Estimation of the methanol loss in the gas hydrate prevention unit using the artificial neural networks: Investigating the effect of training algorithm on the model accuracy

Artificial neural networks (ANNs) can understand the behavior of a given system from historical measurements its associated variables. Adjusting weight and bias ANN model using an optimization algorithm is known as training process. The reliability directly related to success Therefore, this study investigates effect algorithms on prediction accuracy multilayer perceptron (MLPNNs). complex gas hydrate prevention unit simulated MLPNN trained by 20 different algorithms. This gradient-based, evolutionary, Bayesian-based Combining statistical ranking analyses confirms that Levenberg–Marquardt (LM) most efficient technique for model. adjusts bis parameters so highest accurate predictions have been achieved. On other hand, imperialist competitive shows lowest considered task. LM predicts 239 laboratory-measured data sets about methanol (MeOH) loss with absolute average relative deviation 6.4% regression coefficient 0.9643. Coupling developed differential evolution temperature = 263 K pressure 6.92 MPa are optimum condition minimizing MeOH in unit. Economic analysis annual cost daily processing 100 × 106 m3 ~17 million US dollars.