“…In addition, numerical integration of the DAEs system is complicated and time-consuming, to guarantee the performance accurately and capture the process' dynamic features simultaneously [20], especially dealing with highly nonlinear isotherms, due to numerical dispersion (smearing) and oscillation. All that has also greatly increased the difficulty of process optimization [21].To reduce the computation amounts for simulation and optimization, researchers have proposed a variety of different surrogate models, such as the polynomial surface response model (PRSM) [22], Kriging model [23], proper orthogonal decomposition [24][25][26], polynomial regression model (PNR), support vector regression, and artificial neural network (ANN) model [27,28]. The surrogate model is essentially a black-box model, which is built from a known sample of input-output data points, and can be used to predict the output response at untried points/configurations [25,29].…”