The development of process optimization is essential for optimal energy consumption, production cost reduction, and product generation maximization. Modeling and simulation of a large scale gas to liquids (GTL) process involves numerous complex mathematical calculations. Accordingly, finetuning and optimizing the key parameters of the GTL process is computationally very demanding and time consuming. To alleviate this problem, this study first develops an artificial neural network (ANN) model of the GTL process. The inputs to this model are tail gas unpurged ratio, recycled tail gas to FT ratio, H 2 O/C entering the syngas section, and CO 2 removal percentage, and the ANN model quickly yet precisely estimates the wax production rate. This surrogate model is then imbedded into an optimization problem where the purpose is to maximize the wax production rate by finding the optimal values for the key parameters of the GTL process. The genetic algorithm (GA) is applied for effectively searching the parameter space and finding the global optimum solution. Simulation results indicate that an ANN with a structure of 4:7:15:1 achieves the best prediction performance (mean squared error less than 0.0006). The relative error of estimating the optimum value by the ANN is approximately 0.057%, which is an acceptable value. In addition, optimal GTL parameters found by the proposed ANN-GA technique improves the wax production rate (+107 kg/h). Last but not least, the optimization elapsed has been significantly reduced from about several days to less than a few seconds.

High computational loads, time-consuming convergence, and simulation crashes are common when using process simulators for flowsheet optimization. In this paper, by replacing the large-scale physical process simulations by surrogate models, the optimization time and computational load are reduced significantly along with maintaining the accuracy and reliability. A gas-to-liquids (GTL) plant was used as a large-scale process plant case study. The multilayer perceptron neural network (MLP-ANN), radial basis function neural network, support vector machine, and adaptive neuro-fuzzy inference system models were selected as alternative surrogate models. These alternatives were investigated for implementation of the self-optimizing control procedure on the above case study to find the best individual and combined self-optimizing controlled variables. The MLP-ANN surrogate model showed the best performance in predicting the optimum points and for selecting the best self-optimizing CVs. In fact, it even performed better than using the full process simulator.