Development of aggregated random intelligent approach for the modeling of desalination processes

Abstract

The boiling point rise (BPR) is a critical parameter in the operation and optimal design of a multi-stage flash desalination system. Accurate prediction of BPR would increase the efficiency of freshwater production. In this study, original aggregated random intelligent approach (ARIA) models is developed to enhance the prediction of BPR. The ARIA algorithms, similar to a random forest (RF) technique, combine a set of models trained on different subsets of data in an ensemble structure. Five ARIA-type models were developed based on shallow neural network (SNN), deep neural network (DNN), support vector regression (SVR), classification and regression decision tree (CART), and adaptive neuro-fuzzy inference system (ANFIS). Synthesis index (SI) and visual graphs were used to rank models. Results show that the developed ARIA models are far more accurate than their regular counterparts. The ARIAs increase the prediction efficiency of regular modes by up to 25.04 %. The ARIA-ANFIS with the lowest root mean square error (RMSE) of 0.027 °C outperformed other models and regression equations. The developed ARIA-ANFIS decreases the error of corresponding RF prediction by 69.66 % (RMSE = 0.089 °C), which is a significant achievement because RF is a widely recognized model and numerous studies underline the high accuracies of RF applications.