Next Article in Journal
Polyol Formation of Silver@Metal Oxides Nanohybrid for Photocatalytic and Antibacterial Performance
Previous Article in Journal
Synthesis and Antimicrobial Activity of Chalcone-Derived 1,4-Dihydropyridine Derivatives Using Magnetic Fe2O3@SiO2 as Highly Efficient Nanocatalyst
Previous Article in Special Issue
Enhanced Diclofenac Removal from Constructed Wetland Effluent Using a Photoelectrocatalytic System with N-TiO2 Nanocrystal-Modified TiO2 Nanotube Anode and Graphene Oxide/Activated Carbon Photocathode
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Catalysts
Volume 15
Issue 3
10.3390/catal15030282
catalysts-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Machine Learning-Assisted Catalysts for Advanced Oxidation Processes: Progress, Challenges, and Prospects

by
Qinghui Yuan
1,
Xiaobei Wang
2,
Dongdong Xu
2,
Hongyan Liu
1,3,
Hanwen Zhang
1,3,
Qian Yu
1,
Yanliang Bi
1,* and
Lixin Li
4,*
1
Department of Ocean, Binzhou Polytechnic, No. 919 Huanghe 12th Road, Binzhou 256600, China
2
College of Information Engineering, Binzhou Polytechnic, No. 919 Huanghe 12th Road, Binzhou 256600, China
3
Department of Marine Convergence Design Engineering, Pukyong National University, 45, Busan 48513, Republic of Korea
4
School of Environment and Chemical Engineering, Heilongjiang University of Science and Technology, Harbin 150022, China
*
Authors to whom correspondence should be addressed.
Catalysts 2025, 15(3), 282; https://doi.org/10.3390/catal15030282
Submission received: 30 January 2025 / Revised: 9 March 2025 / Accepted: 15 March 2025 / Published: 17 March 2025
(This article belongs to the Special Issue Nanomaterials in Environmental Catalysis)
Browse Figures
Versions Notes

Abstract

:
Advanced oxidation processes (AOPs) are recognized as one of the most effective methods in the field of wastewater treatment, and the selection of catalysts in the oxidation process is very important. In the face of the traditional test trial-and-error method, the method of screening advanced oxidation catalysts is time-consuming and inefficient. This paper examines approximately two decades’ worth of literature pertaining to the development of catalysts facilitated by machine learning. A synopsis of the various advanced oxidation processes and reactive oxygen species (ROS) is provided. Subsequently, it is posited that the swift advancement of machine learning (ML) and its algorithmic classification has significantly propelled the progress in ML-assisted catalyst screening, active site prediction, the discovery of acceleration mechanisms, and catalyst structural research, which are subsequently elucidated. Despite ML’s proven efficacy as a tool within the domain of AOPs’ catalysis, the article concludes by presenting challenges and outlining future development strategies, particularly in light of issues pertaining to data quality and quantity, as well as inherent model limitations.

1. Introduction

Over the past few years, numerous chemical companies have tended to release untreated or partly treated wastewater to maximize profits, resulting in a rise in the volume of wastewater [1]. The existence of multiple contaminants within water bodies contributes to the global pollution and poses significant risks to humanity, wildlife, and other living organisms [2]. Researchers from all walks of life contribute their wisdom to wastewater treatment and develop various methods for wastewater treatment [3]. The most commonly used wastewater treatment technologies can be categorized into physical, chemical, and biological treatment methods [4]. The physical method of wastewater treatment mainly includes precipitation, adsorption, and membrane separation [5,6,7]. It is difficult to remove organic materials from water using conventional physical treatment methods [8,9]. The biological treatment of wastewater can improve efficiency. The biological method mainly includes activated sludge, biofilm, and immobilized microbial water treatment technology. However, high investment and operating costs, susceptibility to sludge swelling, and long pre-preparation cycles are the issues that exist in biological treatment process [10,11,12,13]. However, biological methods are susceptible to environmental influences, have a long start-up time, and have a limited treatment effect on refractory pollutants in water [14]. Chemical treatment technologies are also widely used in wastewater treatment, including traditional oxidation methods, such as Fenton, ozonation, and photolysis, and advanced oxidation processes (AOPs), such as ferrite, photo-Fenton, photocatalysis, the solar-driven process, the ultrasonic process, and the electric Fenton process [15]. Among various chemical treatment methods, AOPs are considered to be the best way to treat wastewater due to their high oxidation efficiency and absences of secondary pollution [16].
Several conditions influence the AOPs efficiency, the most important of which are the properties of the catalysts. There are many influencing factors of catalysts required for catalytic oxidation, including ambient temperature, pH value, ionic strength, and the chemical properties of pollutants [17]. In the study of AOPs’ process catalysts, there are too many variables, scattered tests, and traditional experimental trials and error methods, which lack efficiency, timeliness, and innovation and consume a lot of manpower, material, and financial resources. In this regard, more comprehensive efforts are urgently needed to synthesize the effects of various factors on the catalytic process and to predict future research in this field [18].
Machine learning, as a potent data-driven technology, has experienced rapid development over the past two decades. It can mine latent rules and knowledge from vast amounts of data. Machine learning algorithms have been successfully applied to the classification, regression, clustering, and dimensionality reduction tasks of high-dimensional datasets, demonstrating its extraordinary power and capability across numerous fields. For example, Li et al. [19] discussed machine learning-assisted methods for estimating wetland carbon sinks and the factors affecting the wetland carbon cycle. Zeng et al. [20] utilized machine learning to assist in predicting the phase composition of chemical vapor deposition of boron carbide. Latif et al. [21] developed supervised machine learning models to systematically predict persistent free radicals (PFRs). Huang et al. [22] combined machine learning model for the prediction of time–temperature–transformation diagrams of high-alloy steels, and other researchers have combined machine learning with mine operations and the classification of rock discontinuity tracking [23,24]. In particular, the development of machine learning algorithms has greatly compensated for the shortcomings of traditional trial-and-error experimental methods for catalyst development, as ML models can predict the impact of atomic-level features on real-world performance, thereby bridging the gap between theoretical exploration and practical application [25]. This review summarizes the steps of ML in the application of advanced catalytic process catalysts. Due to the lack of research summaries in this field, this paper has guiding significance for the development of machine learning in advanced catalytic process catalysts.
We searched for nearly two decades of machine learning-assisted catalyst development for advanced oxidation processes. Upon conducting a comprehensive literature review, a collection of studies pertaining to the development of machine learning-assisted catalysts for advanced oxidation processes, spanning from 2000 to 2024, was procured from the Web of Science Core Collection. The most prolific publications, countries, and journals involved in the machine learning-assisted catalyst development for advanced oxidation processes were identified. At the same time, a world map was provided, which shows the global research activity in ML catalysts for AOPs. The results are shown in Figure 1 and Figure 2. From Figure 1, it can be seen that, since 2018, the research on machine learning-assisted catalyst development, especially in the process of AOPs, has exploded exponentially. Figure 2 shows that global research activity is concentrated in China, the United States, and European industrial powerhouses, with Asia dominating.

2. Advanced Oxidation Process of Wastewater Treatment

The AOPs are mainly used, because the process can produce reactive oxygen species (ROS). From a technical perspective, the reactive oxygen species (ROS) encompasses both free radicals and non-radical species. Free radicals are typically associated with molecules possessing one or more unpaired electrons [26], examples of which include hydroxyl (•OH), superoxide (O2•−), peroxyl (RO2•), and hydroperoxyl (HO2•) radicals. These free radicals can readily transform from non-radical precursors, such as hydrogen peroxide (H2O2), hypochlorous acid (HOCl), and ozone (O3), as well as singlet oxygen (1O2) [27]. Principally, transition metals and/or their oxides, such as Fe, Co, and Mn, are acknowledged as potential catalysts for certain oxidants, including hydrogen peroxide (H2O2) [28], peroxymonosulfate (PMS), and peroxydisulfate (PDS) [29]. The main AOPs and ROS types are shown in Table 1. Some mechanisms of advanced oxidation processes are shown in Figure 3 [30].

3. Machine Learning Algorithms

ML is able to guide machines in efficiently processing data. At times, the information extracted from the data is difficult for us to comprehend. This is where machine learning plays a role. With the increasing richness of available datasets, the demand for machine learning continues to rise. Currently, many enterprises place great importance on the role of machine learning in data extraction. The purpose of machine learning is to learn from the data [51,52]. The main types of machine learning models are shown in Figure 4.
ML constitutes a significant domain within the contemporary computing landscape. Extensive research endeavors have been dedicated to the development of intelligent machines. The process of learning, an inherent human trait, has been integrated as a fundamental component into machine systems. Numerous methodologies have been formulated to achieve this objective. Conventional ML algorithms have found application across a multitude of fields [53,54]. Given the multitude of model algorithms available in ML, it is capable of selecting the most appropriate computational model method to suit complex environments, thereby facilitating the resolution of practical problems more effectively [55,56]. For example, the neural network algorithm has a strong nonlinear fitting ability; it can automatically learn and capture the complex nonlinear relationship in the environmental data and fit a suitable model by adjusting the connection weights between neurons [57]. Zhao et al. [58,59] used microbial network analysis, and the rhizosphere microbial community mainly promotes nitrogen metabolism by forming a tightly connected network. Gu et al. [60] utilized support vector regression (SVR) to rapidly and accurately predict the piezoelectric coefficient d33 for the design of lead-free high-performance piezoelectric ceramics. He et al. [61] used a Deep neural network (DNN) optimized by Full spectrum prediction (FSP) to accurately predict the information of ferroelectric phase transitions. Not only that, artificial intelligence is also expected to be applied to the field of microorganisms, such as the regulation of fungal–algae relationships and the auxiliary arbuscular mycorrhizal-assisted ENM fixation [62,63,64].

4. Catalyst Development for ML-Assisted Advanced Oxidation Water Treatment Processes

The intricate and unpredictable nature of experimental conditions poses significant challenges in comprehending the micro-level essence of catalyst activity, resulting in diminished research productivity and a lack of direction [65]. However, with the swift advancement of theoretical computation and data analytics methodologies, density functional theory (DFT) and ML offer promising solutions to these challenges. Grounded in the foundational principles of quantum mechanics (QM), DFT can precisely delineate the electronic configuration of catalyst systems, thereby establishing a robust theoretical groundwork for the subsequent elucidation of the formation mechanisms and attributes of active sites [66]. ML, endowed with its formidable data processing capabilities and pattern recognition strengths, can extract potential regularities from extensive experimental data and theoretical computation outcomes, facilitating the swift prediction of active sites and the efficient selection of catalysts. The synergistic integration of these two methodologies is anticipated to engender revolutionary advancements in the realm of catalyst design and evolution. The general workflow for the establishment of catalyst ML models is delineated in Figure 5. Initially, a dataset encompassing a variety of catalysts must be generated. Subsequently, each catalyst is delineated by its characteristics, which include electronic structural attributes, physical properties, and atomic characteristics [67]. These features are required to encapsulate the pivotal physical and chemical properties of the material, be less computationally intensive than the target properties, and possess the ability to uniquely identify each material. Thereafter, ML techniques can be employed to discern patterns, construct models, or uncover descriptors. These descriptors serve to correlate the properties of the catalysts with their performance metrics, ultimately culminating in the optimization of catalyst structures. Taking the treatment of industrial wastewater as an example, based on existing research findings, Table 2 presents the connections among different types of wastewater, advanced oxidation processes, the selected catalysts, and the machine learning models and algorithms. It is evident that machine learning methods excel at selecting the optimal catalysts based on the various types of industrial wastewater. Duan et al. [68] used Bayesian optimization methods to adjust the hyperparameters of ML models and developed a regression model for elemental composition–catalytic activity to accelerate the discovery of entropy-stabilized oxide catalysts for catalytic oxidation. Liu et al. [69] investigated the NORR limiting potential and the inherent properties of the catalyst using random forest and SISSO machine learning models and rationally designed a Pt-anchored single-atom synthetic electrocatalyst in conjunction with DFT.
Table 2. The specific advanced oxidation processes for treating various types of industrial wastewater using machine learning-assisted catalysts and the ML models and algorithms employed as well as their advantages.
Table 2. The specific advanced oxidation processes for treating various types of industrial wastewater using machine learning-assisted catalysts and the ML models and algorithms employed as well as their advantages.
Industrial Wastewater
Category		Typical
Pollutants		AOPsML Models/AlgorithmsSelected Catalyst TypesML Implementation
Mechanism		Advantages of ML-Driven
Catalyst Screening		References
Pharmaceutical/
Textile Industry
Effluent		Antibiotics (e.g., sulfamethoxazole) pharmaceutical
residue, Azo dyes, recalcitrant organic compounds		Fenton/Photo-Fenton systemsArtificial neural network (ANN), deep neural network (DNN), random forest (RF)Fe-based catalystsUsing RF and ANN to predict the reaction characteristics of AOPs and then optimize the screening of catalysts. DNN predicts the best catalytic outcome.1. Resolves nonlinear relationships between catalyst properties and
activity;
2. Minimizes experimental
redundancy;
3. Optimizes multivariate reaction parameters.		[70,71,72]
Electroplating/
Metallurgical Effluent		Heavy metals (Cr6+, Cu2+) with co-existing organicsElectro-Fenton processes,
photocatalytic
oxidation		Convolutional neural networks (CNN), genetic
algorithms (GA), gradient boosting regression (GBR)		Pt-nanoparticles, TiO2-based nanocompositeCNN is used for image analysis and the extraction of material features. GA detects the binding energy on the Pt surface to enable the catalyst to achieve maximum activity. The powerful search capability of the GBR model can be employed to identify target catalysts.1. Optimize the relationship between structure and performance;
2. Improve the accuracy of
screening.		[73,74,75,76]
Agrochemical
Effluent		Organochlorines (e.g., DDT),
Coronated hydrocarbons		Ozone oxidation, Fenton, UVRandom forest (RF), Bayesian optimization (BO), CNN, ANNTiO2-based nanocompositeThe generalized RF model has good predictive ability and is prospective to predict the photocatalytic degradation performance of organic pollutants with the experimental data of BO.1. Improve the efficiency of catalyst prediction;
2. Uncovers synergistic interfacial effects.		[77,78,79,80]

4.1. Active Site Determination and Catalyst Screening

Generally, catalytic nanoparticles (NPs) are frequently utilized due to their significant specific surface area, which aids in the conservation of costly catalytic metals, such as rhodium and platinum [81]. These nanoparticles are characterized by surfaces that consist of various surface sites, and their catalytic activity catalytic activity is dominated by sparse high-energy active sites. The identification of these active sites is a complex task, which necessitates a combination of experimental and theoretical analysis on practical nanoparticles and model catalysts. Therefore, single crystal (SC) surfaces with distinct surface sites have been extensively utilized to elucidate the contribution of each surface site [82]. After continuous research, it has been found that DFT kinetic analysis is particularly beneficial for elucidating surface sites.
Catalytic activities are frequently governed by a limited number of specific surface sites, and the creation of these active sites is pivotal for the realization of high-performance catalysts [83]. The significant advancements in modern surface science have enabled the replication of catalytic reaction rates through the modeling of the arrangement of surface atoms using well-defined single-crystal surfaces. Nonetheless, this approach encounters limitations when dealing with highly inhomogeneous atomic configurations, such as those found on alloy nanoparticles with atomic-scale defects, where the arrangement cannot be simplified into single crystals. To address this, we propose a machine learning framework based on localized similarity kernels, enabling the systematic analysis of its catalytic behavior through atomically resolved environmental descriptors. Jinnouchi et al. [84] developed a machine learning framework capable of accurately predicting the thermodynamic energetics of catalytic reactions occurring on single-crystalline nanoparticle surfaces. When integrated with kinetic analysis, this approach elucidates atomic-scale active site distributions and predicts catalytic activity trends dependent on nanoparticle size and chemical composition. A schematic representation of the algorithm is provided in Figure 5A. Furthermore, Figure 5B–F presents computational predictions for binding energies of N, O, and NO intermediates during NO direct decomposition, and the formation energies of Rh(1−x)Aux alloys relative to pure Rh and Au bulk phases. Li et al. [85] developed a machine learning framework for the rapid screening of bimetallic catalysts through kinetic analysis and descriptors. A database containing the characteristics of adsorption energy and active sites was constructed for training artificial neural networks. These characteristics are based on chemical adsorption theory, encompassing the properties of adsorption sites and primary metal atoms. The model, trained on the methanol electrooxidation reaction, effectively predicts adsorbate–metal interactions and demonstrates strong predictive capabilities in exploring the chemical space of bimetallic catalysts. The characteristic importance analysis reveals the factors affecting the adsorbate/metal interaction, which helps to break the energy scaling limit.
Catalysts 15 00282 g005
Figure 5. An ML-related algorithm that depends on the catalytic active site determined by its local atomic configuration. (A) Schematic of the algorithm. (B) Mean absolute error σ in the predicted binding energies of O on the Rh(1−x).Au(x) SC surfaces as a function of the number of used training data Ndata that equal the number of basis sets Nbasis in this calculation. (C–E) Predicted binding energies (Eb,NML, Eb,OML, Eb,NOML) of (C) N, (D) O, and (E) NO on Rh(1−x).Au(x) SCs and NPs as a function of those obtained by DFT (Eb,NDFT, Eb,ODFT, Eb,NODFT). R2 indicates the correlation coefficient. (F) Predicted formation energies (ΔEML) of Rh(1−x). Au(x)s SCs and NPs from the pure Rh and Au bulks as a function of those obtained by DFT (ΔEML) [84].
Figure 5. An ML-related algorithm that depends on the catalytic active site determined by its local atomic configuration. (A) Schematic of the algorithm. (B) Mean absolute error σ in the predicted binding energies of O on the Rh(1−x).Au(x) SC surfaces as a function of the number of used training data Ndata that equal the number of basis sets Nbasis in this calculation. (C–E) Predicted binding energies (Eb,NML, Eb,OML, Eb,NOML) of (C) N, (D) O, and (E) NO on Rh(1−x).Au(x) SCs and NPs as a function of those obtained by DFT (Eb,NDFT, Eb,ODFT, Eb,NODFT). R2 indicates the correlation coefficient. (F) Predicted formation energies (ΔEML) of Rh(1−x). Au(x)s SCs and NPs from the pure Rh and Au bulks as a function of those obtained by DFT (ΔEML) [84].
Catalysts 15 00282 g005
Other researchers have also carried out research on the ML−assisted DFT prediction of catalyst active sites. Tamtaji et al. [86] used the density functional theory (DFT) in the Vienna ab initio simulation package (VASP5.4.4). They used the Perdew–Burke–Emzerhof (PBE) functional and a spin-polarized DFT calculation of the cut-off energy of a 500 eV plane wave. Deng et al. [87] explored the structure–property correlation and catalytic activity origin of BACs coordinated by metal dimers and nitrogen-doped graphene (NC) through density functional theory simulation combined with machine learning techniques.

4.2. Finding Descriptors and Patterns in Catalysis Data

Catalytic descriptors are representations of the reaction features extracted from raw experimental data, such as reaction conditions, catalysts, and reactants, and characterizes the target properties in a format suitable for machine recognition, including product efficiency, reaction specificity, surface binding strength, and activation energy, with the format optimized for computational analysis. While the selection of an ML algorithm is of paramount importance, the definition of these descriptors is pivotal to the predictive accuracy of machine learning models; algorithm optimization can only approximate the upper bound of accuracy to the extent possible [87,88]. Furthermore, in the process of rationally designing high-performance catalysts, the key lies in a deep understanding of the quantitative structure–activity relationships, which can reveal the intrinsic connections between different catalytic descriptors.
The catalytic data cover multiple levels of information, including the physical and chemical properties of the catalysts (such as crystal structure, elemental composition, specific surface area, etc.), reaction conditions (temperature, pressure, reactant concentration, etc.), and reaction results (conversion, selectivity, etc.) [89,90]. These data sources are extensive, and there are complex nonlinear relationships between them, which increases the difficulty of data analysis. In the actual catalytic research, due to the limitation of experimental cost and time, the experimental data obtained are often limited. This leads to the sparse distribution of data in high-dimensional space, making it difficult for traditional statistical methods to accurately establish models and discover laws. In the process of experimental measurement, noise will inevitably be introduced, such as instrument error, sample preparation difference, and so on. These noises will interfere with the real signal, further conceal the intrinsic patterns and relationships in the catalytic data, and bring challenges to data analysis.
ML-optimized catalytic descriptors play a critical role in elucidating reaction mechanisms, establishing quantitative structure–activity relationships (QSARs), and enabling predictive modeling of catalytic performance. For example, in heterogeneous catalysis, the electronic structure, geometric structure, and other related parameters of surface-active sites are often used as descriptors to correlate catalyst activity.
The correlation between methane conversion and ethylene selectivity is depicted in the scatter plot (Figure 6a) as a function of temperature. Ishioka et al. [88] employed hierarchical clustering techniques to categorize the data into three clusters, each targeting C2 selectivity. Nineteen physicochemical parameters were chosen as descriptors and, subsequently, predicted using random forest and support vector classifier (SVC) methodologies. The cross-validation scores ranged from 0.67 to 0.84. Through feature importance analysis, five critical descriptors were identified. Leveraging these descriptors and predictive models, 62,196 catalysts exhibiting high C2 selectivity were forecasted, and three novel catalysts were experimentally validated (Figure 6b,c). The physical significance of the descriptors was also elucidated, revealing that the first ionization energy, electron affinity, and electronegativity of that catalysts with high C2 selectivity are generally below the average, while their second ionization energy and density approximate the mean values [88,89,90]. This suggests that fundamental physical parameters can serve as efficacious descriptors for the design of catalytic systems.

4.3. Machine-Learned Interatomic Potentials for Catalyst Simulation

The utilization of QM to simulate catalysts within reaction environments incurs substantial computational expenses. The primary limitation in applying QM to catalytic systems—typically containing hundreds of atoms—stems from the disproportionate rise in computational costs associated with increasing system size. To overcome this scale-related constraint, efforts are underway to leverage ML in designing interatomic potentials: mathematical functions that compute the potential energy of atomic systems. These potentials are refined from data generated through quantum mechanical calculations [89,90]. In comparison to traditional QM methods, ML interatomic potentials (MLIPs) offer a more numerically efficient estimation of interaction energies. Notably, recent advancements have shown that MLIPs hold significant promise, achieving accuracy levels that approach those of DFT.
Chen et al. [91] integrated ML atomic interaction potentials (MLIPs) with active learning methodologies to expedite explicit solvent simulations concerning adsorption and reaction processes on heterogeneous catalysts. The MLIPs, once trained, have the capability to enhance the velocity of molecular dynamics simulations by four orders of magnitude, while preserving a high degree of accuracy in simulating water and metal–water interfaces. Utilizing the ML-accelerated simulations, precise predictions were made for the adsorption energies of CO*, OH*, COH*, HCO, and OCCHO on copper surfaces, as well as the free energy barrier associated with the cleavage of ethylene glycol’s C-H bond. The MLaMD simulation has showcased its potential in modeling intricate catalytic reactions, and it has been employed to ascertain the free energy barrier for the C-H bond cleavage of ethylene glycol on Cu(111) and Pd(111) surfaces. The two-dimensional free energy surface is delineated using C-H and C-Cu bond distances as collective variables. On Cu(111), the C-H bond cleavage is concurrent with ethylene glycol’s approach to the surface, and the total energy barrier amounts to 1.28 electron volts (Figure 7). Conversely, on Pd(111), the C-H bond cleavage coincides with ethylene glycol’s approach to the surface, with an energy barrier of 0.93 electron volts.
Chan et al. [92] discussed some recent work on using ML to combine the accuracy and flexibility of electronic structure calculations with the speed of conventional potentials. Yu et al. [93] summarize and compare the most widely used supervised and unsupervised ML methods and then discuss the ML-based emerging interatomic models, as follows: Gaussian approximation potential, spectral nearest neighbor analysis potential, deep potential molecular dynamics, SCHNET, hierarchical interaction particle neural network, and the fast learning of rare atomic events.

4.4. Accelerating the Discovery of Catalytic Mechanisms

Machine learning techniques can significantly enhance the efficiency of transition state (TS) search and minimum energy path (MEP)-finding algorithms [94]. MEP refers to the lowest energy path connecting two local minima on the potential energy surface, and thus, it is of great importance in dynamic analysis. To accelerate the computational speed of MEP and TS searches, researchers have employed neural networks (NN) trained with DFT to estimate the potential energy surface (PES) and then perform nudged elastic band (NEB) calculations.
ML methodologies exhibit significant potential for simplifying the intricacies inherent in reaction networks, thereby facilitating advancements in mechanism research [94,95]. Despite the capacity of quantum mechanical modeling to elucidate the specifics of small molecule reactions and enhance catalyst efficacy, the substantial computational expenses associated with macromolecular complex reaction networks constrain its widespread application. Researchers have employed the Gaussian process regression (GPR) optimization framework to attain precise and expedient network calculations and predictions. For instance, in the investigation of the syngas reaction on the Rh (111) catalyst (Figure 8A), the free energy of all intermediate species was prognosticated using the GPR method, and the prospective rate-limiting steps were ascertained through a fundamental classification technique. Through the iterative refinement of the GPR model, in conjunction with the application of the nudged elastic band algorithm featuring a climbing image, the viable reaction network was discerned (Figure 8B) [96]. ML also demonstrates analogous potential in the realm of catalytic mechanism discovery, including the domain of advanced oxidation catalysis [97].

5. Future Development Directions and Challenges

It is evident that ML has emerged as a transformative computational paradigm, offering unprecedented opportunities to decipher complex catalytic mechanisms and accelerate materials discovery. However, the development of ML frameworks with quantitative predictive power and their effective deployment in catalysis science encounter significant hurdles, as summarized in Figure 9. These obstacles encompass the paucity of comprehensive datasets that fail to fully encapsulate the input variables and output functionalities of catalytic processes, a dearth of predictive models grounded in reaction kinetics, an insufficient understanding of the actual active sites on solid catalysts, and an inadequate integration of catalyst deactivation phenomena in the prediction models for catalyst screening [98,99]. The following sections analyze these limitations through the lens of contemporary research and propose targeted strategies to advance this interdisciplinary frontier.

5.1. Shared Database Establishment

ML techniques exhibit strong dependence on extensive, high-quality datasets to achieve robust learning outcomes and reliable predictive accuracy [99]. In the context of ML-driven catalyst development, data scarcity emerges as a fundamental constraint, critically impairing model training efficacy and limiting performance optimization. From the perspective of experimental data acquisition, the research and development experiments of catalysts are often accompanied by high costs and complex operation processes.
In terms of theoretical data, while first-principles simulations (e.g., DFT) provide critical theoretical insights into catalyst electronic structures and adsorption energetics, their application to complex catalytic systems, such as multicomponent alloys or hierarchically porous materials with multiscale structural features, incurs prohibitive computational overhead. High-fidelity ab initio simulations of such systems demand extensive computational resources and time investments [100,101]. The lack of data makes it difficult for ML models to fully learn the complex relationship between catalyst structure and performance during the training process, and it is easy to fall into the dilemma of overfitting, that is, overlearning detailed features on limited data and ignoring general rules, resulting in the poor generalization ability of the model and inability to accurately predict the performance of the new catalyst system, which greatly hinders the development process of new and efficient catalysts. Furthermore, data quality challenges, including data noise and errors, are significant and unavoidable issues in the application of ML in the field of catalysts.
To address the challenges of data paucity and inconsistency, it is imperative for researchers and institutions to enhance collaborative efforts and establish a unified, open-access database for AOP catalysts [102]. This database must adhere to standardized protocols for data entry, encompassing comprehensive details, such as the chemical composition, preparation methodologies, microstructural characterization, and conditions and outcomes of catalytic performance assessments. Additionally, the implementation of a rigorous data audit framework is essential to guarantee the precision and dependability of the contributed data. The data-sharing platform facilitates the seamless acquisition and contribution of data among researchers, thereby accelerating the rate of data accumulation and application. For instance, the geometric transformation technique, commonly employed in image data processing, can be adapted to expand microstructural diversity, while preserving catalyst symmetry constraints [103]. Generative adversarial networks can produce synthetic data that mirror real-world data by learning its distribution, thereby further enriching the dataset. Nonetheless, in the application of data augmentation techniques, it is crucial to ensure the generated data’s validity and authenticity and to prevent the introduction of excessive noise or erroneous information.

5.2. Training and Development of Model

Despite the transformative potential of machine learning (ML) in accelerating catalyst discovery and optimization, the inherent complexity of state-of-the-art algorithms poses significant interpretability challenges. For instance, deep neural networks (DNNs), a cornerstone of modern deep learning, demonstrate exceptional predictive accuracy in classifying catalytic behaviors and forecasting activity trends. However, their opaque “black-box” nature stems from highly intricate architectures, characterized by high-dimensional parameter spaces encompassing millions of tunable weights [104].
When researchers explore the relationship between catalyst active sites and reaction selectivity, they typically only obtain the final predictions of the model and cannot understand how the model deduces the role of active sites and their impact on reaction selectivity based on catalyst structure, electron cloud density, reactant concentration, and other information. This limitation restricts theoretical interpretation and the construction of new catalytic theories. In practical applications, when predictions do not match experiments, it is difficult for researchers to troubleshoot and optimize the model, nor can they adjust catalyst design strategies, limiting the application of machine learning in the field of catalysis [105,106]. In specific studies, various machine learning models should be compared, representative models should be selected for training and evaluation based on data characteristics and research questions, the best-performing model should be chosen, and model integration techniques, such as voting, averaging, and stacking, should be used to improve model stability and generalization capabilities.

5.3. Cross-Latitude Fusion

AOPs are complex physical and chemical processes involving multiple scales, from the structure and electronic state of the catalyst active site at the microscopic atomic scale to the morphology and pore structure of the catalyst particles at the mesoscopic scale, and then, to the fluid dynamics and mass transfer and heat transfer process of the reaction system at the macroscopic scale. While ML has demonstrated proficiency in modeling isolated scale-specific behaviors, current ML frameworks lack the systematic integration of cross-scale dependencies. For example, at the micro level, QM calculations can accurately describe the electronic structure and chemical reaction mechanism of the active sites of the catalysts, but how to effectively integrate these microscopic information with the catalytic activity and selectivity observed in macro experiments is still an unsolved problem. While current efforts have emerged to integrate multiscale descriptors into machine learning frameworks [107], most of the methodologies mainly remain in the stage of practical exploration. Establishing cross-scale correlations in multiscale systems necessitates both fundamental mechanistic insights into physicochemical processes across discrete spatiotemporal regimes and innovative advances in designing algorithms, as well as models, to achieve the effective integration and collaborative computation of data at different scales.
In order to establish cross-scale correlation, it is necessary to use a variety of multiscale modeling methods. Combining microscopic simulation methods, such as QM and molecular dynamics (MD), with macroscopic reaction kinetics models, a hierarchical multiscale coupling model was constructed. At the micro level, the QM method can accurately calculate the electronic structure and chemical reaction energy barrier of the active site of the catalysts and provide atomic-scale information for understanding the catalytic reaction mechanism. Molecular dynamics simulation can study the molecular adsorption, diffusion, and reaction process on the surface of the catalysts and reveal the microscopic dynamic behavior. At the macro level, the reaction kinetic model can describe the overall behavior of the reaction system, such as reaction rate, conversion rate, etc., [108,109,110]. Through machine learning algorithms, data transfer and collaborative computing between different scale models can be realized. For example, the reaction rate constant obtained by the microscopic simulation is used as the input parameter of the macroscopic reaction kinetic model, and the microscopic model is verified and modified by the macroscopic experimental data, and the complete correlation from the microstructure to the macroscopic performance is gradually established.

5.4. Reverse Design of Catalysts

Machine learning (ML) algorithms exhibit distinct theoretical foundations and domain-specific applicability, necessitating judicious model selection based on target catalytic properties. Inverse catalyst design is an emerging paradigm in computational catalysis, which refers to the design and development of the catalysts, according to the specific catalyst performance [111]. This process utilizes the powerful data processing and pattern recognition capabilities of ML. In the process of the reverse development and design of ML-assisted AOP catalysts, the shortcomings of this field are also exposed. Selecting the appropriate model is a complex and critical problem. Due to the high nonlinearity, complexity, and diversity of the data involved in this field, the fitting ability and generalization ability of different models are significantly different. For example, the artificial neural network model shows a strong advantage in dealing with complex nonlinear relationships and can automatically learn the deep features in the data, which is very helpful for the analysis of complex reaction mechanisms and the relationship between material structure and performance in the reverse development and design of AOP catalysts. However, it is easy to fall into the local optimal solution during the training process, resulting in overfitting, especially in the case of limited data, and the reverse development and design of AOP catalysts often have limited data. The decision tree model has good interpretability and can intuitively show the classification and decision-making process of data [112,113]. However, in the reverse development and design of AOP catalysts, the fitting ability of complex continuous variables and highly nonlinear data is relatively weak. The support vector machine model has certain advantages in dealing with small sample data and high-dimensional data. In the reverse development and design of AOP catalysts, the processing of small sample experimental data or high-dimensional catalyst characterization data may play a role, but it is very sensitive to the choice of kernel function. Different kernel functions will lead to significant differences in model performance. Therefore, in the practical application of the ML-assisted reverse development and design of AOP catalysts, it is necessary to carefully evaluate and select the most suitable model, according to specific data characteristics, problem types, and research objectives. The effect of the machine learning model depends on the hyperparameters, including the architecture of the neural network, the number of neurons, and the learning rate.
The hyperparameters critically govern both the representational capacity of ML models in capturing catalyst–property relationships and the computational tractability of ML-driven inverse design frameworks for AOP catalysts. Manual hyperparameter adjusting remains computationally prohibitive, which is likely to undermine the reliability of ML-driven inverse design frameworks for advanced oxidation process (AOP) catalysts. In contrast, automated hyperparameter optimization techniques, such as genetic algorithms and particle swarm optimization, can significantly improve the efficiency of the adjustment process and find better parameter configurations. Hyperparameter tuning frameworks, including Hyperopt and Scikit Optimize, simplify the automation of this process [114,115]. Establishing a reasonable search space and termination criteria can prevent the waste of computing resources in the process of ML-assisted AOP catalyst reverse development and design.

6. Conclusions

This article examines the utilization of ML within the domain of AOP catalysts, underscoring its pivotal role in the evolution of catalyst development and the concomitant challenges, scrutinizes the employment of ML-augmented catalysts in the context of wastewater treatment, and elucidates the shortcomings of conventional screening methodologies. These traditional techniques are deemed urgently in need of more efficacious substitutes due to their protracted duration, diminished efficiency, and the intricate nature of the variables involved. The article conducts a review of scholarly literature from the past twenty years, cataloging a variety of ML algorithms, and accentuates their utility in managing high-dimensional datasets. Additionally, it discusses the swift advancement of ML in expediting catalyst screening, forecasting active sites, and conducting structural modeling, encompassing dataset generation, feature extraction, and model training. Moreover, the discourse addresses the significance of physical and chemical properties in forecasting catalyst performance. The application of ML methodologies for predicting active sites and screening catalysts opens up novel research avenues and underscores challenges, such as data integrity, data volume, and limitations inherent to the models.

Author Contributions

Writing—original draft preparation, Q.Y. (Qinghui Yuan); writing—review, X.W., D.X., H.L., H.Z. and Q.Y. (Qian Yu); review and editing, Y.B. and L.L.; funding acquisition, Y.B. and L.L.; supervision, L.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Research Fund of Binzhou Polytechnic (2021yjkt09), Natural Science Foundation of Heilongjiang Province (LH2023E125), and Postdoctoral Research Foundation of Heilongjiang University of Science and Technology (2023BSH03).

Data Availability Statement

No new data were created or analyzed in this study. Data sharing is not applicable to this article.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Saravanan, A.; Deivayanai, V.; Kumar, P.S.; Rangasamy, G.; Hemavathy, R.; Harshana, T.; Gayathri, N.; Alagumalai, K. A detailed review on advanced oxidation process in treatment of wastewater: Mechanism, challenges and future outlook. Chemosphere 2022, 308, 136524. [Google Scholar] [CrossRef] [PubMed]
  2. Wang, B.; Zhang, H.; Wang, F.F.; Xiong, X.G.Y.; Tian, K.; Sun, Y.B.; Yu, T.T. Application of heterogeneous catalytic ozonation for refractory organics in wastewater. Catalysts 2019, 9, 241. [Google Scholar] [CrossRef]
  3. Ma, D.S.; Yi, H.; Lai, C.; Liu, X.G.; Huo, X.Q.; An, Z.W.; Li, L.; Fu, Y.K.; Li, B.S.; Zhang, M.M.; et al. Critical review of advanced oxidation processes in organic wastewater treatment. Chemosphere 2021, 275, 130104. [Google Scholar] [CrossRef] [PubMed]
  4. Fraiha, O.; Hadoudi, N.; Najlae, Z.A.; Salhi, A.; Amhamdi, H.; Mourabit, F.; Ahari, M.H. Comprehensive review on the adsorption of pharmaceutical products from wastewater by clay materials. Desalination Water Treat. 2024, 317, 100114. [Google Scholar] [CrossRef]
  5. Li, L.X.; Han, J.Z.; Huang, X.H.; Qiu, S.; Liu, X.H.; Liu, L.L.; Zhao, M.J.; Qu, J.W.; Zou, J.L.; Zhang, J. Organic pollutants removal from aqueous solutions using metal-organic frameworks (MOFs) as adsorbents: A review. J. Environ. Chem. Eng. 2023, 6, 111217. [Google Scholar] [CrossRef]
  6. Chen, H.J.; Xiao, R.H.; Huang, D.L.; Deng, R.; Li, R.J.; Chen, Y.S.; Zhou, W. Three kinds of apatite adsorbents prepared by co-precipitation for Pb (II) and Cd (II) removal from wastewater: Performance, competitive effects and mechanisms. J. Mol. Liq. 2024, 400, 124478. [Google Scholar] [CrossRef]
  7. Kavitha, E.; Poonguzhali, E.; Nanditha, D.; Kapoor, A.; Arthanareeswaran, G.; Prabhakar, S. Current status and future prospects of membrane separation processes for value recovery from wastewater. Chemosphere 2022, 291, 132690. [Google Scholar] [CrossRef]
  8. Öcal, Z.B.; Öncel, M.S.; Keskinler, B.; Khataee, A.; Karagündüz, A. Sustainable treatment of boron industry wastewater with precipitation-adsorption hybrid process and recovery of boron species. Process Saf. Environ. Prot. 2024, 182, 719–726. [Google Scholar] [CrossRef]
  9. Bartolomeu, M.; Neves, M.G.; Faustino, M.A.; Almeida, A. Wastewater chemical contaminants: Remediation by advanced oxidation processes. Photochem. Photobiol. Sci. 2018, 17, 1573–1598. [Google Scholar] [CrossRef]
  10. Li, L.X.; He, Z.M.; Liang, T.J.; Sheng, T.; Zhang, F.G.; Wu, D.; Ma, F. Colonization of biofilm in wastewater treatment: A review. Environ. Pollut. 2022, 293, 118514. [Google Scholar] [CrossRef]
  11. Li, L.X.; Han, J.Z.; Huang, L.L.; Liu, L.L.; Qiu, S.; Ding, J.; Liu, X.H.; Zhang, J. Activation of PMS by MIL-53 (Fe)@ AC composites contributes to tetracycline degradation: Properties and mechanisms. Surf. Interfaces 2024, 51, 104521. [Google Scholar] [CrossRef]
  12. Li, L.X.; Chai, W.; Sun, C.Y.; Huang, L.L.; Sheng, T.; Song, Z.W.; Ma, F. Role of microalgae-bacterial consortium in wastewater treatment: A review. J. Environ. 2024, 360, 121226. [Google Scholar] [CrossRef] [PubMed]
  13. Zhang, C.; Zhou, M.W.; Zhang, Y.M.; Li, D.P.; Hou, N.; Zhao, X.Y. Novel strategies for enhancing energy metabolism and wastewater treatment in algae-bacteria symbiotic system through carbon dots-induced photogenerated electrons: The definitive role of accelerated electron transport. Chem. Eng. J. 2024, 500, 157016. [Google Scholar] [CrossRef]
  14. Crini, G.; Lichtfouse, E. Advantages and disadvantages of techniques used for wastewater treatment. Environ. Chem. Lett. 2019, 17, 145–155. [Google Scholar] [CrossRef]
  15. Ahmed, M.B.; Zhou, J.L.; Ngo, H.H.; Guo, W.; Thomaidis, N.S.; Xu, J. Progress in the biological and chemical treatment technologies for emerging contaminant removal from wastewater: A critical review. J. Hazard. Mater. 2017, 323, 274–298. [Google Scholar] [CrossRef]
  16. Ponnusami, A.B.; Sinha, S.; Ashokan, H.; Paul, M.V.; Hariharan, S.P.; Arun, J.; Gopinath, K.P.; Le, Q.H.; Pugazhendhi, A. Advanced oxidation process (AOP) combined biological process for wastewater treatment: A review on advancements, feasibility and practicability of combined techniques. Environ. Res. 2023, 237, 116944. [Google Scholar] [CrossRef]
  17. Zhang, S.Z.; Chen, S.; Jiang, H. A new tool to predict the advanced oxidation process efficiency: Using machine learning methods to predict the degradation of organic pollutants with Fe-carbon catalyst as a sample. Chem. Eng. Sci. 2023, 280, 119069. [Google Scholar] [CrossRef]
  18. Liu, S.Q.; Long, Z.Q.; Liu, H.Z.; Zhang, J.; Zhang, G.M.; Liang, J.S. Machine learning predict the degradation efficiency of aqueous refractory organic pollutants by ultrasound-based advanced oxidation processes. J. Water Process. Eng. 2024, 66, 106022. [Google Scholar] [CrossRef]
  19. Li, L.X.; Xu, H.B.; Zhang, Q.; Zhan, Z.S.; Liang, X.W.; Xing, J. Estimation methods of wetland carbon sink and factors influencing wetland carbon cycle: A review. Carbon Res. 2024, 3, 50. [Google Scholar] [CrossRef]
  20. Zeng, Q.; Gao, Y.; Guan, K.; Liu, J.; Feng, Z. Machine learning and a computational fluid dynamic approach to estimate phase composition of chemical vapor deposition boron carbide. J. Adv. Ceram. 2021, 10, 537–550. [Google Scholar] [CrossRef]
  21. Latif, J.; Chen, N.; Saleem, A.; Li, K.; Qin, J.J.; Yang, H.Q.; Jia, H.Z. Machine learning for persistent free radicals in biochar: Dual prediction of contents and types using regression and classification models. Carbon Res. 2024, 3, 39. [Google Scholar] [CrossRef]
  22. Huang, X.Y.; Wang, H.; Xue, W.H.; Ullah, A.; Xiang, S.; Huang, H.L.; Meng, L.; Ma, G.; Zhang, G.Z. A combined machine learning model for the prediction of time-temperature-transformation diagrams of high-alloy steels. J. Alloys Compd. 2020, 823, 153694. [Google Scholar] [CrossRef]
  23. Ma, X.; Chen, P.; Chen, C.; Feng, W.; Wang, Y. Prediction of surrounding rock stability of coal roadway based on machine learning and its application. J. Min. Sci. Technol. 2023, 8, 156–165. [Google Scholar]
  24. Wang, K.; Li, K.; Du, F.; Zhang, X.; Wang, Y.; Zhou, J. Prediction of coal-gas compound dynamic disaster based on convolutional neural network. J. Min. Sci. Technol. 2023, 8, 613–622. [Google Scholar]
  25. Zhang, X.L.; Zhu, T.; Yuan, B.; Wang, M.D.; Zhang, B.X. Combustion performance of sub-nanometer Pd/S-1 catalyst for ventilation air methane. J. Min. Sci. Technol. 2023, 8, 339–347. [Google Scholar] [CrossRef]
  26. Fernández-Castro, P.; Vallejo, M.; San Román, M.F.; Ortiz, I. Insight on the fundamentals of advanced oxidation processes. Role and review of the determination methods of reactive oxygen species. J. Chem. Technol. Biotechnol. 2015, 90, 796–820. [Google Scholar] [CrossRef]
  27. Clennan, E.L.; Pace, A. Advances in singlet oxygen chemistry. Tetrahedron 2005, 61, 6665–6691. [Google Scholar] [CrossRef]
  28. Battistella, B.; Ray, K. O2 and H2O2 activations at dinuclear Mn and Fe active sites. Coord. Chem. Rev. 2020, 408, 213176. [Google Scholar] [CrossRef]
  29. Ding, Y.B.; Wang, X.R.; Fu, L.B.; Peng, X.Q.; Pan, C.; Mao, Q.H.; Wang, C.J.; Yan, J.C. Nonradicals induced degradation of organic pollutants by peroxydisulfate (PDS) and peroxymonosulfate (PMS): Recent advances and perspective. Sci. Total Environ. 2021, 765, 142794. [Google Scholar] [CrossRef]
  30. Wang, J.L.; Wang, S.Z. Reactive species in advanced oxidation processes: Formation, identification and reaction mechanism. Chem. Eng. J. 2020, 401, 1385–8947. [Google Scholar] [CrossRef]
  31. Pignatello, J.J.; Oliveros, E.; MacKay, A. Advanced oxidation processes for organic contaminant destruction based on the Fenton reaction and related chemistry. Crit. Rev. Environ. Sci. Technol. 2006, 36, 1–84. [Google Scholar] [CrossRef]
  32. Rachmilovich-Calis, S.; Masarwa, A.; Meyerstein, N.; Meyerstein, D.; Van Eldik, R. New mechanistic aspects of the Fenton reaction. Chem. Eur. J. 2009, 15, 8303–8309. [Google Scholar] [CrossRef] [PubMed]
  33. Brillas, E.; Sirés, I.; Oturan, M.A. Electro-Fenton process and related electrochemical technologies based on Fenton’s reaction chemistry. Chem. Rev. 2009, 109, 6570–6631. [Google Scholar] [CrossRef]
  34. Wu, M.Q.; Guo, X.; Cao, Y.D.; Yu, H.C.; Hu, Z.R.; Yang, Y.; Yao, T.J.; Wu, J. Cascading H2O2 photosynthesis and fenton reaction for self-sufficient photo-fenton reactions: A review of recent advances. Chem. Eng. J. 2024, 489, 151091. [Google Scholar] [CrossRef]
  35. Dai, C.L.; Feng, Z.H.; Hu, Q.M.; Qiu, J.Y.; You, J.H.; Guo, R.; Liu, X.W.; Zhang, H.Z. Recent progress in modification and composite strategies of graphitic carbon nitride as catalysts for heterogeneous photo-Fenton reaction. Mater Sci. Semicond. Process 2023, 167, 107807. [Google Scholar] [CrossRef]
  36. Dong, C.C.; Fang, W.Z.; Yi, Q.Y.; Zhang, J.L. A comprehensive review on reactive oxygen species (ROS) in advanced oxidation processes (AOPs). Chemosphere 2022, 308, 136205. [Google Scholar] [CrossRef]
  37. Pandis, P.K.; Kalogirou, C.; Kanellou, E.; Vaitsis, C.; Savvidou, M.G.; Sourkouni, G.; Zorpas, A.A.; Argirusis, C. Key points of advanced oxidation processes (AOPs) for wastewater, organic pollutants and pharmaceutical waste treatment: A mini review. Chem. Eng. 2022, 6, 8. [Google Scholar] [CrossRef]
  38. Wang, X.C.; Jiang, J.N.; Zhou, M.H.; Dewil, R. Recent advances in H2O2-based advanced oxidation processes for removal of antibiotics from wastewater. Chin. Chem. Lett. 2023, 34, 107621. [Google Scholar] [CrossRef]
  39. Tian, F.X.; Ye, W.K.; Xu, B.; Hu, X.J.; Ma, S.X.; Lai, F.; Gao, Y.Q.; Xing, H.B.; Xia, W.H.; Wang, B. Comparison of UV-induced AOPs (UV/Cl2, UV/NH2Cl, UV/ClO2 and UV/H2O2) in the degradation of iopamidol: Kinetics, energy requirements and DBPs-related toxicity in sequential disinfection processes. Chem. Eng. J. 2020, 398, 125570. [Google Scholar] [CrossRef]
  40. Wang, H.; Zhang, S.Y.; He, X.W.; Yang, Y.Y.; Yang, X.T.; Van Hulle, S.W. Comparison of macro and micro-pollutants abatement from biotreated landfill leachate by single ozonation, O3/H2O2, and catalytic ozonation processes. Chem. Eng. J. 2023, 452, 139503. [Google Scholar] [CrossRef]
  41. Yuan, F.; Hu, C.; Hu, X.X.; Qu, J.H.; Yang, M. Degradation of selected pharmaceuticals in aqueous solution with UV and UV/H2O2. Water Res. 2009, 43, 1766–1774. [Google Scholar] [CrossRef]
  42. Jung, Y.J.; Kim, W.G.; Yoon, Y.; Kang, J.W.; Hong, M.Y.; Kim, H.W. Removal of amoxicillin by UV and UV/H2O2 processes. Sci. Total Environ. 2012, 420, 160–167. [Google Scholar] [CrossRef] [PubMed]
  43. Al-Mamun, M.R.; Kader, S.; Islam, M.S.; Khan, M.Z. Photocatalytic activity improvement and application of UV-TiO2 photocatalysis in textile wastewater treatment: A review. J. Environ. Chem. Eng. 2019, 7, 103248. [Google Scholar] [CrossRef]
  44. ES, E. Photocatalytic degradation of amoxicillin, ampicillin and cloxacillin antibiotics in aqueous solution using UV/TiO2 and UV/H2O2/TiO2 photocatalysis. Desalination 2010, 252, 46–52. [Google Scholar]
  45. Hua, I.; Hoffmann, M.R. Optimization of ultrasonic irradiation as an advanced oxidation technology. Environ. Sci. Technol. 1997, 31, 2237–2243. [Google Scholar] [CrossRef]
  46. Lin, C.C.; Hsiao, H.H. Degradation of Rhodamine B in water by heat/persulfate process. J. Taiwan Inst. Chem. Eng. 2022, 132, 104190. [Google Scholar] [CrossRef]
  47. Xu, Q.H.; Shi, F.; You, H.; Wang, S.T. Integrated remediation for organic-contaminated site by forcing running-water to modify alkali-heat/persulfate via oxidation process transfer. Chemosphere 2021, 262, 128352. [Google Scholar] [CrossRef]
  48. Yang, Y.; Lu, X.L.; Jiang, J.; Ma, J.; Liu, G.; Cao, Y.; Liu, W.L.; Li, J.; Pang, S.Y.; Kong, X.J.; et al. Degradation of sulfamethoxazole by UV, UV/H2O2 and UV/persulfate (PDS): Formation of oxidation products and effect of bicarbonate. Water Res. 2017, 118, 196–207. [Google Scholar] [CrossRef]
  49. Xiao, Y.J.; Zhang, L.F.; Zhang, W.; Lim, K.Y.; Webster, R.D.; Lim, T.T. Comparative evaluation of iodoacids removal by UV/persulfate and UV/H2O2 processes. Water Res. 2016, 102, 629–639. [Google Scholar] [CrossRef]
  50. Chen, G.Y.; Yu, Y.; Liang, L.; Duan, X.G.; Li, R.; Lu, X.K.; Yan, B.B.; Li, N.; Wang, S.B. Remediation of antibiotic wastewater by coupled photocatalytic and persulfate oxidation system: A critical review. J. Hazard. Mater. 2021, 408, 124461. [Google Scholar] [CrossRef]
  51. Mahesh, B. Machine learning algorithms-a review. Int. J. Sci. Res. 2020, 9, 381–386. [Google Scholar] [CrossRef]
  52. Mahmoodzadeh, A.; Fakhri, D.; Mohammed, A.H.; Mohammed, A.S.; Ibrahim, H.H.; Rashidi, S. Estimating the effective fracture toughness of a variety of materials using several machine learning models. Eng. Fract. Mech. 2023, 286, 109321. [Google Scholar] [CrossRef]
  53. Shinde, P.P.; Shah, S. A review of machine learning and deep learning applications. In Proceedings of the 2018 Fourth International Conference on Computing Communication Control and Automation (ICCUBEA), Pune, India, 16–18 August 2018; pp. 1–6. [Google Scholar]
  54. Feng, Y.; Wang, Q.H.; Wu, D.; Luo, Z.; Chen, X.J.; Zhang, T.Y.; Gao, W. Machine learning aided phase field method for fracture mechanics. Int. J. Eng. Sci. 2021, 169, 103587. [Google Scholar] [CrossRef]
  55. Sharma, N.; Sharma, R.; Jindal, N. Machine learning and deep learning applications-a vision. Glob. Transit. Proc. 2021, 2, 24–28. [Google Scholar] [CrossRef]
  56. Zhang, W.Y.; Su, Y.M.; Jiang, Y.F.; Hu, Z.Q.; Bi, J.T.; He, W.T. Data-driven fatigue crack propagation and life prediction of tubular T-joint: A fracture mechanics based machine learning surrogate model. Eng. Fract. Mech. 2024, 311, 110556. [Google Scholar] [CrossRef]
  57. Wu, Y.; Feng, J. Development and application of artificial neural network. Wirel. Pers. Commun. 2018, 102, 1645–1656. [Google Scholar] [CrossRef]
  58. Zhao, X.Y.; Meng, X.W.; Li, Q.L.; Ho, S.H. Nitrogen metabolic responses of non-rhizosphere and rhizosphere microbial communities in constructed wetlands under nanoplastics disturbance. J. Hazard. Mater. 2025, 484, 136777. [Google Scholar] [CrossRef]
  59. Zhao, X.Y.; Guo, M.R.; Wang, Y.N.; Jin, M.; Hou, N.; Wu, H.M. Toxic effects of nanoplastics on biological nitrogen removal in constructed wetlands: Evidence from iron utilization and metabolism. Water Res. 2024, 256, 121577. [Google Scholar] [CrossRef]
  60. Gu, W.; Yang, B.; Li, D.; Shang, X.; Zhou, Z.; Guo, J. Accelerated design of lead-free high-performance piezoelectric ceramics with high accuracy via machine learning. J. Adv. Ceram. 2023, 12, 1389–1405. [Google Scholar] [CrossRef]
  61. He, J.; Wang, C.; Li, J.; Liu, C.; Xue, D.; Cao, J.; Bai, Y. Machine learning assisted prediction of dielectric temperature spectrum of ferroelectrics. J. Adv. Ceram. 2023, 12, 1793–1804. [Google Scholar] [CrossRef]
  62. Zhang, X.; Wang, L.; Ma, F.; Yang, J.X. Effects of arbuscular mycorrhizal fungi on CH4 emissions from rice paddies. Int. J. Phytorem. 2017, 19, 39–45. [Google Scholar] [CrossRef] [PubMed]
  63. Wang, L.; Yang, D.G.; Ma, F.; Wang, G.; You, Y.Q. Recent advances in responses of arbuscular mycorrhizal fungi-Plant symbiosis to engineered nanoparticles. Chemosphere 2022, 286, 131644. [Google Scholar] [CrossRef] [PubMed]
  64. Wang, G.; Wang, L.; Ma, F.; Yang, D.G.; You, Y.Q. Earthworm and arbuscular mycorrhiza interactions: Strategies to motivate antioxidant responses and improve soil functionality. Environ. Pollut. 2021, 272, 115980. [Google Scholar] [CrossRef] [PubMed]
  65. Zahrt, A.F.; Henle, J.J.; Rose, B.T.; Wang, Y.; Darrow, W.T.; Denmark, S.E. Prediction of higher-selectivity catalysts by computer-driven workflow and machine learning. Science 2019, 363, 6424. [Google Scholar] [CrossRef]
  66. Sun, M.; Dougherty, A.W.; Huang, B.; Li, Y.; Yan, C.H. Accelerating atomic catalyst discovery by theoretical calculations-machine learning strategy. Adv. Energy Mater. 2020, 10, 1903949. [Google Scholar] [CrossRef]
  67. Fang, Z.L.; Li, S.Y.; Zhang, Y.J.; Wang, Y.X.; Meng, K.; Huang, C.Y.; Sun, S.R. The DFT and machine learning method accelerated the discovery of DMSCs with high ORR and OER catalytic activities. J. Phys. Chem. Lett. 2024, 15, 281–289. [Google Scholar] [CrossRef]
  68. Duan, X.; Li, Y.; Zhao, J.; Zhang, M.Y.; Wang, X.P.; Zhang, L.; Ma, X.X.; Qu, Y.; Zhang, P.F. Machine Learning Accelerated Discovery of Entropy-Stabilized Oxide Catalysts for Catalytic Oxidation. J. Am. Chem. Soc. 2024, 147, 651–661. [Google Scholar] [CrossRef]
  69. Liu, J.Y.; Guo, H.Q.; Xiong, Y.L.; Chen, X.; Yu, Y.F.; Wang, C.H. Rational design of Pt-anchored single-atom alloy electrocatalysts for NO-to-NH3 conversion by density functional theory and machine learning. Chin. J. Catal. 2024, 62, 243–253. [Google Scholar] [CrossRef]
  70. Acosta-Angulo, B.; Lara-Ramos, J.; Niño-Vargas, A.; Diaz-Angulo, J.; Benavides-Guerrero, J.; Bhattacharya, A.; Cloutier, S.; Machuca-Martínez, F. Unveiling the potential of machine learning in cost-effective degradation of pharmaceutically active compounds: A stirred photo-reactor study. Chemosphere 2024, 358, 142222. [Google Scholar] [CrossRef]
  71. Xie, Y.; Yang, F.; Liu, J.; Ren, M.J. Prediction of the degradation of organic pollutants by metal-activated peracetic acid using machine learning. J. Environ. Chem. Eng. 2024, 12, 113700. [Google Scholar] [CrossRef]
  72. Yadav, K.K.; Elboughdiri, N.; Fetimi, A.; Bhutto, J.K.; Merouani, S.; Tamam, N.; Alreshidi, M.A.; Rodríguez-Díaz, J.M.; Benguerba, Y. Enhanced wastewater treatment by catalytic persulfate activation with protonated hydroxylamine-assisted iron: Insights from a deep learning-based numerical investigation. Chemosphere 2024, 360, 142367. [Google Scholar] [CrossRef] [PubMed]
  73. Back, S.; Yoon, J.; Tian, N.; Zhong, W.; Tran, K.; Ulissi, Z.W. Convolutional Neural Network of Atomic Surface Structures to Predict Binding Energies for High-Throughput Screening of Catalysts. J. Phys. Chem. Lett. 2019, 10, 4401–4408. [Google Scholar] [CrossRef] [PubMed]
  74. Ooka, H.; Wintzer, M.E.; Nakamura, R. Non-Zero Binding Enhances Kinetics of Catalysis: Machine Learning Analysis on the Experimental Hydrogen Binding Energy of Platinum. ACS Catal. 2021, 11, 6298–6303. [Google Scholar] [CrossRef]
  75. Bai, Y.; Wilbraham, L.; Slater, B.J.; Zwijnenburg, M.A.; Sprick, R.S.; Cooper, A.I. Accelerated Discovery of Organic Polymer Photocatalysts for Hydrogen Evolution from Water through the Integration of Experiment and Theory. J. Am. Chem. Soc. 2019, 141, 9063–9071. [Google Scholar] [CrossRef]
  76. Mai, H.; Le, T.C.; Chen, D.; Chen, D.H.; David, A.W.; Rachel, A.C. Machine learning for electrocatalyst and photocatalyst design and discovery. Chem. Rev. 2022, 122, 13478–13515. [Google Scholar] [CrossRef]
  77. Jiang, Z.; Hu, J.; Zhang, X.; Zhao, Y.; Fan, X.; Zhong, S.; Zhang, H.; Yu, X. A generalized predictive model for TiO2-Catalyzed photo-degradation rate constants of water contaminants through artificial neural network. Environ. Res. 2020, 187, 109697. [Google Scholar] [CrossRef]
  78. Jiang, Z.; Hu, J.; Samia, A.; Yu, X. Predicting active sites in photocatalytic degradation process using an interpretable molecular-image combined convolutional neural network. Catalysts 2022, 12, 746. [Google Scholar] [CrossRef]
  79. Shang, Q.; Liu, X.; Zhang, M.; Zhang, P.; Ling, Y.; Cui, G.; Liu, W.; Shi, X.; Yue, J.; Tang, B. Photocatalytic degradation of ofloxacin antibiotic wastewater using TS-1/C3N4 composite photocatalyst: Reaction performance optimisation and estimation of wastewater component synergistic effect by artificial neural network and genetic algorithm. Chem. Eng. J. 2022, 443, 136354. [Google Scholar] [CrossRef]
  80. Alshehri, F.; Shahfahad; Rahman, A. Optimizing Machine Learning Models with Bayesian Techniques for Prediction of Groundwater Quality Index in Southwest Saudi Arabia. Earth Syst. Environ. 2024, 8, 1417–1436. [Google Scholar] [CrossRef]
  81. Biswas, P.; Wu, C.Y. Nanoparticles and the environment. J. Air Waste Manag. 2005, 55, 708–746. [Google Scholar] [CrossRef]
  82. Lauher, J.W.; Fowler, F.W.; Goroff, N.S. Single-crystal-to-single-crystal topochemical polymerizations by design. Acc. Chem. Res. 2008, 41, 1215–1229. [Google Scholar] [CrossRef] [PubMed]
  83. Chen, P.; Fryling, M.A.; McCreery, R.L. Electron transfer kinetics at modified carbon electrode surfaces: The role of specific surface sites. Anal. Chem. 1995, 67, 3115–3122. [Google Scholar] [CrossRef]
  84. Jinnouchi, R.; Asahi, R. Predicting catalytic activity of nanoparticles by a DFT-aided machine-learning algorithm. J. Phys. Chem. Lett. 2017, 8, 4279–4283. [Google Scholar] [CrossRef]
  85. Li, Z.; Wang, S.W.; Chin, W.S.; Achenie, L.E.; Xin, H.L. High-throughput screening of bimetallic catalysts enabled by machine learning. J. Mater. Chem. A 2017, 5, 24131–24138. [Google Scholar] [CrossRef]
  86. Tamtaji, M.; Kazemeini, M.; Abdi, J. DFT and machine learning studies on a multi-functional single-atom catalyst for enhanced oxygen and hydrogen evolution as well as CO2 reduction reactions. Int. J. Hydrogen Energy 2024, 80, 1075–1083. [Google Scholar] [CrossRef]
  87. Deng, C.F.; Su, Y.; Li, F.H.; Shen, W.F.; Chen, Z.F.; Tang, Q. Understanding activity origin for the oxygen reduction reaction on bi-atom catalysts by DFT studies and machine-learning. J. Mater. Chem. A 2020, 8, 24563–24571. [Google Scholar] [CrossRef]
  88. Ishioka, S.; Fujiwara, A.; Nakanowatari, S.; Takahashi, L.; Taniike, T.; Takahashi, K. Designing catalyst descriptors for machine learning in oxidative coupling of methane. ACS Catal. 2022, 12, 11541–11546. [Google Scholar] [CrossRef]
  89. Liu, W.X.; Zhu, Y.; Wu, Y.Q.; Chen, C.; Hong, Y.; Yue, Y.N.; Zhang, J.C.; Hou, B. Molecular dynamics and machine learning in catalysts. Catalysts 2021, 11, 1129. [Google Scholar] [CrossRef]
  90. Mortazavi, B.; Zhuang, X.; Rabczuk, T.; Shapeev, A.V. Atomistic modeling of the mechanical properties: The rise of machine learning interatomic potentials. Mater. Horiz. 2023, 10, 1956–1968. [Google Scholar] [CrossRef]
  91. Chen, B.W.J.; Zhang, X.L.; Zhang, J. Accelerating explicit solvent models of heterogeneous catalysts with machine learning interatomic potentials. Chem. Sci. 2023, 14, 8338–8354. [Google Scholar] [CrossRef]
  92. Chan, H.; Narayanan, B.; Cherukara, M.J.; Sen, F.G.; Sasikumar, K.; Gray, S.K.; Chan, M.K. Sankaranarayanan SK. Machine learning classical interatomic potentials for molecular dynamics from first-principles training data. J. Phys. Chem. C 2019, 123, 6941–6957. [Google Scholar] [CrossRef]
  93. Yu, W.; Ji, C.Y.; Wan, X.H.; Zhang, Z.F.; Robertson, J.; Liu, S.; Guo, Y.Z. Machine-learning-based interatomic potentials for advanced manufacturing. IJMSD 2021, 1, 159–172. [Google Scholar] [CrossRef]
  94. Sheppard, D.; Terrell, R.; Henkelman, G. Optimization methods for finding minimum energy paths. J. Chem. Phys. 2008, 128, 13. [Google Scholar] [CrossRef] [PubMed]
  95. Goldsmith, B.R.; Esterhuizen, J.; Liu, J.X.; Bartel, C.J.; Sutton, C. Machine learning for heterogeneous catalyst design and discovery. AIChE J. 2018, 64, 2311–2323. [Google Scholar] [CrossRef]
  96. Ma, S.; Liu, Z.P. Machine learning for atomic simulation and activity prediction in heterogeneous catalysis: Current status and future. ACS Catal. 2020, 10, 13213–13226. [Google Scholar] [CrossRef]
  97. Mohan, O.; Choksi, T.S.; Lapkin, A.A. The design and optimization of heterogeneous catalysts using computational methods. Catal. Sci. Technol. 2024, 14, 515–532. [Google Scholar]
  98. Li, H.B.; Jiao, Y.; Davey, K.; Qiao, S.Z. Data-driven machine learning for understanding surface structures of heterogeneous catalysts. Angew. Chem. Int. Ed. 2023, 62, e202216383. [Google Scholar]
  99. Guan, Y.N.; Chaffart, D.; Liu, G.H.; Tan, Z.Y.; Zhang, D.S.; Wang, Y.J.; Li, J.D.; Ricardez-Sandoval, L. Machine learning in solid heterogeneous catalysis: Recent developments, challenges and perspectives. Chem. Eng. Sci. 2022, 248, 117224. [Google Scholar] [CrossRef]
  100. Yang, W.; Fidelis, T.T.; Sun, W.H. Machine learning in catalysis, from proposal to practicing. ACS Omega 2019, 5, 83–88. [Google Scholar] [CrossRef]
  101. Zhang, T.S.; Shi, Y.B.; Liu, Y.; Yang, J.Y.; Guo, M.R.; Bai, S.W.; Hou, N.; Zhao, X.Y. A study on microbial mechanism in response to different nanoplastics concentrations in constructed wetland and its carbon footprints analysis. Chem. Eng. J. 2024, 480, 148023. [Google Scholar]
  102. Wang, R.P.; Chen, H.L.; He, Z.X.; Zhang, S.Y.; Wang, K.; Ren, N.Q.; Ho, S.H. Discovery of an end-to-end pattern for contaminant-oriented advanced oxidation processes catalyzed by biochar with explainable machine learning. Environ. Sci. Technol. 2024, 58, 16867–16876. [Google Scholar] [CrossRef] [PubMed]
  103. Pascacio, P.; Vicente, D.J.; Salazar, F.; Guerra-Rodríguez, S.; Rodríguez-Chueca, J. Predictive modeling of Enterococcus sp. removal with limited data from different advanced oxidation processes: A machine learning approach. J. Environ. Chem. Eng. 2024, 12, 112530. [Google Scholar] [CrossRef]
  104. Liu, Y.H.; Li, Z.Y.; Cao, C.Q.; Zhang, X.Z.; Meng, S.Q.; Davari, M.D.; Xu, H.J.; Ji, Y.; Schwaneberg, U.; Liu, L. Engineering of Substrate Tunnel of P450 CYP116B3 though Machine Learning. Catalysts 2023, 13, 1228. [Google Scholar] [CrossRef]
  105. Jaffari, Z.H.; Abbas, A.; Lam, S.M.; Park, S.; Chon, K.; Kim, E.S.; Cho, K.H. Machine learning approaches to predict the photocatalytic performance of bismuth ferrite-based materials in the removal of malachite green. J. Hazard. Mater. 2023, 442, 130031. [Google Scholar] [CrossRef]
  106. Oviedo, L.R.; Durzian, D.M.; Montagner, G.E.; Ruiz, Y.P.; Galembeck, A.; Pavoski, G.; Espinosa, D.C.; Nora, L.D.D.; da Silva, W.L. Supported heterogeneous catalyst of the copper oxide nanoparticles and nanozeolite for binary dyes mixture degradation: Machine learning and experimental design. J. Mol. Liq. 2024, 402, 124763. [Google Scholar] [CrossRef]
  107. Liu, M.H.; Li, N.; Meng, S.Y.; Yang, S.L.; Jing, B.J.; Zhang, J.Y.; Jiang, J.Z.; Qiu, S.; Deng, F.X. Bio-inspired Cu2O cathode for O2 capturing and oxidation boosting in electro-Fenton for sulfathiazole decay. J. Hazard. Mater. 2024, 478, 135484. [Google Scholar] [CrossRef]
  108. Li, L.X.; Liang, T.J.; Zhao, M.J.; Lv, Y.; Song, Z.W.; Sheng, T.; Ma, F. A review on mycelial pellets as biological carriers: Wastewater treatment and recovery for resource and energy. Bioresour. Technol. 2022, 355, 127200. [Google Scholar] [CrossRef]
  109. Yazdi, A.A.; Pirbazari, A.E.; Saraei, F.E.; Esmaeili, A.; Pirbazari, A.E.; Kohnehsari, A.A.; Derakhshesh, A. Design of 2D/2D β-Ni (OH)2/ZnO heterostructures via photocatalytic deposition of nickel for sonophotocatalytic degradation of tetracycline and modeling with three supervised machine learning algorithms. Chemosphere 2024, 352, 141328. [Google Scholar]
  110. Cüce, H.; Özçelik, D. Application of machine learning (ML) and artificial intelligence (AI)-based tools for modelling and enhancing sustainable optimization of the classical/photo-Fenton processes for the landfill leachate treatment. SDGs 2022, 14, 11261. [Google Scholar] [CrossRef]
  111. Yu, G.F.; Wu, Y.Q.; Cao, H.B.; Ge, Q.F.; Dai, Q.; Sun, S.H.; Xie, Y.B. Insights into the mechanism of ozone activation and singlet oxygen generation on N-doped defective nanocarbons: A DFT and machine learning study. Environ. Sci. Technol. 2022, 56, 7853–7863. [Google Scholar] [CrossRef]
  112. Liu, J.X.; Jia, H.; Mei, M.; Wang, T.; Chen, S.; Li, J.P. Efficient degradation of diclofenac by digestate-derived biochar catalyzed peroxymonosulfate oxidation: Performance, machine learning prediction, and mechanism. Process Saf. Environ. Prot. 2022, 167, 77–88. [Google Scholar] [CrossRef]
  113. Wang, L.J.; Yang, T.Y.; Wei, M.J.; Guan, R.Q.; Wei, W.; Jiang, J.Z. Machine learning and DFT dual-guidance of carbon dots implanted SrTiO3 hollow nanosphere for efficient all-pH-value photocatalysis. J Mater. Sci. Technol. 2025, 217, 169–181. [Google Scholar] [CrossRef]
  114. Bergstra, J.; Komer, B.; Eliasmith, C.; Yamins, D.; Cox, D.D. Hyperopt: A python library for model selection and hyperparameter optimization. CSD 2015, 8, 014008. [Google Scholar] [CrossRef]
  115. Liu, X.J.; Xu, P.C.; Zhao, J.J.; Lu, W.C.; Li, M.J.; Wang, G. Material machine learning for alloys: Applications, challenges and perspectives. J. Alloys Compd. 2022, 921, 165984. [Google Scholar] [CrossRef]
Catalysts 15 00282 g001
Figure 1. Literature analysis from 2000 to 2024.
Figure 1. Literature analysis from 2000 to 2024.
Catalysts 15 00282 g001
Catalysts 15 00282 g002
Figure 2. The global research activity in ML catalysts for AOPs.
Figure 2. The global research activity in ML catalysts for AOPs.
Catalysts 15 00282 g002
Catalysts 15 00282 g003
Figure 3. Elucidation of advanced oxidation process mechanisms. (a) Hydroxyl radicals are generated as intermediates in electrocatalytic oxidation processes. (b) Catalytic mechanism of ozone. (c) Possible activation pathways of persulfate. (d) Reactive species generated under ultrasonic irradiation [30].
Figure 3. Elucidation of advanced oxidation process mechanisms. (a) Hydroxyl radicals are generated as intermediates in electrocatalytic oxidation processes. (b) Catalytic mechanism of ozone. (c) Possible activation pathways of persulfate. (d) Reactive species generated under ultrasonic irradiation [30].
Catalysts 15 00282 g003
Catalysts 15 00282 g004
Figure 4. Tree diagram of main model algorithm of machine learning.
Figure 4. Tree diagram of main model algorithm of machine learning.
Catalysts 15 00282 g004
Catalysts 15 00282 g006
Figure 6. ML and physical quantities derived from the periodic table explore the descriptors of product selectivity in the oxidative coupling of methane (OCM) reaction. (a) C2 selectivity versus CH4 conversion of the oxidative coupling of methane obtained by high-throughput experiments. The color represents the temperature. (b) Importance analysis by random forest classification. (c) Average density, electronegativity, electron affinity, and first and second ionization energies of predicted high C2 selectivity catalysts. In each plot, a horizontal line represents the descriptor average over all data for comparison. ion-e_1st = first ionization energy; ion-e_2nd = second ionization energy; e-aff-ev = electron affinity; Pauling_e-neg = Pauling electronegativity; Group = periodic table group. Reproduced with permission [88].
Figure 6. ML and physical quantities derived from the periodic table explore the descriptors of product selectivity in the oxidative coupling of methane (OCM) reaction. (a) C2 selectivity versus CH4 conversion of the oxidative coupling of methane obtained by high-throughput experiments. The color represents the temperature. (b) Importance analysis by random forest classification. (c) Average density, electronegativity, electron affinity, and first and second ionization energies of predicted high C2 selectivity catalysts. In each plot, a horizontal line represents the descriptor average over all data for comparison. ion-e_1st = first ionization energy; ion-e_2nd = second ionization energy; e-aff-ev = electron affinity; Pauling_e-neg = Pauling electronegativity; Group = periodic table group. Reproduced with permission [88].
Catalysts 15 00282 g006
Catalysts 15 00282 g007
Figure 7. Based on MLaMD, the cleavage dynamics enhancement simulation diagram of the ethylene glycol C-H bond on Cu(111) and Pd(111). (a) Illustrate an atomic-level computational model of ethylene glycol adsorption on Cu(111), with a clear representation of the reagent; ethylene glycol and the surface are represented by spheres; whereas, water molecules are represented by lines. The two collective variables, d(C–H) and d(C–Cu), are marked. Color code for atoms: brown—Cu, grey—C, red—O, white—H. (b) 2D free energy surface for the C–H bond breaking of ethylene glycol over Cu(111) and Pd(111). Red dotted circles mark transition states and red text denotes their energies. Solid black circles mark minima [91].
Figure 7. Based on MLaMD, the cleavage dynamics enhancement simulation diagram of the ethylene glycol C-H bond on Cu(111) and Pd(111). (a) Illustrate an atomic-level computational model of ethylene glycol adsorption on Cu(111), with a clear representation of the reagent; ethylene glycol and the surface are represented by spheres; whereas, water molecules are represented by lines. The two collective variables, d(C–H) and d(C–Cu), are marked. Color code for atoms: brown—Cu, grey—C, red—O, white—H. (b) 2D free energy surface for the C–H bond breaking of ethylene glycol over Cu(111) and Pd(111). Red dotted circles mark transition states and red text denotes their energies. Solid black circles mark minima [91].
Catalysts 15 00282 g007
Catalysts 15 00282 g008
Figure 8. The optimization framework of Gaussian process regression (GPR) was applied to study the reaction of syngas (CO and H2) on Rh (111) catalyst under experimental operating conditions (573 K and 1 atm). (A) Comprehensive reaction network encompasses syngas transformation into multiple products, including CO2, H2O, methanol (CH3OH), acetaldehyde (CH3CHO), methane (CH4), and ethanol (CH3CH2OH). Surface intermediates are restricted to species containing ≤2 carbon and ≤2 oxygen atoms. (B) Machine learning-driven reduced reaction network for syngas reactivity identifies acetaldehyde and CO2 as dominant products, consistent with experimental validation on Rh(111). This simplification is achieved through the iterative GPR-guided pruning of kinetically irrelevant pathways, prioritizing thermodynamically favorable and experimentally observed routes. Oxygen atom = red sphere; rhodium atom = green sphere; carbon atom = grey sphere; hydrogen atom = white sphere [96].
Figure 8. The optimization framework of Gaussian process regression (GPR) was applied to study the reaction of syngas (CO and H2) on Rh (111) catalyst under experimental operating conditions (573 K and 1 atm). (A) Comprehensive reaction network encompasses syngas transformation into multiple products, including CO2, H2O, methanol (CH3OH), acetaldehyde (CH3CHO), methane (CH4), and ethanol (CH3CH2OH). Surface intermediates are restricted to species containing ≤2 carbon and ≤2 oxygen atoms. (B) Machine learning-driven reduced reaction network for syngas reactivity identifies acetaldehyde and CO2 as dominant products, consistent with experimental validation on Rh(111). This simplification is achieved through the iterative GPR-guided pruning of kinetically irrelevant pathways, prioritizing thermodynamically favorable and experimentally observed routes. Oxygen atom = red sphere; rhodium atom = green sphere; carbon atom = grey sphere; hydrogen atom = white sphere [96].
Catalysts 15 00282 g008
Catalysts 15 00282 g009
Figure 9. Challenges and opportunities of applying machine learning to advanced oxidation process catalysts.
Figure 9. Challenges and opportunities of applying machine learning to advanced oxidation process catalysts.
Catalysts 15 00282 g009
Table 1. Common advanced oxidation processes and corresponding reactive oxygen species in water treatment.
Table 1. Common advanced oxidation processes and corresponding reactive oxygen species in water treatment.
AOP TypesROS TypesOther Occurring MechanismsReference
Fenton reaction•OHIron coagulation
Iron sludge-induced adsorption		[31,32,33]
Photo-Fenton reaction•OHIron coagulation
Iron sludge-induced adsorption
UV photolysis		[34,35]
O3•OHDirect O3 oxidation[36,37]
O3/H2O2•OHDirect O3 oxidation
H2O2oxidation		[38,39,40]
O3/UV•OHUV photolysis[39,40]
UV/H2O2•OHUV photolysis
H2O2oxidation		[41,42]
UV/TiO2•OHUV photolysis[43,44]
Ultrasonic irradiation•OHAcoustic cavitation generates transient high temperatures (≥5000 K) and pressures (≥1000 atm) and produce H+ and HO2−, besides OH−[45]
Heat/persulfateSO4−Persulfate oxidation[46,47]
UV/persulfateSO4−Persulfate oxidation
UV photolysis		[48]
Fe(II)/persulfateSO4−Persulfate oxidation
Iron coagulation
Iron sludge-induced adsorption		[49,50]
OH−/persulfateSO4−/•OHPersulfate oxidation
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Yuan, Q.; Wang, X.; Xu, D.; Liu, H.; Zhang, H.; Yu, Q.; Bi, Y.; Li, L. Machine Learning-Assisted Catalysts for Advanced Oxidation Processes: Progress, Challenges, and Prospects. Catalysts 2025, 15, 282. https://doi.org/10.3390/catal15030282

AMA Style

Yuan Q, Wang X, Xu D, Liu H, Zhang H, Yu Q, Bi Y, Li L. Machine Learning-Assisted Catalysts for Advanced Oxidation Processes: Progress, Challenges, and Prospects. Catalysts. 2025; 15(3):282. https://doi.org/10.3390/catal15030282

Chicago/Turabian Style

Yuan, Qinghui, Xiaobei Wang, Dongdong Xu, Hongyan Liu, Hanwen Zhang, Qian Yu, Yanliang Bi, and Lixin Li. 2025. "Machine Learning-Assisted Catalysts for Advanced Oxidation Processes: Progress, Challenges, and Prospects" Catalysts 15, no. 3: 282. https://doi.org/10.3390/catal15030282

APA Style

Yuan, Q., Wang, X., Xu, D., Liu, H., Zhang, H., Yu, Q., Bi, Y., & Li, L. (2025). Machine Learning-Assisted Catalysts for Advanced Oxidation Processes: Progress, Challenges, and Prospects. Catalysts, 15(3), 282. https://doi.org/10.3390/catal15030282

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop