Process optimization is necessary in order to decrease energy consumption and production costs. Using surrogate models, rather than mathematical modeling or simulator software, is an effective method to decrease the calculations and the time needed for optimization. Developing an offline data-based surrogate model for the whole response space requires generating a big data set. This itself involves numerous calculations and, therefore, would be too time-consuming. In this paper, the utilization of an online optimization algorithm is addressed for large-scale processes with a high computational burden. In this algorithm, by using of Latin hypercube sampling (LHS) method and the grey wolf meta-heuristic optimization algorithm (GWO) in combination with the support vector machine (SVM), a suitable balance between exploration and exploitation abilities is achieved. For comparison, the value of the objective function in the estimated global optimum point (GOP) and the number of objective function evaluations (NFEs) required to converge to GOP are investigated. The large-scale gas to liquids (GTL) process plant is chosen as a case study. The results showed that in the online method, while decreasing NFEs to less than one-tenth of the offline method, the GOP is found with a relative error of 0.1 percent.

Keywords