Digital imaging-in-flow (FlowCAM) and probabilistic machine learning to assess the sonolytic disinfection of cyanobacteria in sewage wastewater.

Assessing the cyanobacteria disinfection in sewage and its compliance with international-standards requires determining the concentration and viability, which can be achieve using Imaging Flow Cytometry device called FlowCAM. The objective is to thoroughly investigate the sonolytic morphological changes and disinfection-performance towards toxic cyanobacteria existing in sewage using the FlowCAM. After optimizing the process conditions, over 80% decline in cyanobacterial cell counts was observed, accompanied by an additional 10-15% of cells exhibiting injuries, as confirmed through morphological investigation. Moreover, for the first time, the experimentally collected data was utilized to build deep-learning probabilistic-neural-networks (PNN) and natural-gradient-boosting (NGBoost) models for predicting disinfection efficiency and ABD area as target outputs. The findings suggest that the NGBoost model exhibited superior prediction performance for both targets, with high test coefficient of determination (R2 > 0.87) and lower test errors (RMSE < 7.10, MAE < 4.14). The confidence interval examination in NGBoost prediction performance showed a minute variation from the experimentally calculated values, suggesting a high accuracy in model prediction. Finally, SHAP analysis suggests the sonolytic time alone contributes around 50% to the cyanobacteria disinfection. Overall, the findings demonstrate the effectiveness of the FlowCAM device and the potential of machine-learning modeling in predicting disinfection outcomes.

Machine learning analysis to interpret the effect of the photocatalytic reaction rate constant (k) of semiconductor-based photocatalysts on dye removal.

Photocatalytic reactions with semiconductor-based photocatalysts have been investigated extensively for application to wastewater treatment, especially dye degradation, yet the interactions between different process parameters have rarely been reported due to their complicated reaction mechanisms. Hence, this study aims to discern the impact of each factor, and each interaction between multiple factors on reaction rate constant (k) using a decision tree model. The dyes selected as target pollutants were indigo and malachite green, and 5 different semiconductor-based photocatalysts with 17 different compositions were tested, which generated 34 input features and 1527 data points. The Boruta Shapley Additive exPlanations (SHAP) feature selection for the 34 inputs found that 11 inputs were significantly important. The decision tree model exhibited for 11 input features with an R2 value of 0.94. The SHAP feature importance analysis suggested that photocatalytic experimental conditions, with an importance of 59%, was the most important input category, followed by atomic composition (39%) and physicochemical properties (2%). Additionally, the effects on k of the synergy between the metal cocatalysts and important experimental conditions were confirmed by two feature SHAP dependence plots, regardless of importance order. This work provides insight into the single and multiple factors that affect reaction rate and mechanism.

Long short-term memory models of water quality in inland water environments.

Water quality is substantially influenced by a multitude of dynamic and interrelated variables, including climate conditions, landuse and seasonal changes. Deep learning models have demonstrated predictive power of water quality due to the superior ability to automatically learn complex patterns and relationships from variables. Long short-term memory (LSTM), one of deep learning models for water quality prediction, is a type of recurrent neural network that can account for longer-term traits of time-dependent data. It is the most widely applied network used to predict the time series of water quality variables. First, we reviewed applications of a standalone LSTM and discussed its calculation time, prediction accuracy, and good robustness with process-driven numerical models and the other machine learning. This review was expanded into the LSTM model with data pre-processing techniques, including the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise method and Synchrosqueezed Wavelet Transform. The review then focused on the coupling of LSTM with a convolutional neural network, attention network, and transfer learning. The coupled networks demonstrated their performance over the standalone LSTM model. We also emphasized the influence of the static variables in the model and used the transformation method on the dataset. Outlook and further challenges were addressed. The outlook for research and application of LSTM in hydrology concludes the review.

Machine learning approaches to predict the photocatalytic performance of bismuth ferrite-based materials in the removal of malachite green.

This study focuses on the potential capability of numerous machine learning models, namely CatBoost, GradientBoosting, HistGradientBoosting, ExtraTrees, XGBoost, DecisionTree, Bagging, light gradient boosting machine (LGBM), GaussianProcess, artificial neural network (ANN), and light long short-term memory (LightLSTM). These models were investigated to predict the photocatalytic degradation of malachite green from wastewater using various NM-BiFeO3 composites. A comprehensive databank of 1200 data points was generated under various experimental conditions. The ten input variables selected were the catalyst type, reaction time, light intensity, initial concentration, catalyst loading, solution pH, humic acid concentration, anions, surface area, and pore volume of various photocatalysts. The MG dye degradation efficiency was selected as the output variable. An evaluation of the performance metrics suggested that the CatBoost model, with the highest test coefficient of determination (0.99) and lowest mean absolute error (0.64) and root-mean-square error (1.34), outperformed all other models. The CatBoost model showed that the photocatalytic reaction conditions were more important than the material properties. The modeling results suggested that the optimized process conditions were a light intensity of 105 W, catalyst loading of 1.5 g/L, initial MG dye concentration of 5 mg/L and solution pH of 7. Finally, the implications and drawbacks of the current study were stated in detail.

An open-source deep learning model for predicting effluent concentration in capacitive deionization.

To effectively evaluate the performance of capacitive deionization (CDI), an electrochemical ion separation technology, it is necessary to accurately estimate the number of ions removed (effluent concentration) according to energy consumption. Herein, we propose and evaluate a deep learning model for predicting the effluent concentration of a CDI process. The developed deep learning model exhibited excellent prediction accuracy for both constant current and constant voltage modes (R2 ≥ 0.968), and the accuracy increased with the data size. This model was based on the open-source language, Python, and the code has since been distributed with proper instructions for general use. Owing to the nature of the data-oriented deep learning model, the findings of this study are not only applicable to conventional CDI but also to various types of CDI (membrane CDI, flow CDI, faradaic CDI, etc.). Therefore, by referring to the examples shown in this study, we hope that this open-source deep learning code will be widely used in CDI research.

Prediction of biogas production in anaerobic co-digestion of organic wastes using deep learning models.

Interest in anaerobic co-digestion (AcoD) has increased significantly in recent decades owing to enhanced biogas productivity due to the utilization of different organic wastes, such as food waste and sewage sludge. In this study, a robust AcoD model for biogas prediction is developed using deep learning (DL). We propose a hybrid DL architecture, i.e., DA-LSTM-VSN, wherein a dual-stage-attention (DA)-based long short-term memory (LSTM) network is integrated with variable selection networks (VSNs). To enhance the model predictability, we perform hyperparameter optimization. The model accuracy is validated using long-term AcoD monitoring data measured over two years of municipal wastewater treatment plant operation and then compared with those of two other DL-based models (i.e., DA-LSTM and the standard LSTM). In addition, the feature importance (FI) is analyzed to investigate the relative contribution of input variables to biogas production prediction. Finally, we demonstrate the successful application of the validated DL model to the AcoD process optimization. Results show that the model accuracy improved significantly by incorporating DA into LSTM, i.e., the coefficient of determination (R2) increased from 0.38 to 0.68; however, the R2 can be further increased to 0.76 by combining DA-LSTM with a VSN. For the biogas prediction of the AcoD model, the VSN contributes significantly by employing the discontinuous time series of measurement data on biodegradable organic-associated variables during AcoD. In addition, the VSN allows the AcoD model to be interpretable via FI analysis using its weighted input features. The FI results show that the relative importance is vital to variables associated with food waste leachate, whereas it is marginal for those associated with the primary and chemically assisted sedimentation sludges. In conclusion, the AcoD model proposed herein can be utilized in practical applications as a robust tool because it can provide the optimal sludge conditions to improve biogas production. This is because it facilitates the time-series biogas prediction at the full scale using unprocessed datasets with either missing value imputation or outlier removal.

Prediction of antibiotic-resistance genes occurrence at a recreational beach with deep learning models.

Antibiotic resistance genes (ARGs) have been reported to threaten the public health of beachgoers worldwide. Although ARG monitoring and beach guidelines are necessary, substantial efforts are required for ARG sampling and analysis. Accordingly, in this study, we predicted ARGs occurrence that are primarily found on the coast after rainfall using a conventional long short-term memory (LSTM), LSTM-convolutional neural network (CNN) hybrid model, and input attention (IA)-LSTM. To develop the models, 10 categories of environmental data collected at 30-min intervals and concentration data of 4 types of major ARGs (i.e., aac(6'-Ib-cr), blaTEM, sul1, and tetX) obtained at the Gwangalli Beach in South Korea, between 2018 and 2019 were used. When individually predicting ARGs occurrence, the conventional LSTM and IA-LSTM exhibited poor R2 values during training and testing. In contrast, the LSTM-CNN exhibited a 2-6-times improvement in accuracy over those of the conventional LSTM and IA-LSTM. However, when predicting all ARGs occurrence simultaneously, the IA-LSTM model exhibited a superior performance overall compared to that of LSTM-CNN. Additionally, the influence of environmental variables on prediction was investigated using the IA-LSTM model, and the ranges of input variables that affect each ARG were identified. Consequently, this study demonstrated the possibility of predicting the occurrence and distribution of major ARGs at the beach based on various environmental variables, and the results are expected to contribute to management of ARG occurrence at a recreational beach.