Dye Pollutant Removal from Synthetic Wastewater: A New Modeling and Predicting Approach Based on Experimental Data Analysis, Kriging Interpolation Method, and Computational Intelligence Techniques

Abstract

In the present study, a new approach by coupling the interpolation method with computation-based technique (data-mining algorithms and an optimization algorithm) is introduced for modeling and optimization removal of Reactive Orange 7 (RO7) dye removal from synthetic wastewater. To this end, four significant factors like pH, electrolyte concentration, current density, and electrolysis time are considered as input variables. Thus, modeling of RO7 removal is implemented using eight data mining algorithms including multivariate linear regression (MLR), ridge regression (RR), multivariate nonlinear regression (MNLR), artificial neural network (ANN), classification and regression tree (CART), k nearest neighbor (KNN), random forest (RF), and support vector machine (SVM). These algorithms require a large data set for creating reliable results. However, creating a large number of experimental data request consuming high cost and time. Hence, the interpolation methods of kriging (KRG) and inverse distance weight (IDW) are applied for generating more data, whereas KRG has more accuracy than IDW by increasing the 47.080, 36.914, and 1.77% in MAE, RMSE, and R values, respectively. Then, the data mining algorithms are used for modeling the decolorization efficiency (DE) based on the original data and new data from KRG. It is found that using new data leads to significantly increasing accuracy (94.47, 96.43, 1.52, and 2.77% for MAE, RMSE, R and R 2 , respectively) of DE modeling. Also, SVM has demonstrated the highest accuracy out of all data mining algorithms (by increasing the 97.13, 98.30, and 14.42% in MAE, RMSE, and R 2 values, respectively). Another challenge in the removal of RO7 from synthetic wastewater is predicting the maximum removal amount and optimal input variables. For this purpose, the hybrid of SVM and whale optimization algorithm (WOA) is employed. Finally, SVM-WOA has predicted the maximum of DE (91%) by optimal values of 4.2, 1.5 g/L, 4.2 mA/cm 2 , and 18 min for pH, C, I, and Time, respectively. In light of the high performance of the introduced approach for modeling removal process and predicting optimal conditions of removal process, this approach can be suggested for the removal of other pollutants from wastewater when the number of experimental data set is limited.

Two Birds with One Stone: A Dual-Functional Washing Fluid for Shoreline Oil Spill Response

“…Temperature plays a key role in affecting the transport behavior of pollutants. − The washing performance of the MoS 2 fluid regarding the temperatures is shown in Figure b. The MoS 2 fluid was more efficient for the cleanup of the oiled sands at higher temperatures, removing 42.5% of oil at 5 °C and 70.2% of oil at 25 °C.…”

An Efficient Water Quality Prediction and Assessment Method Based on the Improved Deep Belief Network—Long Short-Term Memory Model

“…The main advantage of methods based on data-driven models is that they can accurately predict the "black box" events and are suitable for systems whose internal mechanisms are too complex to be described using mathematical expressions. Data-driven models have been widely used in many fields since they can learn data relationships using various types of neural network models [11]. In [12], the long short-term memory (LSTM) neural network model was used for time-series prediction, accurately revealing the future development trend of water quality and indicating the application potential of the LSTM model in drinking-water quality prediction.…”

Coupling Machine and Deep Learning with Explainable Artificial Intelligence for Improving Prediction of Groundwater Quality and Decision-Making in Arid Region, Saudi Arabia

“…Over the past 20 years, the use of AI approaches has expanded across various fields. In the context of simulating and predicting water quality, various machine learning algorithms, including adaptive boosting (Adaboost) [24], gradient boosting (GBM) [25], extreme gradient boosting (XGBoost) [26], decision tree (DT) [27], extra trees (ExT) [28], radial basis function (RBF) [29], artificial neural network (ANN) [29,30], random forest (RF) [31], deep feed-forward neural network (DFNN) [23], and convolutional neural network (CNN) [22] have been examined for their efficacy. However, researchers still face the challenge of determining the most suitable techniques for a given problem.…”