Experimental Study and Modeling of Ultrafiltration of Refinery Effluents Using a Hybrid Intelligent Approach

Abstract

This study aims at indicating the capability of a state-of-the-art computational intelligence approach for predicting pseudosteady flux and pseudosteady fouling at different operating condition (temperature (T), transmembrane pressure (TMP), cross-flow velocity (CFV), and feed pH) as well as for permeate flux decline at the mentioned operational conditions with processing filtration time. To train and test these models, the experimental data collected during the polyacrylonitrile (PAN) UF process to treat the oily wastewater of a Tehran refinery have been used. The proposed method utilizes a least-squares support vector machine (LSSVM) to carry out nonlinear modeling. The genetic algorithm (GA) was employed to tune the optimal model parameters. GA-LSSVM has the competence of describing the nonlinear behavior. The accuracy of the proposed GA-LSSVM models is very satisfactory and quantified by statistical parameters. Finally, the results obtained by implementing various sensitivity analysis techniques portrayed that T and TMP have the most significant influence on pseudosteady flux and pseudosteady fouling, correspondingly, in comparison with other factors involved in the addressed treatment process.

Application of artificial neural network coupled with genetic algorithm and simulated annealing to solve groundwater inflow problem to an advancing open pit mine

“…The general relationship between input and output in an ANN model can be expressed as (Fazeli et al, 2013):…”

“…The objective is to find the value of the weight that minimises differences between the real output and the predicted output in the output layer in order to minimise the mean square errors (MSEs), the average squared error between the network predictions and the target outputs . In order to find the optimal weight ( ) and the bias ( ), training or learning processes must be implemented to minimise the error (Fazeli et al, 2013).…”

Application of foam floods for enhancing heavy oil recovery through stability analysis and core flood experiments

“…[69][70][71] During recent decades, various predictive models have been proposed, including artificial neural network, [72,73] adaptive neuro-fuzzy inference systems, [74,75] and the least square supported vector machine (LS-SVM). [65,76] The main advantage of LS-SVM model is that it does not need to determine the network topology and it can automatically be established as the training process finishes. In addition, the LS-SVM is a two parametric model that this prevents the network to over-fit.…”

Connectionist model for predicting minimum gas miscibility pressure: Application to gas injection process

“…In the current study, Least Square Support Vector Machine (LS-SVM) and evolutionary 45 algorithms (for example, Genetic Algorithm (GA) and Imperialist Competitive Algorithm), both addressed 46 in previous literatures, have been employed to estimate the MMP. A set of laboratorial data accessible in 47 the open literature was gained to test the reliability of the proposed HGAPSO-LSSVM model which its 48 generated results have been compared with the other proposed intelligent approaches. Moreover, the 49 performances of both implemented solutions certify statistically the strong potential of models in predic- 50 tion of the MMP.…”

“…By reaching to the MMP, the displacement is piston-like 67 and the oil recovery is 100% at 1 pore volume of the injected gas, 68 if the displacement process is represented as a one dimensional, 69 two-phase, dispersion-free flow [2-4]. 70 Optimum displacement efficiency of gas flooding happens at 71 whereas following constraints should be considered [45][46][47][48][49][50]: The assembled real database was separated into three subsets. 280 The first, which is utilized in the process of training, contains 281 80% of the whole data (68 data point) and is equal to 55 data lines.…”