Search

As the world grapples with the challenges of energy transition and industrial decarbonization, the development of carbon capture technologies presents a promising solution. The Scalable Modeling, Artificial Intelligence (AI), and Rapid Theoretical calculations, referred as SMART here, is an interdisciplinary approach that combines high-throughput calculation and data-driven modeling with expertise from chemical, materials, environmental, computer and data science and engineering, leading to the development of advanced capabilities in simulating and optimizing carbon capture processes. This perspective discusses the state-of-the-art material discovery research enabled by high-throughput calculation and data-driven modeling. Further, we propose a framework for material discovery, and illustrate the synergies among deep learning models, pretrained models, and comprehensive data sets, emerging as a robust framework for data-driven design and development in carbon capture. In essence, the adoption of the SMART approach promises a revolutionary impact on efforts in energy transition and industrial decarbonization.

Wearable strain sensors that detect joint/muscle strain changes become prevalent at human-machine interfaces for full-body motion monitoring. However, most wearable devices cannot offer customizable opportunities to match the sensor characteristics with specific deformation ranges of joints/muscles, resulting in suboptimal performance. Adequate wearable strain sensor design is highly required to achieve user-designated working windows without sacrificing high sensitivity, accompanied with real-time data processing. Herein, wearable Ti 3 C 2 T x MXene sensor modules are fabricated with in-sensor machine learning (ML) models, either functioning via wireless streaming or edge computing, for full-body motion classifications and avatar reconstruction. Through topographic design on piezoresistive nanolayers, the wearable strain sensor modules exhibited ultrahigh sensitivities within the working windows that meet all joint deformation ranges. By integrating the wearable sensors with a ML chip, an edge sensor module is fabricated, enabling in-sensor reconstruction of high-precision avatar animations that mimic continuous full-body motions with an average avatar determination error of 3.5 cm, without additional computing devices., (© 2022. The Author(s).)

Motivation: Protein activity is a significant characteristic for recombinant proteins which can be used as biocatalysts. High activity of proteins reduces the cost of biocatalysts. A model that can predict protein activity from amino acid sequence is highly desired, as it aids experimental improvement of proteins. However, only limited data for protein activity are currently available, which prevents the development of such models. Since protein activity and solubility are correlated for some proteins, the publicly available solubility dataset may be adopted to develop models that can predict protein solubility from sequence. The models could serve as a tool to indirectly predict protein activity from sequence. In literature, predicting protein solubility from sequence has been intensively explored, but the predicted solubility represented in binary values from all the developed models was not suitable for guiding experimental designs to improve protein solubility. Here we propose new machine learning (ML) models for improving protein solubility in vivo., Results: We first implemented a novel approach that predicted protein solubility in continuous numerical values instead of binary ones. After combining it with various ML algorithms, we achieved a R2 of 0.4115 when support vector machine algorithm was used. Continuous values of solubility are more meaningful in protein engineering, as they enable researchers to choose proteins with higher predicted solubility for experimental validation, while binary values fail to distinguish proteins with the same value-there are only two possible values so many proteins have the same one., Availability and Implementation: We present the ML workflow as a series of IPython notebooks hosted on GitHub (https://github.com/xiaomizhou616/protein_solubility). The workflow can be used as a template for analysis of other expression and solubility datasets., Supplementary Information: Supplementary data are available at Bioinformatics online., (© The Author(s) 2019. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.)

The adsorption of six heavy metals (lead, cadmium, nickel, arsenic, copper, and zinc) on 44 biochars were modeled using artificial neural network (ANN) and random forest (RF) based on 353 dataset of adsorption experiments from literatures. The regression models were trained and optimized to predict the adsorption capacity according to biochar characteristics, metal sources, environmental conditions (e.g. temperature and pH), and the initial concentration ratio of metals to biochars. The RF model showed better accuracy and predictive performance for adsorption efficiency (R 2  = 0.973) than ANN model (R 2  = 0.948). The biochar characteristics were most significant for adsorption efficiency, in which the contribution of cation exchange capacity (CEC) and pH H2O of biochars accounted for 66% in the biochar characteristics. However, surface area of the biochars provided only 2% of adsorption efficiency. Meanwhile, the models developed by RF had better generalization ability than ANN model. The accurate predicted ability of developed models could significantly reduce experiment workload such as predicting the removal efficiency of biochars for target metal according to biochar characteristics, so as to select more efficient biochar without increasing experimental times. The relative importance of variables could provide a right direction for better treatments of heavy metals in the real water and wastewater., (Copyright © 2019 Elsevier B.V. All rights reserved.)

Benefitting from the interlaced networking structure of carbon nanotubes (CNTs), the composites of CNTs/polydimethylsiloxane (PDMS) have found extensive applications in wearable electronics. While hierarchical multiscale simulation frameworks exist to optimize the structure parameters, their wide applications were hindered by the high computational cost. In this study, a machine learning model based on the artificial neural networks (ANN) embedded graph attention network, termed as AGAT, was proposed. The datasets collected from the micro-scale and the macro-scale simulations are utilized to train the model. The ANN layer within the model framework is trained to pass the information from micro-scale to macro-scale, while the whole model is aimed to predict the electro-mechanical behavior of the CNTs/PDMS composites. By comparing the AGAT model with the original multiscale simulation results, the data-driven strategy is shown to be promising with high accuracy, demonstrating the potential of the machine-learning-enabled approach for the structure optimization of CNT-based composites.

For practical applications, molecules often exist in an aggregate state. Therefore, it is of great value if one can predict the performance of molecules when forming aggregates, for example, aggregation-induced emission (AIE) or aggregation-caused quenching (ACQ). Herein, a database containing AIE/ACQ molecules reported in the literature is first established. Through training, these machine learning (ML) models can build up the structure-property relationship and thus implement fast prediction of AIE/ACQ properties. To this end, a multi-modal approach is proposed, multiple prediction methods are compared and designed, and thus an ensemble strategy is developed. First, multiple molecular descriptors are considered at the same time, major features are extracted by dimensionality reduction, and multi-modal features are synthesized. Then, several state-of-the-art methods are designed and compared to analyze the advantages of the different methods. Finally, the ensemble strategy combines the advantages of the multiple methods to obtain the final prediction result. The reliability of this approach in an unknown molecular space is further verified by three newly designed molecules. Reasonable consistency between model predictions and experimental outcomes is obtained. The result indicates that ML can be a powerful tool to predict molecular properties in the aggregated state, thus accelerating the development of solid-state optical materials.

To accurately estimate the age of individual tree and to achieve full-cycle sustainable management of natural Larix gmelinii forest in Great Xing' an Mountains of northeastern China, we constructed individual tree age prediction model using stepwise regression and random forest algorithms based on 44 fixed plots data and 280 standard tree cores obtained from the Pangu Forest Farm. We analyzed the influence of stand structure, site conditions, and competition index on the accuracy of model prediction. The model was evaluated by the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). The results showed that the random forest model had the highest prediction accuracy when number of decision trees was 1500 and number of node contention variables was 8. The random forest model had better accuracy and prediction ability than the stepwise regression model, with R², RMSE and MAE of 0.5882, 9.9259 a, 8.1155 a. Diameter at breast height was the most important factor affecting age prediction (83.8%), followed by tree height (34.4%), elevation (17.9%), and basal area per hectare (17.5%). The random forest algorithm exhibited better adaptability and modeling effect on constructing a predictive model for individual tree age. This research contributed to improving the accuracy of growth and harvest estimation for L. gmelinii, and could provide a reference for other scientific studies related to tree age estimation in forests. [ABSTRACT FROM AUTHOR]

Background: Suicide attempt is one of the most severe consequences for patients with mood disorders. This study aimed to perform deep learning and machine learning while using text mining to identify patients with suicide attempts and to compare their effectiveness.Methods: A total of 13,100 patients with mood disorders were selected. Two traditional text mining methods, logistic regression and Support vector machine (SVM), and one deep learning model (Convolutional neural network, CNN) were adopted to perform overall analysis and gender-specific subgroup analysis of patients to identify suicide attempts. The classification effectiveness of these models was evaluated by accuracy, F1-value, precision, recall, and the area under Receiver operator characteristic curve (ROC).Results: CNN's results were greater than the other two for all indicators except recall which was slightly smaller than SVM in male subgroup analysis. The accuracy values of the CNN were 98.4 %, 98.2 %, and 98.5 % in the overall analysis and the subgroup analysis for males and females, respectively. The results of McNemar's test showed that CNN and SVM models' predictions were statistically different from the logistic regression model's predictions in the overall analysis and the subgroup analysis for females (P < 0.050).Limitations: A fixed number of features were selected based on document frequency to train models; this was a single-site study.Conclusions: CNN model was a better way to detect suicide attempts in patients with mood disorders prior to hospital admission, saving time and resources in recognizing high-risk patients and preventing suicide. [ABSTRACT FROM AUTHOR]

UAV base station platform has become the current research hotspot of assisting ground base station for wireless coverage.At present, the most important issue is how to make path planning to provide the stable communication guarantee for multiple mobile users. In this article, we model the air-to-ground channel to describe the path loss between the UAV platform and the user and build a simulation environment for training based on the OpenAI-GYM architecture. In addition, this paper proposes a reinforcement learning algorithm based on intrinsic rewards, which uses the mean square error of the state prediction results to quantify the novelty of the state. Algorithms enable agents to efficiently carry out strategy iterations. Experiments results showed that our algorithm has a higher score and takes less time. [ABSTRACT FROM AUTHOR]

The real biological effect is not generated by the total content of heavy metals (HMs), but rather by bioavailable content. A new bioavailability-based ecological risk assessment (BA-based ERA) framework was developed for deriving bioavailability-based soil quality criteria (BA-based SQC) and accurately assessing the ecological risk of soil HMs at a multi-regional scale in this study. Through the random forest (RF) models and BA-based ERA framework, the 217 BA-based SQC for HMs in 31 Chinese provinces were derived and the BA-based ERA was comprehensively assessed. This study found that bioavailable HMs extraction methods (BHEMs) and total HMs content play the predominant role in affecting HMs (As, Cd, Cr, Cu, Ni, Pb, and Zn) bioavailability by explaining 27.55–56.11% and 9.20–62.09% of the variation, respectively. The RF model had accurate and stable prediction ability for the bioavailability of soil HMs with the mean R 2 and RMSE of 0.83 and 0.43 for the test set, respectively. The results of BA-based ERA showed that bioavailability could avoid the overestimation of ecological risks to some extent after reducing the uncertainty of soil differences. This study confirmed the feasibility of using bioavailability for ERA and will utilised to revise the soil environmental standards based on bioavailability for HMs. [Display omitted] • A new bioavailability-based ecological risk assessment framework was developed. • The feasibility of using bioavailability for ERA of soil HMs was confirmed. • The RF model had accurate prediction ability for the bioavailability of soil HMs. • The BA-based ERA for different geographic regions in China was assessed. [ABSTRACT FROM AUTHOR]

Soil and crop management are facing challenges, in particular an increasing pressure to feed a growing population and improve food safety. Smart farming assists in dealing with these issues by incorporating information and communication technologies into farming practice. The Multiponics Vertical Farming (MVF) system is a very promising concept in smart farming that saves space, transportation cost and delivery time. Complicated systems and messy data can be processed by advanced technologies to achieve greater accuracy and efficiency in modelling and computing. For dynamic and complex problems, Artificial Intelligence (AI) provides rapid evaluation and aids in designing effective solutions. This study discusses how AI is adopted in soil management and MVF for tasks including classification, detection and forecasting. Three categories of AI and machine learning techniques, Support Vector Machine (SVM), Decision Tree (DT) and Neural Network (NN)‐based models, that are widely used will be presented. Future perspectives on urban farming are also discussed. [ABSTRACT FROM AUTHOR]

With 'smart' being the order of the day, the shift in the landscape of a typical production‐oriented manufacturing environment to a more data‐oriented, automated and smart manufacturing is imminent. However, what is meant by smart manufacturing? And how can smart manufacturing contribute to a bigger picture by acting as enablers of smart cities? Given the paucity in literature that seeks to make sense in this direction, herein, first, six indices that represent or define a smart city are identified. Then, a holistic perspective of smart manufacturing is presented by collectively dwelling into the concepts of cyber physical production systems (CPPS) and industrial symbiosis—the recent and ongoing developments, applications, and relevant examples. In each subsequent section, the Review addresses how smart manufacturing contributes to smart cities, not just from a technology perspective, but also by satisfying the ergonometric factors and sustainability issues which are equally important indices that make up a smart city. A brief overview of Singapore as a smart nation and smart manufacturing hub is presented toward the end, along with highlights of a real‐world smart manufacturing platform called the Model Factory and its relevant modules. [ABSTRACT FROM AUTHOR]

[Display omitted] • Developed models showed excellent prediction accuracy across 10 key outputs. • Gradient boosting regression outperformed other model types. • Feedstock particle size and gasifying agent most strongly influenced predictions. • Global and local model interpretability methods have been combined. Machine learning has been regarded as a promising method to better model thermochemical processes such as gasification. However, their black box nature can limit how much one can trust and learn from the developed models. Here seven different machine learning methods have been adopted to model the gasification of biomass and waste across a wide range of operating conditions. Gradient boosting regression has been found to outperform the other model types with a coefficient of determination (R2) of 0.90 when averaged across ten key gasification outputs. Global and local model interpretability methods have been used to illuminate the developed black box models. The studied models were most strongly influenced by the feedstock's particle size and the type of gasifying agent employed. By combining global and local interpretability methods, the understanding of black box models has been improved. This allows policy makers and investors to make more educated decisions about gasification process design. [ABSTRACT FROM AUTHOR]

• A new online learning framework was developed for optimizing sugar production (SP) • Efficacy of SP from biomass was significantly improved by model-guided experiments • Impacts of experimental conditions and their interactions on SP were well explored • Generic modelling framework was provided for biomass valorization related researchers • Potential implication analysis has shown valuable insights for future industrialization Conversion of oil palm mesocarp fibers (MF) into fermentable sugar through catalytic pretreatment coupled with enzymatic hydrolysis is a promising solution for biomass valorization. However, lack of understanding of the complicated conversion process and optimization of critical reaction conditions limits the efficacy of the sugar production (SP). In this study, we aim to develop an online framework incorporating orthogonal experimental design and machine learning (ML) algorithm to optimize important conditions and identify their interaction impacts on overall sugar yield from MF. The results show that both solid reduction (SR) and sugar conversion (SC) increase to 73.85% and 51.91% from ML-based optimization with relative errors of 20.58%±0.70% and 7.50%±3.68% from experimental validation, respectively, after two rounds of online learning. Model-based interpretation suggests that substrate and catalyst concentrations are two negatively related conditions with both SR and SC, while the solid concentration has a piecewise linear folded relationship. Moreover, potential implications on environments and economy of the presented efficiency (51.91%) of overall sugar yield were evaluated based on the amount of MF generated in Malaysia. The ML-optimized conversion strategy can save about 70 million tons of sugar cane and achieve a profit of US$ 1,450 million annually with a low operation cost (40 US$/ton MF conversion). It provides insights that help us to identify important conditions for improving conversion efficiency and promoting future industrialization. [Display omitted] [ABSTRACT FROM AUTHOR]

• Optimizing integrated solar-bioenergy system for emission and cost reductions. • LSTM-RNN model for system dynamics prediction. • Monte Carlo simulation to generate datasets for machine learning model development. • Combined LSTM-RNN and Monte Carlo simulation for robust system evaluation. • Carbon tax impacts optimal system configuration and trade-off avoidance consideration. The combination of solar and biomass energy systems is regarded as a highly promising technology for tackling the challenges related to greenhouse gas emissions from energy generation and the increasing costs of energy production. This research centers on an integrated solar-bioenergy system, which includes a concentrated solar tower, thermal energy storage, and a combined cycle gas turbine. The system was evaluated using a multi-objective optimization approach considering life cycle assessment and cost-benefit analysis. The long short-term memory recurrent neural network algorithm with 5.1 % average error had been employed to capture the intricate temporal dependencies and dynamics of the system. The scenarios are expanded by using the Monte Carlo approach to address the challenges of limited specialized models and experiments for the system. The optimal solution is determined through the technique for order preference by similarity to ideal solution method. Carbon tax significantly influenced the results of the multi-objective optimization. The optimal configuration of the system could avoid the trade-off phenomenon when treating the carbon tax as revenue. The best scenario of the system with the cumulative reduction in global warming potential amounted to 415,960 tons of CO 2 -eq and a 30-year net present worth of €4,298 million. Without considering the carbon tax as revenue, the trade-off is present. The best scenario of the system with the cumulative reduction in global warming potential amounted to 132,615 tons of CO 2 -eq and net present worth of €3,042 million. The findings highlight the robust prospects of the system across environmental and economic dimensions. [ABSTRACT FROM AUTHOR]

In this work, we present a physics-informed recurrent neural network (PIRNN) modeling approach, and a PIRNN-based predictive control scheme for a general class of nonlinear dynamic systems. Specifically, we first develop a hybrid data-driven and physics-guided modeling framework that integrates measurement data and mechanistic mathematical models to construct high-fidelity RNN models. Then, we derive a generalization error bound of the PIRNN model based on a nominal system model via the Rademacher complexity technique from statistical machine learning theory. Subsequently, the PIRNN model is utilized in Lyapunov-based model predictive controllers and applied to a chemical reactor example with Gaussian measurement noise to demonstrate its improved noise rejection and generalization performance in comparison to the purely data-driven and the purely physics-guided RNN-based predictive control schemes. • Physics-informed RNN modeling framework for nonlinear dynamic systems. • Theoretical generalization error bound for physics-informed RNN. • Model predictive control scheme using physics-informed RNN models. • Application to a chemical reactor example to demonstrate its effectiveness. [ABSTRACT FROM AUTHOR]

[Display omitted] • This study compared expert-based supervised pre-training models with self-supervised pre-training models for various datasets. • The research examines model efficacy in predicting adsorption energy during CO 2 reduction via M−X−C single-atom catalysts. • The optimal pre-training approach pinpointed two superior catalysts: Zn-NP 3 and Sn-P 4. • The findings underscore the transformative potential of self-supervised pre-training in material discovery. Machine learning, by virtue of its powerful predictive capabilities in discerning material properties, has emerged as an indispensable tool in the pursuit of new material discovery and design. Its specific impact on advanced catalyst design is unlocking breakthroughs across various scientific disciplines. However, the success of machine learning (ML) in predicting material properties is strongly influenced by the size of the dataset. Small datasets can lead to overfitting, resulting in suboptimal predictive performance. In this study, we compare various pre-training models that leverage expert knowledge with those based on large, mixed datasets for specific downstream property prediction tasks. We argue that transfer learning, particularly self-supervised learning based on large datasets, offers more generalizable, scalable, and accurate predictions than supervised pre-training models. Our analysis demonstrates the advantages of self-supervised pre-training, particularly in cases where expert knowledge is limited or insufficient. We evaluate the performance of these models on the challenging task of predicting the adsorption energy associated with CO 2 reduction by M−X−C single-atom catalysts. Using the optimal model, the self-supervised pre-training model, we conducted a screening of single-atom catalysts for CO 2 reduction to CO, leading to the discovery of two new, more efficient catalysts, namely Zn-NP 3 and Sn-P 4. Our findings highlight the potential of self-supervised pre-training models in advancing material property prediction and guiding the discovery of novel materials. [ABSTRACT FROM AUTHOR]

Energy storage systems (ESS) are becoming more prevalent and indispensable in modern electrical infrastructure. The process of choosing the proper type of ESS technology for the application is the first step in designing the best performing ESS. However, the selection process involves a variety of factors, and currently there lacks a sophisticated and systematic framework for convenient energy storage selection. This paper develops a data-driven optimization framework for selecting energy storage systems for general energy system applications. In the framework, a supervised classification machine learning method is proposed for the first time to quantify the technical suitability of energy storage technologies for different applications. The DOE Global Energy Storage Database provided the basic information for machine learning, and the Random Forest Classifier had the best prediction performance for this dataset. The probability of technical suitability can be predicted and further incorporated in a multi-objective optimization for technology recommendation based on integrated technical, economic, and environmental criteria. As a demonstration of the methodology, the prediction of technical suitability and recommendation of technology selection were conducted for eleven common energy storage applications. The result showed that the prediction of technical suitability and the multi-objective optimization yielded reasonable recommendations compared with the literature. In conclusion, the proposed data-driven decision-making approach can be used as a convenient tool to help decision-makers make informed decisions when choosing the energy storage system for general applications. [Display omitted] • A data-driven decision-support framework for energy storage selection is developed. • Data exploration and visualization of the DOE energy storage project database are presented. • Novel quantification of energy storage technical suitability using machine learning is proposed. • Random forest classifier has the best prediction performance among the tested models. [ABSTRACT FROM AUTHOR]

[Display omitted] • Building a multi-layer automated machine learning framework for AD systems. • GBM model was more suitable to learn AD data and to simulate and predict AD. • HRT was key factor with an optimal range of 33–45 d. • Methanocelleus was the key microorganisms for AD. • Biogas production is best predicted by archaea + environmental factors. Exploring key factors has important guidance for understanding complex anaerobic digestion (AD) systems. This study proposed a multi-layer automated machine learning framework to understand the complex interactions in AD systems and explore key factors at the environmental factor, microorganisms and system levels. The first layer of the framework identified hydraulic residence time (HRT) as the most important environmental factor, with an optimal range of 33–45 d. In the second layer of the framework, Methanocelleus (optimal relative abundance (ORA) = 3.0%) and Candidatus_Caldatribacterium (ORA = 1.7%) were found to be the key archaea and bacteria, respectively. Furthermore, the prediction of key microorganisms based on environmental factors and remaining microbial data showed the essential roles of Methanothermobacter and Acetomicrobium. The third layer for finding the optimal combination of data variables for predicting biogas production demonstrated that combined Archaea genera and environmental factors should be achieved for the most accurate prediction (root mean square error (RMSE) = 84.21). GBM had the best model performance and prediction accuracy among all the built-in models. Based on the optimal GBM model, the analysis at the system level showed that HRT was the most important variable. However the most important microorganism, Methanocelleus , within the appropriate survival range is also essential to achieve optimal biogas production. This research explores key parameters at various levels through automated machine learning techniques, which are expected to provide guidance in understanding the complex architecture of industrial and laboratory AD systems. [ABSTRACT FROM AUTHOR]

• How machine learning models can aid pathway engineering is described. • The workflow of developing and validating machine learning models is illustrated. • Key innovations needed to generate large dataset are discussed. Machine learning is transforming many industries through self-improving models that are fueled by big data and high computing power. The field of metabolic engineering, which uses cellular biochemical network to manufacture useful small molecules, has also witnessed the first wave of machine learning applications in the past five years, covering reaction route design, enzyme selection, pathway engineering and process optimization. This review focuses on pathway engineering, and uses a few recent studies to illustrate (1) how machine learning models can be useful in overcoming an evident rate-limiting step, and (2) how the models may be used to exhaustively search – or guide optimization algorithms to search – a large design space when the cellular regulation of the reaction network is more convoluted. [ABSTRACT FROM AUTHOR]

Impurity detection involves detecting small impurities in the liquid inside an opaque glass bottle with complex textures by looking through the bottleneck. Sometimes experts have to observe continuous frames to determine the existence of an impurity. In recent years, region-based convolutional neural networks have gained incremental successes in common object detection tasks. However, sequential impurity detections present more challenging issues than detecting targets in a single frame, because consecutive motions and appearance changes of impurities cannot be captured using those common object detectors. In this paper, we propose a simple and controllable ensemble architecture to alleviate this problem. Specifically, a siamese fusion network is used to generate impurity proposals, then an attention model based on visual features and trajectories is proposed to localize a unique region proposal in each frame, finally, a sequential region proposal classifier using a long-term recurrent convolutional network is applied to refine impurity detection performances. The proposed method achieves 79.81 % mAP on IML-DET datasets, outperforming a comparable state-of-the-art Mask R-CNN model. [ABSTRACT FROM AUTHOR]

The optimization of water permeability and salt rejection of thin film nanocomposite (TFN) membranes is of great interests for reverse osmosis (RO) desalination. Based on literature data, machine learning was used to form prediction models of water permeability and salt pass rate for TFN RO membranes. A literature review was done to examine key parameters in membrane transport. Gradient boosting tree model was employed to learn from relevant variables such as loading, size, pore size of nanoparticles, and properties of the membranes. The results suggest that while porous nanoparticles perform better than nonporous ones, factors including loading, size and hydrophilicity are the primary factors that influence membrane performances. Ways to optimize the parameters for improved membrane performance were discussed using partial dependence plot analysis. The optimized properties were also compared with aquaporin-based membranes and implications for future development were discussed. Image 1 • Machine learning was used to understand TFN membranes. • Gradient boosted tree model was evaluated and chosen. • Porosity, size, hydrophilicity and loading were the key factors. • Optimized nanoparticles: pore size 5–7 Å, size 150 nm, loading ~0.1 wt%. • Comparisons with aquaporin-based membranes and implications were discussed. [ABSTRACT FROM AUTHOR]

• The pyrolysis process of lignocellulosic biomass was modeled by machine learning. • Random forest showed good prediction ability for biochar yield and carbon contents. • Comprehensive and appropriate inputs were critical to predict the target outputs. • Pyrolysis temperature was most significant for the change of both yield and carbon contents. • Analysis of Partial Dependence Plot provided inside information for pyrolysis process. In the study, machine learning was used to develop prediction models for yield and carbon contents of biochar (C-char) based on the pyrolysis data of lignocellulosic biomass, and explore inside information underlying the models. The results suggested that random forest could accurately predict biochar yield and C-char according to biomass characteristics and pyrolysis conditions. Furthermore, the relative contribution of pyrolysis conditions was higher than that of biomass characteristics for both yield (65%) and C-char (53%). For biomass characteristics, structural information was more important than elements compositions for accurately predicting biochar yield and it was inverse for C-char. The partial dependence plot analysis showed the impact way of each influential factor on the target variable and the interactions among these factors in the pyrolysis process. The present work provided new insights for understanding pyrolysis process of biomass and improving biochar yield and C-char. [ABSTRACT FROM AUTHOR]

• Conversion of plastic waste into biofuel by pyrolysis was compressively reviewed. • Effects of conditions, catalyst, plastic, and reactor type on biofuel were discussed. • The state-of-the-art of machine learning in plastic pyrolysis was reviewed. • A critical discussion of recent challenges with future perspectives was presented. It is urgent to promote the technology development of plastic waste (PW) valorization to mitigate the present serious plastic pollution. Pyrolysis has been widely used to convert PW into high-value-added products, especially liquid fuels, in a sustainable manner. However, achieving a high yield of liquid fuel with good quality remains challenging due to the difficulty in optimizing pyrolysis process and synthesizing outstanding catalysts to narrow the product distribution. This work comprehensively reviewed PW pyrolysis from both technical and computational (machine learning modeling) aspects, with a critical discussion of recent challenges to find new insights for improving the conversion efficiency and promoting commercialization. Results indicated that the impacts of various factors, including PW type, process condition, catalyst, and reactor type, on the PW pyrolysis were extensively investigated by the research community. Machine learning methods have also been frequently applied to predict, interpret, and optimize the PW pyrolysis. However, more efforts can be made in the future regarding catalyst synthesis, selection of co-pyrolysis additives, mechanism of catalyst deactivation, and design of renewable energy supply system for PW pyrolysis plants. Additionally, more attention should be paid to enlarging the data size, improving model interpretability, and exploring innovative ways of machine learning application (e.g., active learning) in process optimization and catalyst design for PW pyrolysis. [ABSTRACT FROM AUTHOR]

• Novel deep learning model specialized for dynamic flow reactor modelling. • Model architecture developed by integrating evolving deep learning methods and conventional reaction engineering theories. • High data efficiency allows to train with minimal number of process data. • High model interpretability allows to extract learned reaction kinetics and reactor residence time distribution. Modeling is essential for designing, scaling up, controlling, and optimizing a reactor or process involving reactions. However, developing high-fidelity mechanistic models from first principles for reactor systems involving complex physiochemical phenomena is usually time- and resource-consuming. Therefore, machine learning models using data-driven methods can help in such cases to fill the gap between the complex system and our limited knowledge. Currently, most research works use generic off-the-shelf machine learning models to model reactor behavior. Such models frequently face problems related to data limitations, dynamics, model accuracy, and model interpretability. Considering the increasing need for data-driven models, especially in the fine chemicals and pharmaceutical industry, this work presents a new machine learning model architecture specially for the dynamic modeling of general flow reactors. Derived from the conventional residence time distribution reactor model, our generalized reactor neural ODE (GRxnODE) can achieve, without any prior knowledge of reaction kinetics, higher model accuracy, data efficiency, and model interpretability than commonly used data-driven models. The well-trained model can predict dynamic reactor response and learn reaction kinetics and reactor RTD from process data. The source codes of the model are publicly available. [ABSTRACT FROM AUTHOR]

As an important subtopic within phytoremediation, hyperaccumulators have garnered significant attention due to their ability of super-enriching heavy metals. Identifying the factors that affecting phytoextraction efficiency has important application value in guiding the efficient remediation of heavy metal contaminated soil. However, it is challenging to identify the critical factors that affect the phytoextraction of heavy metals in soil–hyperaccumulator ecosystems because the current projections on phytoremediation extrapolations are rudimentary at best using simple linear models. Here, machine learning (ML) approaches were used to predict the important factors that affecting phytoextraction efficiency of hyperaccumulators. ML analysis was based on 173 data points with consideration of soil properties, experimental conditions, plant families, low-molecular-weight organic acids from plants, plant genes, and heavy metal properties. Heavy metal properties, especially the metal ion radius, were the most important factors that affect heavy metal accumulation in shoots, and the plant family was the most important factor that affect the bioconcentration factor, metal extraction ratio, and remediation time. Furthermore, the Crassulaceae family had the highest potential as hyperaccumulators for phytoremediation, which was related to the expression of genes encoding heavy metal transporting ATPase (HMA), Metallothioneins (MTL), and natural resistance associated macrophage protein (NRAMP), and also the secretion of malate and threonine. New insights into the effects of plant characteristics, experimental conditions, soil characteristics, and heavy metal properties on phytoextraction efficiency from ML model interpretation could guide the efficient phytoremediation by identifying the best hyperaccumulators and resolving its efficient remediation mechanisms. [Display omitted] • Machine learning was used to predict factors that affect phytoremediation. • Metal ion radius was the most important factor affects HM accumulation in shoots. • Plant family was the most important factor that affect BCF and MER. • Crassulaceae family had the highest potential as hyperaccumulators. [ABSTRACT FROM AUTHOR]

[Display omitted] • Machine learning was used to predict the maturity of compost from food waste. • Germination index and C/N were predicted with R2 of >0.93. • Different model architectures provided comprehensive feature importance analysis. • Early warning and process regulation were provided based on prediction models. • A prediction app was developed to illustrate automatic reactor composting. Reactive composting is a promising technology for recovering valuable resources from food waste, while its manual regulation is laborious and time-consuming. In this study, machine learning (ML) technologies are adopted to enable automated composting by predicting compost maturity and providing process regulation. Four machine learning algorithms, namely random forest (RF), extreme gradient boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and Multilayer Perceptron (MLP) are employed to predict the seed germination index (GI) and C/N ratio. Based on the best fusion model with the highest R2 of 0.977 and 0.986 for the multi-task prediction of GI and C/N ratio, the critical factors and their interactions with maturity are identified. Moreover, the ML model is validated on a composting reactor and the ML-based prediction application can provide regulation to ensure food waste decompose within the required time. In conclusion, this compost maturity prediction system automates the reactive composting, thus reducing labor costs. [ABSTRACT FROM AUTHOR]

The conversion of wet waste (e.g., food waste, sewage sludge, and animal manure) into bioenergy is a promising strategy for sustainable energy generation and waste management. Although experimental efforts have driven waste conversion technologies (WCTs) to various degrees of maturity, computational modeling has equally contributed to this endeavor. This review focuses on the application of modeling techniques, including computational fluid dynamics (CFD), process simulation (PS), and machine learning (ML) on WCTs including anaerobic digestion, hydrothermal carbonization, gasification, pyrolysis and incineration. It addresses in a concise manner on how CFD models aid in understanding of the complex process and their molecular kinetics; while PS and ML models help in understanding the reaction kinetics, variable-response relationship, techno-economic assessment and sensitivity analysis. Relevant modeling approaches with their pros and cons are summarized and case studies are presented for each WCT. Moreover, a comparative evaluation among the three modeling techniques, along with their recent and ongoing developments are highlighted. Hybrid frameworks derived by combining mechanistic and ML models are proposed, which are expected to advance future wet waste valorization strategies for sustainable clean energy production and waste management. [ABSTRACT FROM AUTHOR]

Carbon capture technologies have been extensively investigated as indispensable tools for reducing CO 2 emissions. In particular, CO 2 capture using solid waste-derived porous carbons (SWDPCs) has attracted significant research attention as one of the most promising and sustainable approaches to simultaneously mitigate climate change and address solid waste management challenges. Considerable research has recently been conducted on the thermal and chemical treatments of solid waste for upcycling into porous carbons (PCs) for effective and selective CO 2 capture. In this review, we discuss the synergistic benefits of employing SWDPCs for CO 2 capture and introduce innovative approaches for converting solid waste into PCs with the desired physical and chemical properties. The performance of SWDPCs for CO 2 capture is comprehensively discussed in terms of the synthesis route, CO 2 capture capacity, process cyclability, and sample optimization guided by machine learning. Furthermore, the mechanisms of CO 2 capture on PCs are discussed based on pore structures and incorporated surface functional groups. The life-cycle environmental impact of the PCs synthesized from solid waste and their practical applications for CO 2 capture are also evaluated. The overall environmental benefits of the proposed SWDPC-based CO 2 capture approach are analyzed in relation to the United Nations Sustainable Development Goals. Furthermore, the remaining challenges in upcycling solid waste into high-performance CO 2 adsorbents are discussed, and potential solutions are proposed. • GHG emission reductions and better solid waste management are required to achieve the UN SDGs. • Synthesis of solid-waste-based CO 2 adsorbents may be optimized using machine learning. • The life-cycle environmental impact was assessed for sustainable development. • A circular carbon economy is required to address climate change and waste pollution. [ABSTRACT FROM AUTHOR]

[Display omitted] • Wastes to hydrogen was studied for both energy and environmental benefits. • Application of data-driven approaches for aiding SCWG process was investigated. • Machine learning based predictive and optimization models were developed. • Catalyst descriptors and critical process factors based on feature engineering were revealed. • Optimal experimental conditions and catalysts for SCWG of wet wastes were identified. Hydrogen production from wet organic wastes through supercritical water gasification (SCWG) promotes sustainable development. However, it is always time-consuming and expensive to achieve optimal SCWG conditions and suitable catalysts for different wastes to produce H 2 -rich syngas. Herein, we developed a unified machine learning (ML) framework to predict syngas composition from SCWG processes inclusive of non-catalytic and catalytic systems. The neural network (NN) model which was the core of the framework exhibited generalizable and satisfactory accuracy (R 2 > 0.85) for all systems evaluated. The model also aided in the screening of catalyst and provided optimal conditions for accelerating experiments to produce H 2 -rich syngas by maximizing H 2 yield and minimizing CO 2 yield. ML model-based exploration found that SCWG temperature and solid content in the feedstock were the two most important factors affecting syngas composition. ML-based optimization suggested that Fe-based catalyst exhibited a greater potential to promote SCWG of wet wastes under optimal operational conditions. Besides, a web-based graphic user interface was developed by embedding the developed NN model for free access to the SCWG scientific community. [ABSTRACT FROM AUTHOR]

[Display omitted] • A hybrid data-driven and mechanistic modelling framework was developed for HTG. • The developed GBR model could predict the syngas composition with R2 > 0.9. • GBR-optimization for producing H 2 -rich syngas was validated by mechanistic simulation. • Data-driven and mechanistic modelling can complement each other to aid waste to energy. Recent advances in machine learning (ML) have witnessed a profound interest and application in the domain of waste to energy. However, their black-box nature renders challenges for ubiquitous acceptance. To address this issue, we developed a novel and first-of-its-kind hybrid data-driven and mechanistic modelling approach for hydrothermal gasification (HTG) of wet waste, in which a gradient boost regressor (GBR) integrated optimization model was first developed to predict and optimize the yield of syngas from HTG of wet waste, and then the predictions of the GBR model were validated and interpreted via mechanistic simulations in Aspen Plus. Results showed that the GBR model had a prediction performance with test R 2 > 0.90. GBR-based feature analysis identified that reaction temperature and feedstock solid content were the two significant features necessary to achieve high H 2 yield in syngas. Moreover, the GBR-based optimization provided optimal process conditions for H 2 -rich syngas production, which was validated over mechanistic simulations in Aspen Plus with an error of less than 20%. Interpretation from the mechanistic simulation revealed that steam and dry methane reforming and CO 2 methanation were the most significant reactions in the overall HTG process, responsible to produce H 2 -rich syngas. Integrated modelling approach as presented in this study shows how data-driven and mechanistic models complement each other and can aid the acceleration of experimental design of HTG by ML in general. [ABSTRACT FROM AUTHOR]

[Display omitted] • Dataset about HTL of biomass for bio-oil production was systematically complied. • Multi-task ML models were successfully developed for bio-oil prediction. • ML-based reverse optimization was achieved to guide high-quality bio-oil production. • Optimal solutions from reverse optimization were experimentally verified. Hydrothermal liquefaction (HTL) of biomass with high moisture (e.g., algae, sludge, manure, and food waste) is a promising and sustainable approach to produce renewable energy (bio-oil) and protect the environment. However, the production of bio-oil with high yield and preferable properties such as low nitrogen content (N_oil) is time/labor-consuming using the traditional HTL experimental method. To this context, machine learning (ML) algorithms were employed to aid the bio-oil production with the consideration of related factors in HTL, including biochemical and elemental compositions of biomass, process parameters, and solvents. Results showed that the random forest (RF) algorithm was the best one (average R 2 = 0.80) for the multi-task prediction of bio-oil yield (Yield_oil), N_oil, and energy recovery (ER_oil), hence employed for post feature interpretation and optimization. Feature importance indicates that both Yield_oil and ER_oil follow the trend of lipid content in biomass > temperature > retention time, while N_oil follows the trend of N content in biomass > temperature > retention time. Then ML-based optimization was conducted to guide the experimental research to produce bio-oil with high yield and low N content. The HTL experiment verification based on the optimal solutions from the ML-based Particle Swarm Optimization achieved the maximum Yield_oil (54.30%) and minimal N_oil (2.60%) from model biomass (composed of protein content 28%, lipid 48%, and carbohydrates 21%) at 300 °C and 30 min. The experiment verification was successful as the results were comparable to the modeling results, with errors of less than 7% for Yield_oil and ER_oil, and 23% for N_oil, and the N_oil of the experiment N_oil (2.60%) was interestingly lower than the modeled one (3.37%). This work provides new insight and strategy to accelerate the engineered HTL for desired bio-oil production. [ABSTRACT FROM AUTHOR]

• SMOTE-REG is proposed for generating samples for small dataset to address regression tasks. • SMOTE-REG discovers new samples that match the original data distribution. • Deep neural network is applied for regression based on synthetic samples. • Experimental results demonstrate significant performance on prediction and inference. • The method enables deep learning for solving problems with a small dataset. Precise control over dimension of nanocrystals is critical to tune the properties for various applications. However, the traditional control through experimental optimization is slow, tedious and time consuming. Herein a robust deep neural network-based regression algorithm has been developed for precise prediction of length, width, and aspect ratios of semiconductor nanorods (NRs). Given there is limited experimental data available (28 samples), a Synthetic Minority Oversampling Technique for regression (SMOTE-REG) is employed first for data generation. Deep neural network is further applied to develop regression model which demonstrated the well performed prediction on both the original and generated data with a similar distribution. The prediction model is further validated with additional experimental data, showing accurate prediction results. Additionally, Local Interpretable Model-Agnostic Explanations (LIME) is used to interpret the weight for each sample, corresponding to its importance towards the target dimension, which is well validated by experimental observations. [ABSTRACT FROM AUTHOR]

[Display omitted] • A three-step framework MIGRATE for smart energy management in manufacturing is proposed. • Industrial big data and machine learning foster the development of MIGRATE. • Supervised learning predicts machine-specific load profiles via energy disaggregation. • Unsupervised learning estimates the machine's activity state from disaggregated load. • Framework developed is purely data-driven and cross-deployable. The dynamic nature of chemical processes and manufacturing environments, along with numerous machines, their unique activity states, and mutual interactions, render challenges to energy monitoring at a machine level. In this study, we introduce MIGRATE (Machine learnInG foR smArT Energy), a novel three-step framework to predict the machine-specific load profiles via energy disaggregation, which are in turn used to predict the machine's activity state and the respective production capacities. Various supervised tree-based and recurrent neural network algorithms were evaluated on their capacities to predict load profiles and production capacities of four machines investigated in this study. Light gradient boosting machines and ensemble bi-directional long-term short memory were identified as the respective best performing algorithms with a mean absolute error and root mean squared error of 0.035 and 0.105 (units in Watts) for the disaggregation studies and 1.639 and 11.401 (units in quantities of samples processed) for production estimation. Four unsupervised machine learning algorithms were evaluated to cluster the machine's activity state from their disaggregated load profiles, where the gaussian mixture model had a superior performance with the V score and Fowlkes Mallows index of 0.852 and 0.983, respectively. The MIGRATE framework is purely data-driven, cross-deployable and serves as promising catalyst to foster smart energy management practices and sustainable productions in the chemical and industrial manufacturing processes. [ABSTRACT FROM AUTHOR]

In industrial steel plate production, process parameters and steel grade composition significantly influence the microstructure and mechanical properties of the steel produced. But determining the exact relationship between process parameters and mechanical properties is a challenging process. This work aimed to devise a deep learning model, to predict mechanical properties of industrial steel plate including yield strength (YS), ultimate tensile strength (UTS), elongation (EL), and impact energy (A kv); based on the process parameters as well as composition of raw steel, and apply it online to a real steel manufacturing plant. An optimal deep neural network (DNN) model was formulated with 27 inputs parameters, 2 hidden layers each having 200 nodes and 4 output parameters (27 × 200 × 200 × 4) with an initial learning rate 0.0001, using Adam optimizer and subjected to Z pre-processing method, to yield an accurate model with R2 = 0.907. The tuned DNN model, had a root mean square error of 21.06 MPa, 16.67 MPa, 2.36%, and 39.33 J, and root mean square percentage error of 4.7%, 2.9%, 7.7%, and 16.2%, for YS, UTS, EL and A kv respectively. Through comparative analysis, it was found that the accuracy of DNN model was higher than other classic machine learning algorithms. To interpret the model assumptions and findings, several local linear models were devised and analyzed to establish the link between process parameters and mechanical properties. Finally the tuned DNN model was deployed in the real-steel plant for online monitoring and control of steel mechanical properties, and to guide the production of targeted steel plates with tailored mechanical properties. Unlabelled Image • An online prediction system of the mechanics of the hot-rolled steel plate is developed. • Real-world data (11,101 groups) were used to avoid overfitting and allow broad applicability. • Deep neural network is trained to achieve high prediction accuracy. • Impacts of model hyperparameters are discussed in detail. • Physical metallurgy knowledge learned by deep learning is interpreted. [ABSTRACT FROM AUTHOR]

Hydrothermal carbonization (HTC) is a promising technology for valuable resources recovery from high-moisture wastes without pre-drying, while optimization of operational conditions for desired products preparation through experiments is always energy and time consuming. To accelerate the experiments in an efficient, sustainable, and economic way, machine learning (ML) tools were employed for bridging the inputs and outputs, which can realize the prediction of hydrochar properties, and development of ML-based optimization for achieving desired hydrochar. The results showed that deep neural network (DNN) model was the best one for joint prediction of both fuel properties (FP) and carbon capture and storage (CCS) stability of hydrochar with an average R 2 and root mean squared error (RMSE) of 0.91 and 3.29. The average testing prediction errors for all the targets were below 20%, furtherly within 10% for HHV, carbon content and H/C predictions. ML-based feature analysis unveiled that both elementary composition and temperature were crucial to FP and CCS. Furthermore, a ML-based software interface was provided for practitioners and researchers to freely access. The insights and Pareto solution provided from ML-based multi-objective optimization benefitted desired hydrochar preparation for the potential application of fuel substitution or carbon sequestration in soil. Image 1 • ML models and software were developed to predict characteristics of hydrochar. • The optimal DNN exhibited the highest accuracy for the multi-task prediction. • Integration of ML and MOO models was conducted to maximize hydrochar performance. • ML-based feature importance and MOO are beneficial to preparing desired hydrochar. [ABSTRACT FROM AUTHOR]

• Datasets for hydrochar and pyrochar were compiled by systematic literature review. • Machine learning models with multi-task prediction were developed and optimized. • Fuel properties of chars were predicted by optimal models with R2 of ~0.90. • Feature impacts on targets were explored with model explainer based on game theory. • Performance of models was further improved after feature re-examination. Conversion of wet organic wastes into renewable energy is a promising way to substitute fossil fuels and avoid environmental deterioration. Hydrothermal carbonization and pyrolysis can convert wet organic wastes into hydrochar and pyrochar, which are potential fossil fuel alternatives due to their comparable fuel properties. Machine learning (ML) has strong prediction ability after being trained with historic dataset and facilitates good understanding of the impact of input features on output targets through a data-driven approach. In this study, ML models for multi-task prediction of fuel properties of the chars were developed and optimized based on two datasets for hydrochar and pyrochar. Feature importance and correlation were explored based on optimized ML model, and feature re-examination was conducted for model improvement. Results showed that support vector regression model with optimal hyper-parameters exhibited better generalized performance for prediction of both hydrochar and pyrochar properties with the best average R2 of 0.90 and 0.94. ML-based feature analysis indicated that process temperature and carbon content in the feedstock were the significant features impacting fuel properties of both chars, while nitrogen content was another important input feature for hydrochar and hydrogen content for pyrochar. The accuracy (especially for pyrochar), generalization ability, and computational speed of models were further improved after feature re-examination. The intuitions obtained from feature analysis provided meaningful insights to select input features for prediction performance improvement and computational cost saving, and might guide experiments to produce chars with desired quality. [ABSTRACT FROM AUTHOR]