Fig 3 - uploaded by J. Navarro-Laboulais
Content may be subject to copyright.
Scheme showing the process for preparing data, training the artificial neural network (ANN) and using it to estimate the initial parameters. The same operational conditions set to obtain the experimental data are used in the simulation procedure. Then the initial values determined by the ANN are used in the Levenberg-Marquardt (LM) algorithm for the definitive fitting of the parameters and their errors.
Source publication
A procedure for the determination of initial parameter values for quadratically convergent optimization methods is proposed using artificial neural networks coupled with a non-stationary gas-liquid reaction model. The evaluation of the regression and the mean squared error coefficients of the neural network during its training process allow the par...
Context in source publication
Citations
... In ANN modeling, the mathematical description involved in the process is not required which makes it useful in simulating and up-scaling complex photochemical systems (Despagne and Massart, 1998). Also, ANN is a strong tool to model the complicated reaction mechanisms in AOPs due to the lack of selectivity in the hydroxyl radical activity (Madaeni et al., 2012;Manokaran et al., 2014;Ferre-Aracil et al., 2015;Gazi et al., 2017), while other simple phenomenological kinetic models are not useful (Brandão et al., 2013). The AOPs performance are affected by several operational parameters, while the ability of a simple kinetic model is restricted to the simulation of the process during the reaction time under a certain operational condition (conventionally optimum) which its related experimental data has been analyzed. ...
This study deals with the modeling and simulation of an efficient pilot plant photo-chemical wastewater treatment reactor. Treatment of an azo dye (i.e. direct red 23) was performed using a UV/H2O2 process in a jet mixing photo-reactor with 10-L volume. To model the reactor and simulate the treatment process, six important, influential physical and chemical factors such as nozzle angle (θN), nozzle diameter (dN), flow-rate (Q), irradiation time (t), H2O2 initial concentration ([H2O2]0), and pH, were taken into account. In this regard, artificial neural networks (ANNs) were employed as a powerful modeling methodology. Six different ANN architectures were constructed and most appropriate numbers for hidden neuron and learning iteration were determined based on minimization of the mean square error (MSE) function related to the testing data sets. Furthermore, simulation of the reactor efficiency, as well as sensitivity analysis, was performed via the cross-validation outputs. It was found that a three-layered feed-forward ANN composes ten hidden neurons, calibrated at 100th iteration using “trainlm” as learning algorithm and “tansig” and “purelin” as transfer functions in the hidden and output layers can model the process as the best case. The order of importance for variation of the key factors were indicated as [H2O2]0> t > pH > Q > θN > dN.
... The input parameters were temperature, substrate concentration, pH, and hydrolysis time while the output layer was reducing sugar concentration. The first layer receives information from the input parameters, and each next layer receives data from the input layer and all previous layers (Aracil et al., 2015;Nikzad et al., 2015). ...
The efficient production of reducing sugars is an extremely important requirement in the utilization of microalgae as a feedstock in bioethanol production. In this study, for the first time, the time course of reducing sugar production during starch hydrolysis of mixed microalgal biomass under different operational conditions was modeled by two different methods; a-Michaelis-Menten kinetic model and b-artificial neural network (ANN) method. The results from both models revealed that predicted values are in good agreement with the experimental results. Also, sensitivity analysis indicated that the kinetic model results are less sensitive to Km and Ki than to . The applied ANN was a feed-forward back propagation network with Levenberg-Marquardt algorithm. It was found that the order of relative importance of the input variables on reducing sugar concentration predicted by ANN model was as follows: pH>substrate concentration>temperature>hydrolysis time. Subsequently, the results indicated that the maximum reducing sugar yield (96.3%) was achieved by adding enzymes with the sequence of first cellulases, at 50 °C, pH 4.5, second α -amylase, at 70 °C, and pH 6 with a substrate concentration of 50 g/L. These findings may be useful for improving the enzymatic hydrolysis of mixed microalgae for bioethanol production.
... y A0 is the mole fraction of ozone gas at the reactor inlet which is measured at the by-pass of the reactor previously to the introduction into it, y A1 is the mole fraction of ozone gas in the bottom gas mixing chamber of the reactor, y A2 is the mole fraction of ozone gas inside the reactor, i.e. in the bubbles, C A is the molar concentration of dissolved ozone in the liquid, C B is the organic bulk matter concentration and y A3 is the mole fraction of ozone gas measured at the outlet of the upper gas mixing chamber in the reactor (see Ferre-Aracil et al., 2015 and references therein). ...
In the present study, a feed-forward dense multilayer artificial neural network (ANN) trained with backpropagation algorithm (BP) is used to determine the kinetic parameters governing the performance of the oil-drilling cuttings (ODC) ozonation. Ozonation tests of ODC, pre-treated with surfactant (SDS) and diluted with synthetic seawater, are conducted on a semi-batch bubble flow reactor. The performance of ozonation experiments on semi-batch bubble flow reactors is evaluated by measuring the removal efficiency of the total organic carbon (TOC). The experimental datasets are employed to calibrate two mathematical models of increasing complexity: a tank-in-series (TSM), and a computational fluid dynamics (CFM) model, both combining multiphase transport with reactive processes, involving a high number of unknown parameters, and being able to provide numerous simulated datasets for various values of dimensionless parameters. These data are employed for the training and validation of the neural network. The trained neural network is then interpreted through the Shapley additive explanation (SHAP) method to provide insights about the experimental parameters.
Novel magnetic bayberry-like Fe3O4/Bi2S3 microspheres (Fe3O4/Bi2S3 MSs) combining highly effective adsorption and photocatalytic regeneration were prepared by a facile hydrothermal method. The resultant Fe3O4/Bi2S3 MSs were characterized by XRD, SEM, EDS, VSM, BET and DRS. Fe3O4/Bi2S3 MSs possessed a relative large surface area of 36.0 m² g⁻¹ and narrow pore size distribution around 4.72 nm. The equilibrium and kinetics of adsorption process followed the Langmuir isotherm model and pseudo-second-order kinetic model, respectively. Maximum adsorption capacity of CR as 92.24 mg g⁻¹ was achieved on Fe3O4/Bi2S3 MSs, while only 66.28 mg g⁻¹ was found on Bi2S3 MSs. High saturation magnetization, low coercivity and remnant magnetization values of Fe3O4/Bi2S3 MSs indicated that easy separation and fast re-dispersion in aqueous solution can be realized. What’s more important, Fe3O4/Bi2S3 MSs with a stronger absorption in the visible light region can be regenerated by photocatalysis under simulated solar light irradiation. Fe3O4/Bi2S3 MSs also showed excellent stability and reusability for continuous removal of congo red dye by synergistic adsorption and photocatalytic regeneration. As a result, Fe3O4/Bi2S3 MSs provided effective and conveniently recyclable materials for environmental remediation by means of providing the facile preparation, easy magnetic separation, high adsorption, and simple regeneration.
