the uncertainty quantification stage. The second stage uses the generalized polynomial chaos approach to quantify uncertainty of the simplified model from the first stage. A rigorous simulation illustrates the proposed approach using an interface between MATLAB and HYSYS for three complex chemical processes. The proposed method was compared with conventional approaches, such as the Quasi Monte-Carlo sampling-based method and standard polynomial chaos-based method. The results revealed the clear advantage of the proposed approach in terms of the computational efforts.