Machine Learning for Advanced Emission Monitoring and Reduction Strategies in Fossil Fuel Power Plants

Abstract

Fossil fuel power plants are a significant contributor to global carbon dioxide (CO₂) and nitrogen oxide (NO_x) emissions. Accurate monitoring and effective reduction of these emissions are crucial for mitigating climate change. This systematic review examines the current state of research on the application of machine learning techniques in evaluating the emissions from fossil fuel power plants. This review first briefly introduces the continuous emission monitoring (CEM) systems and predictive emission monitoring (PEM) systems that are commonly used in power plants and highlights that machine learning models can significantly improve PEM systems through their capability to process and interpret large datasets intelligently to transform traditional emission monitoring systems by enhancing their precision, effectiveness, and cost-efficiency. Compared to previously published review articles, the key contribution and innovation in this present review is the discussion of machine learning models in CO₂/NO_x emissions according to the different algorithms used, including their advantages and disadvantages in a systematic way, which aims to help future researchers to develop more effective machine learning models. The most popular machine learning model includes reinforcement learning, a forward neural network, a long short-term memory neural network, and support vector regression. While each model method has its own advantages and disadvantages, we noted that training data quality, as well as the proper selection of model parameters, plays an important role. The challenges and research gaps, such as model transferability, a deep understanding of the physics of CO₂/NO_x emissions, and the availability of high-quality data for training machine learning models, are identified, and recommendations as well as potential future research directions to address these challenges are proposed and discussed.

1. Introduction

The urgency of tackling climate change through sustainable practices has never been more critical, as evidenced by the rising global focus on reducing greenhouse gas emissions. The increasing concentration of greenhouse gases has become the key reason for global warming and climate change, which consequentially affect the development of our society in the 21st century [1]. Greenhouse gases refer to gases in the Earth’s atmosphere that trap heat to increase the temperature of the Earth. Greenhouse gases include many different types of gases, mainly carbon dioxide (CO₂), methane (CH₄), and fluorinated gases. Out of all types of greenhouse gases, carbon dioxide (CO₂) accounts for approximately 80% of all greenhouse gases [2]. In addition, nitrogen oxides (NO_x) act as major indirect greenhouse gases. Both CO₂ and NO_x are mostly emitted from fossil fuel power plants, which are the major generators of electricity worldwide. The generation of CO₂ is majorly from the combustion of fossil fuels, while NO_x is generated from the air’s reaction with oxygen (O₂) and is normally formed under high temperatures in the combustion process. There are some common ways to help reduce the emission of NO_x, including gas recirculation, reduced air preheating, and a reduced firing rate [3]. The reduction of anthropogenic CO₂/NO_x emissions can effectively reduce the trend of climate change, as evidenced by the abundance of reviews in the literature [4,5,6]. In 2021, global CO₂ emissions increased by 6% to the highest-ever level of 36.3 billion tons [7]. Based on the most recent data from the U.S. Energy Information Administration, approximately 60% of the electricity generation was from fossil fuels—coal, natural gas, petroleum, and other gases [8]. Furthermore, the emission of greenhouse gases, such as CO₂, from developing countries is rapidly increasing along with their economic growth [9,10,11]. It is estimated that approximately 28% of global total energy consumption will be from fossil fuels by 2040 [12]. Although renewable energy is gaining momentum as an alternate energy source, it is widely reported that fossil fuels are still primarily used to meet the ever-growing energy consumption requirement in the world [9,13,14]. Aside from CO₂ emissions, NO_x, which is mostly emitted from the combustion process in fossil fuel power plants, also contributes to global climate change. In addition, it has numerous hazardous effects on human beings, vegetation, and animals [15,16,17]. The emission of CO₂ mostly depends on the amount of fossil fuels consumed, while NO_x is commonly generated under high temperatures and pressures [18]. Furthermore, the consumption of fossil fuels produces not only greenhouse gases but also other pollutants that are harmful to human health and the environment [19]. Thus, given the considerable scale of CO₂/NO_x emissions from power plants, reducing the carbon footprint is crucial for mitigating the impacts of climate change and transitioning towards a more sustainable energy future for our society [20]. To achieve the goal of reducing CO₂ and NO_x emissions, one primary step is to accurately and effectively monitor CO₂ and NO_x through the accurate and reliable monitoring of carbon emissions, especially regarding the processes of fossil fuel power plants, as this is essential for ensuring compliance with environmental regulations, reporting emissions to relevant authorities, and informing emission reduction strategies [14].

There are two major ways to monitor CO₂ emissions from fossil fuel plants: continuous emission monitoring (CEM) and predictive emission monitoring (PEM) [21]. Continuous Emission Monitoring is one of the most widely used methods for monitoring carbon emissions in fossil fuel power plants. CEM is designed to continuously measure the concentration of pollutants in the flue gas emitted from power plant stacks. Such a system comprises all the essential equipment needed to measure the concentration of gas or particulate matter, as well as the emission rate, by utilizing pollutant analyzer readings and a conversion formula, chart, or software to generate outcomes in the relevant emission standard units [22]. CEM provides real-time, accurate measurements of CO₂ emissions, enabling power plant operators to demonstrate compliance with environmental regulations and report emissions to the relevant authorities. As mentioned before, the complex nature of CEM data, which include large volumes of high-frequency measurements, can pose challenges for data management, analysis, and interpretation. Predictive Emission Monitoring is an alternative approach to monitoring carbon emissions in fossil fuel power plants. Unlike CEM, which directly measures pollutant concentrations, PEM uses process parameters and emission factors to estimate emissions. These systems rely on mathematical models that correlate process variables, such as fuel consumption, boiler temperature, and flue gas flow rate, with emission levels. The input data of the control system, historical data, and process parameters are fed in to connect with the computation algorithm to obtain the prediction of CO₂ and NO_x emissions. Thus, the accuracy of PEM largely depends on how well the built-in mathematical methods consider and model the above process variables [23,24].

PEM offers several advantages over CEM, including lower installation and maintenance costs and greater flexibility in monitoring multiple emission sources [23]. The installation, operation, and maintenance of CEM require expensive hardware, regular calibration, and skilled personnel to ensure accurate and consistent measurements. It is estimated that the capital cost of PEM can be 50% lower than CEM [25,26]. In addition, its operational cost is estimated to be 10–20% lower than CEM [27]. However, the accuracy of PEM is highly dependent on the quality of the underlying models and the availability of reliable process data. Inaccurate or incomplete data can lead to significant discrepancies between estimated and actual emissions, potentially compromising compliance with environmental regulations. In addition, the complexity of power plant process operations and the variability of fuel quality and combustion conditions make it difficult to analyze and interpret the vast amounts of data generated by CEM [23]. Furthermore, PEM may not provide the same level of temporal resolution as CEM, which can limit the ability to detect short-term emission spikes or anomalies [26]. From what we have reviewed above, it can be shown that PEM is more robust and cost-effective compared to CEM. However, traditional PEM relies largely on process parameters for accurate and effective model development. However, these models may not capture the full complexity of the emission formation process, leading to potential discrepancies between estimated and actual emissions [23].

The emergence of machine learning can address the above challenges for PEM by providing more reliable process parameters and advanced model algorithms for data processing and interpretation, which significantly help improve the accuracy and effectiveness of PEM. Machine learning is an area of artificial intelligence (AI) and computer science concerned with the processing of data and algorithms that allow AI to mimic the way humans learn, with the goal of gradually increasing its accuracy [28,29]. The nature of ML’s capability to process and interpret large datasets intelligently enables it to be a promising technology to transform emission monitoring systems by enhancing their precision, effectiveness, and cost-efficiency [30,31]. Figure 1 shows the general flow process of machine learning model development, where the feedback loops across the model development phases are key enablers for monitoring the capability and reliability of the developed machine learning model. “Data preparation” is the first step for developing a typical machine learning model, and the data can be historical data generated from power plants in long-term operation. “Data processing” cleans and filters out any deficient data by using different machine learning algorithms, which will be detailed in later sections. After parameter tuning, the machine learning model is trained and tested for validation. This process may need to be repeated until a desired level of model accuracy is achieved.

The utilization of AI technologies should help to improve the efficiency of fossil fuel power plants for emission control by overcoming the limitations of traditional methods and providing power plant operators with more accurate, reliable, and actionable emission data. By leveraging AI-based approaches, power plants can optimize their emission control strategies, reduce costs associated with monitoring and compliance, and ultimately contribute to the global effort to combat climate change [33]. The purpose of this paper is to systematically review the current research of utilizing machine learning technology in enhancing carbon emission monitoring systems in fossil fuel power plants based on the different categories of machine learning methods (e.g., reinforcement learning, artificial neural networks, support vector regression, etc.). The features of each machine learning method are summarized and their advantages and disadvantages are compared. After reviewing the status of current research, challenges and gaps are identified for future research directions. This paper aims to offer insightful perspectives for researchers, power plant operators, and policymakers to empower them to make well-informed choices regarding the integration and advancement of machine learning technologies in monitoring CO₂ and NO_x emissions from fossil fuel power plants. In this paper, the primary focus is understanding the role of machine learning applications in the prediction and evaluation of CO₂ and NO_x emissions, which forms the basis for further reducing CO₂ and NO_x emissions. When machine learning capabilities are well-understood and validated in CO₂ and NO_x emissions, the possibility of optimizing related control processes in fossil fuel power plants will be possible to further help achieve the goal of reducing greenhouse gases and pollutant emissions eventually.

2. Materials and Methods

2.1. Literature Search Strategy

To ensure methodological rigor, a systematic review of studies concerning the application of machine learning to the evaluation and mitigation of CO₂/NO_x emissions in fossil fuel power plants was conducted following the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) protocol guidelines (Supplementary Materials) [34]. To achieve a comprehensive review, a multi-database search strategy was implemented, leveraging scholarly databases including Scopus, Web of Science, and Google Scholar (see Figure 2). The search process employed a comprehensive series of keywords and phrases relevant to machine learning, CO₂/NO_x emissions, and fossil fuel power plants. We strategically utilized Boolean operators (AND, OR) to refine and expand the search parameters, ensuring thorough coverage of the literature. We tested numerous combinations to ensure a broad and inclusive search. For example, the search terms included “machine learning AND CO₂ emissions”, “NO_x reduction OR fossil fuels”, and “power plants AND emission control”.

The period designated for the literature review extended from 2008 to 2023. This interval was chosen to encapsulate the progressive discourse surrounding the application of machine learning to assess and reduce CO₂/NO_x emissions in fossil fuel power plants over the preceding 15 years. This timeframe is notably significant, as it includes the increasing global focus on sustainability, marked by the transition from the United Nations Millennium Development Goals to the Sustainable Development Goals, which have been in effect since 2015 (https://sdgs.un.org/publications/getting-started-sdgs-18003, accessed on 1 January 2024). A few important but older papers were also reviewed by the authors.

2.2. Inclusion and Exclusion Criteria

To ensure the relevance and quality of the sources, specific inclusion and exclusion criteria were established (Table 1).

Inclusion Criteria:

• –. Published between 2008 and 2023.

• –. Articles, textbooks, and case studies specifically addressing the integration of machine learning with CO₂/NO_x emissions management in fossil fuel power plants.

• –. Works offering empirical data, theoretical frameworks, or comprehensive case studies that explore methodologies, challenges, and outcomes pertinent to sustainability efforts within the energy sector.

• –. Sources must be available in English to ensure the review is accessible to an international audience.

Exclusion Criteria:

• –. Publications outside the 2008–2023 timeframe.

• –. Studies that do not focus on the nexus of machine learning, CO₂/NO_x emissions, and fossil fuel power plants, thereby ensuring a targeted review scope (examples of excluded topics might include unrelated applications of machine learning or general environmental studies).

• –. Sources that are not peer-reviewed, such as blogs, opinion pieces, and non-academic publications, are omitted to maintain the academic rigor of this review.

3. Results and Discussion

Figure 3 shows the keyword network diagram of the retrieved papers, which reveals the subtopics related to this study. The bigger the dot, the more frequently the keyword appeared in publications from the literature search. We further noticed that connections are focused on “machine learning” and “plant”. Connections that centered around machine learning models focused on the development of different methods (e.g., the extreme learning method, the support vector method, neural networks, etc.) to predict and optimize the monitoring of CO₂/NO_x emissions. The key facilities of power plants, such as the boiler, are of particular attention to researchers as their design and process are directly related to CO₂/NO_x emission levels from power plants.

The following section presents the discussion of machine learning model applications for evaluating and potentially reducing CO₂/NO_x emissions from power plants based on different types of machine learning techniques. The presented machine learning techniques mostly include reinforcement learning, neural networks, support vector regression, and the extreme learning method. The characteristics, as well as the advantages and disadvantages, of each machine learning technique are also discussed.

3.1. Reinforcement Learning

Reinforcement learning (RL) is a sub-branch of machine learning that allows computer programs to make decisions by varying the experimental settings and learning from mistakes to achieve the most optimal results in a “trial and error” process by maximizing the reward function. RL has witnessed success in complex systems, such as power plants that include multiple interfaces of process control in their operation. The process of RL-based decision-making allows researchers to predict the behavioral characteristics of power plants and better formulate control processes to reduce emissions.

Researchers have observed that RL has been used to evaluate the carbon emissions of power plants in recent years. The authors of Ref. [35] developed a model-based offline reinforcement learning (RL) framework, which utilizes historical operational data and estimated an improvement of a mere 0.5% in the combustion efficiency of 600 megawatts (MW), which is validated against experiment results. The authors of [36] designed a reinforcement learning framework combining long short-term memory convolutional networks and deep Q-networks to achieve a balance between emission control and boiler efficiency in the field of combustion optimization. The reinforcement learning optimizer contributes to the online development of policies based on static and dynamics stages, which results in a 5% improvement in combustion efficiency and a 15% reduction in NOx emissions compared to experimental data.

The RL method can be very well-designed for handling non-deterministic environments, such as control process conditions for operations (e.g., optimizing the combustion process for carbon emissions in power plants for real-time decision-making) under uncertainties. This non-deterministic environment may include errors in measurement and communication, as well as a lack of complete knowledge of the measured quantity. However, when applying the RL method, one needs to pay close attention to the quality of the reward function, which is used to set goals for the learned skill and direct the agent toward the best policy (e.g., scaler feedback to show how well the agent is doing at a particular step). We found that few of the above-discussed studies thoroughly showed how the reward function is optimized and designed.

3.2. Neural Network

A Neural Network (NN) is one of the most widely used machine learning methods for carbon emission evaluation in industry. A NN is developed by connecting nodes of layered structures that resemble human brains. By selecting process parameters of interest for carbon emissions, researchers apply layers of nodes that enable adaptive learning processes similar to human brains [33].

A forward neural network (FNN), as the basic type of NN, is observed to be used by researchers to develop carbon and NOx emissions. The basic structure of the “input to output” structure allows one to apply appropriate process parameters in a power plant for PEM. The authors of Ref. [37] illustrated the potential of utilizing an artificial neural network in forecasting NO_x emissions from coal-fired combustion boilers. The highlight of this research is that the authors gave good thoughts and explanations on how to select parameters of interest for building up the model. Initially, they relied on expert and previous knowledge to select parameters. Later, it was found that using fewer parameters produced better results with higher accuracy. Thus, the parameters to be considered can be significant for building a high-confidence machine learning model for the prediction of combustion efficiency and NO_x emissions. Similarly, Ref. [38] used a forward neural network (FNN) to evaluate NO_x emissions based on a 210 MW fossil fuel power plant combustion boiler and found that the oxygen concentration in flue gas, fossil fuel properties, fossil fuel flow, boiler load, the air distribution scheme, the flue gas outlet temperature, and nozzle tilt may need to be considered as important parameters for developing a machine learning model. The authors of Ref. [39] used artificial neural networks (ANNs) to develop a machine learning model with a worldwide assembly database to predict general air pollutants, which shows great model reliability with less than 1.0% difference between predicted and estimated values.

The reliability of the NN method depends largely on the quality of the input process data. The authors of Ref. [40] used real operational data from commercial power plants to develop an NN algorithm to estimate NO_x emissions. When considering additional parameters aside from the real operational data, the authors also considered additional factors, such as the impact of dynamic coal and limestone properties, for a more comprehensive prediction of NO_x emissions. The authors found that the model’s prediction accuracy was improved by 42%.

As discussed before, the layers of the node structure are the cornerstone of the NN method. Optimized selection of process parameters for carbon emission processing is very important. Ref. [23] developed a neural network-based model using eight process parameters as inputs to predict NO_x emissions by meeting the regulatory requirements for the precision of regulators. In their study, the optimal number of hidden layers was determined, and it was found that the Nadam method, which is basically the combined Nesterov trick and the Adam optimization method, for optimization yields the best model performance. The same research team developed a predictive emissions model using a gradient-boosting machine learning method. It is found that the machine learning model created by XGBoost has a Root-mean-square deviation (RMSE) that is lower than the model created by a feedforward neural network. XGBoost is a gradient-boosting framework and frequently outperforms deep learning when working with tabular data, while a neural network is better suited for unstructured data. Botros created a FNN for three gas turbine engines at natural gas compressor stations in Canada, spanning from 2008 to 2010 [41,42]. These models utilized either four or seven process parameters as inputs, featured a single hidden layer with two units, and had one output dedicated to NO_x emissions. The models employed the Sigmoid activation function. The primary objective of these predictive models was to assess and contrast the predicted values of NO_x emission factors established by the US EPA.

Long Short-Term Memory Networks represent a particular kind of Recurrent Neural Network designed to tackle the issue of the vanishing gradient problem that plagues conventional Recurrent Neural Networks. The authors of [43] developed a long short-term memory (LSTM) neural network to improve and optimize the NOx emissions of a fossil fuel power plant based on a “t-fired coal boiler” running under a steady state. The authors claimed to achieve an improvement of approximately 5%. However, the researchers did not sufficiently discuss how the overfitting or hypermeter tuning issues, which are commonly encountered for LSTMs, are addressed. LSTMs may not be an ideal machine learning model where robust and time-efficient models are required (e.g., for real-time evaluation of carbon emissions in power plants), as they require long training times and are computationally complex.

3.3. Support Vector Method

The Support Vector Method (SVM) is a method of machine learning applied to tasks involving predicting continuous values, which makes this especially suitable for predicting the real-time evaluation of CO₂/NO_x emissions in fossil fuel power plants. The authors of Ref. [44] developed an SVM-based machine learning model to predict gas emissions from power plants. The authors compared the results of CEM. The results show that the SVM-based model has the capability for online prediction, with an average prediction accuracy of 95%.

By utilizing week-long CEM data, a novel ensemble model based on the SVM is introduced for the prediction of NO_x emissions [45]. It needs to be pointed out that the collected data for building up the model are limited. This can arise from the limitations of time, operating conditions, and financial resources [46]. The authors of Ref. [47] proposed an SVM to study the correlation between NO_x emissions and the operation parameters of a 300 MW fossil fuel power boiler with an error of 1.58%. The experiment validated the model’s accuracy and further showed that NO_x emissions could be reduced by approximately 18% through process optimization in the fossil fuel plant operations used in the machine learning model.

As discussed above, the SVM is memory-efficient, which means that it has great potential in the development of real-time CO₂/NO_x emissions evaluation. However, one needs to note that the SVM is not suitable for large datasets and datasets that contain high levels of noise; this requires a power plant to provide high-quality data for developing the SVM model.

3.4. Extreme Learning Method

Compared with the traditional NN method, the Extreme Learning Method (ELM) uses a single-layer forward NN method, which allows the ELM to avoid training the input layer, only training the output for improved training velocity while maintaining the general performance of the developed model.

Researchers utilize the advantage of the fast training velocity of the ELM, which bypasses iterative optimization methods such as gradient descent, for developing CO₂/NO_x emissions from power plants with high computational efficiency. In the study of [48], the authors used the ELM to model and optimize NO_x emissions with real data from CEM in a coal-fired power plant. The new proposed ELM demonstrated good generalization compared to the widely used NN and SVM by examining the relationship between the NO_x emissions and the operational parameters in the model. It was found that the MRE from the ELM was 1.13%, which was better than the NN and the SVM. This is an exemplary study that shows that the ELM-based machine learning models can not only predict NO_x emissions but also have the potential to serve as an alternative and effective approach to optimize the operational parameters of the fossil fuel power plant (e.g., gas flowrates, temperature, and pressure, out of a total of six selected parameters) to reduce NO_x emissions. Using similar techniques, Ref. [49] constructed a bootstrap aggregated ELM model comprising multiple ELM models, which were created through bootstrap re-sampling replication of the original training data to forecast the CO₂ capture rate and level. By utilizing the mean square error of the prediction results from various hidden neurons as a comparison metric, the bootstrap aggregated ELM model outperformed a single ELM model in delivering more precise and dependable predictions along with model prediction confidence intervals. The authors of Ref. [50] further improved the controlling and monitoring system of the combustion furnace by using ELM and NN algorithms in machine learning. It is revealed that the ELM outperformed the NN as it required less training time and exhibited lower average errors.

The authors of Ref. [48] introduced a NO_x prediction model utilizing an ELM algorithm for coal-fired boilers at a power station. This model was subsequently employed to enhance operations for the reduction of NO_x emissions. The ELM algorithm is characterized as a single hidden layer feedforward neural network. The development of the model was based on ten days of operational data. The NO_x emissions forecasted by the model exhibited a good match with the actual NO_x values in terms of mean absolute error (MAE) and had an R-value of 98%. Furthermore, Ref. [51] compared the model’s accuracy with different machine learning algorithms, and it was found that the ELM has superior performance in terms of prediction accuracy and computational time.

Although the ELM is very popular among researchers for developing a computationally efficient machine learning model for CO₂/NO_x emissions, as we discussed above, one needs to note that attention needs to be paid to the overfitting issue inherited by the ELM. In addition, the ELM is characterized by its opaque nature, which presents challenges in areas where clarity and understanding are essential, which is especially important for the carbon emission evaluation process.

Table 2 shows a summary of the applications of machine learning methods in CO₂/NO_x emission evaluation. Forward neural networks and Extreme learning methods seem to be better when addressing the real-time prediction of CO₂/NO_x emissions due to their computational efficiency.

3.5. Machine Learning Coupled with Computational Fluid Dynamics (CFD)

In power plants, the documentation of historical data may be sparsely available; thus, using validated CFD models and generating large datasets for training machine learning models are also used by researchers for model development. The CFD method, commonly performed by using the commercial software ANSYS 2024 or the free software OPENFOAM v11, is capable of generating training data to be used by machine learning models through the development of a physics model. For example, to address the problem of limited understanding of combustion slagging in boiler combustion in fossil fuel power plants, Ref. [52] developed a neural network-based machine learning model by using real data from CEM. Furthermore, the authors proposed a non-dominated sorting genetic algorithm (NSGAII) for multi-objective optimization by combining CFD to improve the efficiency of fossil fuel-based boiler furnace combustion. The authors of Ref. [46] utilized a self-designed deep belief network (DBN) to forecast NO_x emissions from coal-fired boilers. Their DBN-based models demonstrated superior prediction accuracy and increased reliability compared to alternative CO₂/NO_x emission prediction models by achieving an accuracy of approximately 0.9. The authors mentioned that their input data for training were validated against CFD for data quality examination. Although CFD can be used as an alternative for gathering historical data from power plants for model development, generating CFD data and validating results may take additional computational time.

3.6. Comparison of Different Machine Learning Methods

There have been quite a few studies in the open literature that compare different machine learning methods (e.g., NN, SVM, ELM, etc.) and their applications in predicting CO₂/NO_x emissions in fossil fuel power plants. The authors of Ref. [53] successfully developed a model to analyze the correlation between operational parameters and NO_x emissions using machine learning techniques. They specifically applied this model to a 660 WM boiler. The input data underwent Principal Component Analysis (PCA), which is a dimensionality reduction technique that is frequently used to make large datasets smaller, and were processed before being utilized in three different models: LSTM, RNN, and SVM. The comparison of the three NO_x emission models revealed that the LSTM model outperformed the others in addressing long-term memory issues.

The authors of Ref. [54] introduced an online combustion optimization system for the purpose of reducing NO_x emissions in a 490 MW tangential coal-fired power plant. The effectiveness of this system was confirmed through the utilization of various models such as support vector regression, random forests, and kernel partial least squares. These models were employed to validate the operational sub-regions of the plant’s neural network model, which could be fed back to improve the efficiency of CEM for processing real-time data. The authors also demonstrated that a 22.5% reduction in NO_x could be achieved.

The authors of Ref. [55] examined various techniques to predict NO_x emissions from ten large-scale gas turbines. The methods evaluated comprised linear regression, kernel ridge regression, support vector regression, and a neural network with diverse optimization strategies. The models were trained on 1000 samples and validated on another 1000 samples. Among the methods tested, the neural network and support vector regression demonstrated superior accuracy in predicting NO_x levels.

The authors of Ref. [56] used different machine learning algorithms to predict CO₂ emissions from power plants in Kuwait using data from the Environmental Protection Agency (EPA) of the United States and the Intergovernmental Panel on Climate Change (IPCC). It is found that the ELM algorithm provides the best prediction accuracy by using the data for five consecutive years. ANN and SVM models also provide reasonable predictions of CO₂ emissions but are less accurate than ELM models.

The authors of Ref. [57] focused on developing a machine learning-based model to optimize NO_x emissions from an in-service fossil fuel power plant. Aside from historical data, the authors also used Computational Fluid Dynamics to obtain the NO_x generation data. They found that the Gradient Boost Regression Tree (GBRT) model had higher accuracy than the ANN and SVM algorithms.

In the process of developing a reliable machine learning-based model for predicting CO₇/NOx emissions, one needs to understand which machine learning algorithm is better and which one is especially suited for the application of power plant data, which is sometimes difficult to evaluate concerning its sensor collecting complexity and uniqueness. The authors of Ref. [58] found that the NN algorithm is more prone to overfitting if training data is insufficient, which is likely to be a common case for power plant data collected by CEM. The authors of Ref. [59] pointed out that the SVM algorithm may suffer from large-dimension input vectors and datasets, increasing the computational cost exponentially. The SVM can overcome the challenge of the NN to give improved solutions for highly nonlinear problems, as in our case for NO_x emissions [60,61,62].

The authors of Ref. [63] modeled and optimized the NO_x emissions of a fossil fuel power plant using three different types of machine learning algorithms, including a recurrent neural network (RNN), long short-term memory (LSTM), and a gate recurrent unit (GRU) with input parameters including boiler load, excess air ratio, flue gas temperature, fuel flow rate, and the operational settings of the combustion equipment. Using the RNN algorithm exhibits the highest prediction accuracy, while experimental results reveal that the GRU-based NO_x prediction model delivers the most accurate predictions among the models proposed. By using their developed model, a reduction of 17–19% in NO_x emissions is eventually achieved.

From the above research, it is not conclusive which machine learning method is most optimal for the evaluation of CO₂/NO_x emissions. The fact that each predictive model is specifically designed for a power plant means that the process conditions, facility design, running capacity, and so forth can be very different from one to another. This gives rise to an additional challenge in machine learning model development, and more focus should be given to developing an industry-accepted standardized model that could be applied to general fossil fuel power plants in the future.

3.7. Challenges and Limitations of the Current Machine Learning Model in CO₂/NO_x Emission Evaluation for Fossil Fuel Power Plants

Although there has been considerable research on developing predictive machine learning models for CO₂/NO_x emissions over recent years due to the rapid development of AI technology, there are still challenges and gaps in the research. The following section summarizes the key challenges and research gaps that are recommended to be addressed in the future:

• Addressing the challenge of knowledge transferability across machine learning models across different designs and processes of fossil fuel power plants. The facilities and equipment of power plants (e.g., combustion chambers, boilers, data collecting systems) are very individually based for power plants, which raises a concern about whether the learnings from these models can be transferred to power plants with more generic conditions when predicting CO₂/NO_x emissions. The production capacity of fossil fuel power plants varies in size [47,53,54] and thus may have different thermal units and systems, which adds difficulty to the knowledge transfer process in model standardization. It is highly possible that machine learning models developed using data from one power plant may not perform well when applied to another plant with different characteristics, such as fuel type, boiler design, or emission control technologies. Furthermore, the development of more advanced machine learning models for better transferability may be an additional challenge for developing countries since developing countries may lack the capital and resources to build up the infrastructure for the data center, which is critical for processing data in a machine learning model.

• Evaluation of error of the machine learning model: The error from the machine learning model can be from several aspects. One needs to evaluate which machine learning algorithm is more applicable to the power plant, which can be dependent on the specific individual features of the power plant. An inappropriate selection of process parameters may result in model inaccuracy as these inappropriate process parameters may not accurately represent the power plant emission system. The improvement and availability of good quality training data can help to reduce model error. Thus, one needs to systematically evaluate the overall machine learning model strategy to reduce model error in general.

• Develop a machine learning model that can predict transient or abnormal conditions in the operation run of the power plant. Almost all the models discussed above [28,35,36,37,38,39,40] used data under normal or steady-state working conditions. However, the operation of the fossil fuel power plant may undergo equipment startup, shutdown, or interruption. These conditions may be of short duration; however, they can involve more complicated parameters that affect CO₂/NO_x emissions and even the subsequent steady-state prediction. Researchers may need to pay more attention to these interruption conditions for prediction model capability development.

• A more fundamental understanding of the physics that influences CO₂/NO_x emissions in the processes of power plants is needed. A power plant is a complicated thermal system and there are numerous factors that may affect the final emissions, so a more fundamental knowledge of the physics helps machine learning model developers to select more relevant input parameters and effective algorithms. For example, the main measures to control the generated NO_x emissions are determined by the combustion process, as seen in [64], in which additional Amine reclaimer waste (ARW) is suggested to be added to combustion to reduce the generation of NO_x. However, a combustion boiler is difficult to study as it involves many components, complicated physics, and safety constraints. The chemical mechanism of combustion is not fully clear, mostly due to the complex physics of the oxidation process of carbon monoxide. There needs to be a greater effort in researching the comprehensive combustion process and related parameters systematically. The availability of high-quality training data for machine learning models. Machine learning models require large amounts of labeled data to learn the underlying patterns and relationships in emission-related variables. However, the collection and annotation of such data can be time-consuming, expensive, and prone to errors [65]. The quality of historical data from power plants may not be well-recorded and usually cannot be validated or assured. As historical data from fossil fuel power plants may not be readily available for model developers, an alternative method is to generate large datasets by using the CFD method. However, CFD-generated data does not solve the issue of data validation, and model prediction may not fully replicate the combustion process of the power plant. At the same time, CFD computation may require considerable computational time [52,66].

• The interpretability and transparency of machine learning models. The machine learning models often operate as “black boxes”, making it difficult for users to understand how the predictions are generated and to trust the results [67]. This lack of interpretability can hinder the adoption of machine learning-based emission monitoring systems by fossil fuel power plant operators and regulators who require a clear understanding of the underlying decision-making processes. As a potential solution, explainable AI (XAI) techniques, such as rule extraction and feature importance analysis, can be employed to improve the interpretability of AI models [68,69]. However, further research is needed to develop and validate XAI methods specifically tailored to the context of carbon emission monitoring in power plants.

• Encouraging the establishment of clear guidelines and standards for sharing data on CO₂/NO_x emissions for future machine learning model development. The development of machine learning models requires large amounts of labeled data to learn the underlying patterns and relationships in emission-related variables [64]. Better utilization of sharing data in monitoring fossil fuel power plants also depends on regulators establishing clear guidelines and standards for machine learning-based model validation, as well as encouraging model data sharing to help address the issue of ambiguity of transferring models for understanding CO₂/NO_x emissions beyond the limitation of design-specific fossil fuel power plants. The encouragement of data sharing can also improve the sustainability of the machine learning model in the long run. Although researchers have validated model accuracy using available data, there is almost no long-term monitoring of model capability after publication. Conducting more long-term tracking and using the developed model for CO₂/NO_x emission control helps to further model validation and evaluate and confirm machine learning-based models with regard to their long-term costs, benefits, and environmental impacts.

Addressing these research gaps will contribute to the development of more robust, reliable, and sustainable machine learning-based solutions for CO₂/NO_x emission evaluation and reduction in fossil fuel power plants, ultimately supporting the global efforts to mitigate climate change and transition towards a future of low-carbon energy.

4. Conclusions

In this review paper, we have provided a comprehensive overview of the applications of machine learning in enhancing carbon emission evaluation and reduction in fossil fuel power plants. The development of a machine learning model can significantly improve the evaluation of CO₂/NO_x emissions compared to traditional emission monitoring methods in power plants. The research status of using different machine learning methods has been reviewed and discussed. The reinforcement learning, neural network, support vector, and extreme learning methods are the most frequently mentioned methods for machine learning. Each machine learning method’s characteristics, as well as its advantages and disadvantages, are also discussed. Forward neural network and Extreme learning methods seem to be better when addressing the real-time prediction of CO₂/NO_x emissions due to their computational efficiency. However, no single method that addresses the model’s accuracy, reliability, and computational efficiency stands out. The machine learning model’s accuracy largely depends on training data quality; however, the availability of historical data for training from power plants can be limited. For a specific power plant, the process conditions, the design of the facility, the running capacity, and other aspects can vary greatly from one location to another. This makes the creation of machine learning models more challenging, and more effort ought to be put into creating a standard model that is accepted by the industry and could be used for all fossil fuel power plants in the future. Researchers can use CFD to address the availability issue of training data that is commonly encountered in power plants, although CFD data generation and validation require prudence and additional computational time. Challenges in model transferability, interpretability, and transparency remain as gaps in the current research. This review also has some limitations, which could be addressed by research in the future. Aside from CO₂/NO_x emissions, future studies could review the applications of machine learning in SO_x and Particulate Matter (PM) emissions, which are very important for the use of fossil fuels for energy production. Furthermore, this review mainly focused on machine learning applications. Future research focusing on the development of other AI techniques is encouraged in general.

Author Contributions

Conceptualization, Z.Z. and H.F.; methodology, Z.Z.; software, Y.N.; validation, Z.Z. and J.L.; formal analysis, Z.Z.; investigation, Z.Z. and J.L.; resources, H.F.; data curation, Y.N.; writing—original draft preparation, Z.Z. and J.L.; writing—review and editing, M.Z., Y.N. and H.F.; visualization, M.Z.; supervision, H.F.; project administration, M.Z. and H.F. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Conflicts of Interest

The authors declare no conflicts of interest.

Footnotes

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.

Figures and Tables

Figure 1. The general flow process of machine learning model development [32].

Figure 2. Flow diagram of the research article selection process.

Figure 3. Machine learning model in the prediction of CO2/NOx emissions from power plants using a keyword network diagram.

Table 1

Inclusion and Exclusion Criteria.

Criterion	Inclusion	Exclusion
Year	2008–2023	<2008, ≥2024
Peer review	Peer-reviewed	Non-peer-reviewed
Language	English	Non-English
Type of article	Journal, proceedings	Book, book chapter, review
Topics	Integration of machine learning with CO2/NOx emissions management in fossil fuel power plants	Only related to 1–2 aspects including machine learning, CO2/NOx emissions management, fossil fuel power plants

Table 2

Status of a machine learning models by the method used by researchers.

Method	References	Characteristics
Reinforcement learning	[35,36]	Good for real-time prediction of CO2/NOx emissions
Forward neural network	[23,37,38,40]	Computationally efficient, straightforward algorithm structure, relatively easy to develop model with sufficient input data
Long short-term memory neural network	[43]	The potential issue of overfitting,Long training time, and computational complexity
Support Vector Method	[44,45,47]	Good for real-time prediction of CO2/NOx emissionsInput data quality requirement is higher
Extreme Learning Method	[48,49,50,51]	Computational efficiency, Potential issue of overfitting

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/app14188442/s1, Table S1: PRISMA Checklist.

References

1. Lamb, W.F.; Wiedmann, T.; Pongratz, J.; Andrew, R.; Crippa, M.; Olivier, J.G.J.; Wiedenhofer, D.; Mattioli, G.; Khourdajie, A.A.; House, J. et al. A review of trends and drivers of greenhouse gas emissions by sector from 1990 to 2018. Environ. Res. Lett.; 2021; 16, 073005. [DOI: https://dx.doi.org/10.1088/1748-9326/abee4e]

2. United States Environmental Protection Agency. Overview of Greenhouse Gases. 2024; Available online: https://www.epa.gov/ghgemissions/overview-greenhouse-gases (accessed on 1 January 2024).

3. World Bank Group. Pollution Prevention and Abatement Handbook; The World Bank Group: Washington, DC, USA, 1998; pp. 245-246.

4. United Nations Climate Change. AR6 Climate Change 2021: The Physical Science Basis; United Nations Climate Change: Bonn, Germany, 2021.

5. International Energy Agency. Global Energy & CO₂ Status Report 2019; International Energy Agency Publications: Paris, France, 2020.

6. Monteiro, F.; Sousa, A. CO₂ Emissions Resulting from Large-Scale Integration of Electric Vehicles Using a Macro Perspective. Appl. Sci.; 2024; 14, 6177. [DOI: https://dx.doi.org/10.3390/app14146177]

7. International Energy Agency. Global Energy Review: CO₂ Emissions in 2021; International Energy Agency Publications: Paris, France, 2022.

8. U.S. Energy Information Administration. Electric Power Monthly; U.S. Energy Information Administration: Washington, DC, USA, 2024.

9. Kanwal, F.; Ahmed, A.; Jamil, F.; Rafiq, S.; Ayub, H.M.U.; Ghauri, M.; Khurram, M.S.; Munir, S.; Inayat, A.; Abu Bakar, M.S. et al. Co-Combustion of Blends of Coal and Underutilised Biomass Residues for Environmental Friendly Electrical Energy Production. Sustainability; 2021; 13, 4881. [DOI: https://dx.doi.org/10.3390/su13094881]

10. Abbasi, F.; Riaz, K. CO₂ emissions and financial development in an emerging economy: An augmented VAR approach. Energy Policy; 2016; 90, pp. 102-114. [DOI: https://dx.doi.org/10.1016/j.enpol.2015.12.017]

11. Guo, B.; Zhou, B.; Zhang, Z.; Li, K.; Wang, J.; Chen, J.; Papadakis, G. A Critical Review of the Status of Current Greenhouse Technology in China and Development Prospects. Appl. Sci.; 2024; 14, 5952. [DOI: https://dx.doi.org/10.3390/app14135952]

12. Conti, J.; Holtberg, P.; Diefenderfer, J.; LaRose, A.; Turnure, J.T.; Westfall, L. International Energy Outlook 2016 with Projections to 2040. USA, 2016. Available online: https://www.osti.gov/servlets/purl/1296780 (accessed on 1 January 2024).

13. Abu Bakar, M.S.; Ahmed, A.; Jeffery, D.M.; Hidayat, S.; Sukri, R.S.; Mahlia, T.M.I.; Jamil, F.; Khurrum, M.S.; Ina, A.; Moogi, S. et al. Pyrolysis of solid waste residues from Lemon Myrtle essential oils extraction for bio-oil production. Bioresour. Technol.; 2020; 318, 123913. [DOI: https://dx.doi.org/10.1016/j.biortech.2020.123913]

14. Mercure, J.F.; Pollitt, H.; Viñuales, J.E.; Edwards, N.R.; Holden, P.B.; Chewpreecha, U.; Salas, P.; Sognnaes, I.; Lam, A.; Knobloch, F. Macroeconomic impact of stranded fossil fuel assets. Nat. Clim. Chang.; 2018; 8, pp. 588-593. [DOI: https://dx.doi.org/10.1038/s41558-018-0182-1]

15. Lasek, J.A.; Lajnert, R. On the Issues of NOx as Greenhouse Gases: An Ongoing Discussion…. Appl. Sci.; 2022; 12, 10429. [DOI: https://dx.doi.org/10.3390/app122010429]

16. Munawer, M.E. Human health and environmental impacts of coal combustion and post-combustion wastes. J. Sustain. Min.; 2018; 17, pp. 87-96. [DOI: https://dx.doi.org/10.1016/j.jsm.2017.12.007]

17. Khodakarami, J.; Ghobadi, P. Urban pollution and solar radiation impacts. Renew. Sustain. Energy Rev.; 2016; 57, pp. 965-976. [DOI: https://dx.doi.org/10.1016/j.rser.2015.12.166]

18. Savickas, D.; Steponavičius, D.; Kemzūraitė, A. A novel approach for analysing environmental sustainability aspects of combine harvester through telematics data. Part I: Evaluation and analysis of field tests. Precis. Agric.; 2024; 25, pp. 100-118. [DOI: https://dx.doi.org/10.1007/s11119-023-10060-6]

19. Savickas, D.; Steponavičius, D.; Kemzūraitė, A. A novel approach for analysing environmental sustainability aspects of combine harvesters through telematics data. Part II: An IT tool for comparative analysis. Precis. Agric.; 2024; 25, pp. 221-234. [DOI: https://dx.doi.org/10.1007/s11119-023-10065-1]

20. International Energy Agency. World Energy Outlook 2021; International Energy Agency Publications: Paris, France, 2021.

21. Hackney, R.; Sadasivuni, S.K.; Rogerson, J.W.; Bulat, G. Predictive Emissions Monitoring System for Small Siemens Dry Low Emissions Combustors: Validation and Application. Turbo Expo: Power for Land, Sea, and Air; American Society of Mechanical Engineers: New York, NY, USA, 2016.

22. Jahnke, J. Continuous Emission Monitoring; John Wiley & Sons, Inc: Hoboken, NJ, USA, 2022.

23. Si, F.; Romero, C.E.; Yao, Z.; Schuster, E.; Xu, Z.; Morey, R.L.; Liebowitz, B.N. Optimization of coal-fired boiler SCRs based on modified support vector machine models and genetic algorithms. Fuel; 2009; 88, pp. 806-816. [DOI: https://dx.doi.org/10.1016/j.fuel.2008.10.038]

24. ABB Measurements & Analytics. Predictive Emission Monitoring Systems: The New Approach for Monitoring Emissions from Industry; ABB Measurements & Analytics: Quebec, QC, Canada, 2019.

25. Ciarlo, G.; Callero, F. Predictive Emission Monitoring Systems (PEMS) Deploying Software-Based Emission Monitoring Systems for Refining Processes. 2013; Available online: https://library.e.abb.com/public/310e4b7c45a742578722e61894999b4e/AT_ANALYTICAL_029-EN_A.pdf (accessed on 1 January 2024).

26. Rockwell Automation Inc. Software CEM Predictive Emissions Monitoring System; Rockwell Automation Inc.: Milwaukee, MI, USA, 2009.

27. Eisenman, T.; Bianchin, D.R.; Triebel, D. Predictive Emission Monitoring (PEM): Suitability and Application in View of. U.S. EPA and European Regulatory Frameworks. Proceedings of the 19th Symp. Ind. Appl. Gas Turbines Committee; Banff, AB, Canada, 19–21 October 2014; 15.

28. Dhall, D.; Kaur, R.; Juneja, M. Machine Learning: A Review of the Algorithms and Its Applications. In Proceedings of ICRIC 2019; Singh, P.K.; Kar, A.K.; Singh, Y.; Kolekar, M.H.; Tanwar, S. Springer International Publishing: Cham, Switzerland, 2020; pp. 47-63.

29. Dutton, D.M.; Conroy, G.V. A review of machine learning. Knowl. Eng. Rev.; 1997; 12, pp. 341-367. [DOI: https://dx.doi.org/10.1017/S026988899700101X]

30. Zhan, C.; Ghaderibaneh, M.; Sahu, P.; Gupta, H. DeepMTL: Deep Learning Based Multiple Transmitter Localization. Proceedings of the 2021 IEEE 22nd International Symposium on a World of Wireless, Mobile and Multimedia Networks (WoWMoM); Pisa, Italy, 7–11 June 2021; pp. 41-50.

31. Ghaderibaneh, M.; Zhan, C.; Gupta, H. DeepAlloc: Deep Learning Approach to Spectrum Allocation in Shared Spectrum Systems. IEEE Access; 2024; 12, pp. 8432-8448. [DOI: https://dx.doi.org/10.1109/ACCESS.2024.3352034]

32. Jiang, W.; Xing, X.; Zhang, X.; Mi, M. Prediction of combustion activation energy of NaOH/KOH catalyzed straw pyrolytic carbon based on machine learning. Renew. Energy; 2019; 130, pp. 1216-1225. [DOI: https://dx.doi.org/10.1016/j.renene.2018.08.089]

33. Sarker, I.H. Machine Learning: Algorithms, Real-World Applications and Research Directions. SN Comput. Sci.; 2021; 2, 160. [DOI: https://dx.doi.org/10.1007/s42979-021-00592-x]

34. PRISMA. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). Available online: https://www.prisma-statement.org/prisma-2020-flow-diagram (accessed on 18 May 2023).

35. Zhan, X.; Xu, H.; Zhang, Y.; Huo, Y.; Zhu, X.; Yin, H.; Zheng, Y. DeepThermal: Combustion Optimization for Thermal Power Generating Units Using Offline Reinforcement Learning. arXiv; 2021; arXiv: 2102.11492[DOI: https://dx.doi.org/10.1609/aaai.v36i4.20393]

36. Cheng, Y.; Huang, Y.; Pang, B.; Zhang, W. ThermalNet: A deep reinforcement learning-based combustion optimization system for coal-fired boiler. Eng. Appl. Artif. Intell.; 2018; 74, pp. 303-311. [DOI: https://dx.doi.org/10.1016/j.engappai.2018.07.003]

37. Smrekar, J.; Assadi, M.; Fast, M.; Kuštrin, I.; De, S. Development of artificial neural network model for a coal-fired boiler using real plant data. Energy; 2009; 34, pp. 144-152. [DOI: https://dx.doi.org/10.1016/j.energy.2008.10.010]

38. Ilamathi, P.; Selladurai, V.; Balamurugan, K.; Sathyanathan, V.T. ANN–GA approach for predictive modeling and optimization of NOx emission in a tangentially fired boiler. Clean Technol. Environ. Policy; 2013; 15, pp. 125-131. [DOI: https://dx.doi.org/10.1007/s10098-012-0490-5]

39. Monteiro, T.O.; Alves, P.A.A.d.S.d.A.N.; Barradas Filho, A.O.; Villa-Vélez, H.A.; Cruz, G. Estimation of the main air pollutants from different biomasses under combustion atmospheres by artificial neural networks. Chemosphere; 2024; 352, 141484. [DOI: https://dx.doi.org/10.1016/j.chemosphere.2024.141484] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/38368962]

40. Adams, D.; Oh, D.-H.; Kim, D.-W.; Lee, C.-H.; Oh, M. Prediction of SOx–NOx emission from a coal-fired CFB power plant with machine learning: Plant data learned by deep neural network and least square support vector machine. J. Clean. Prod.; 2020; 270, 122310. [DOI: https://dx.doi.org/10.1016/j.jclepro.2020.122310]

41. Botros, K.K.; Selinger, C.; Siarkowski, L. Verification of a Neural Network Based Predictive Emission Monitoring Module for an RB211-24C Gas Turbine. Turbo Expo: Power for Land, Sea, and Air; American Society of Mechanical Engineers: New York, NY, USA, 2009.

42. Botros, K.K.; Cheung, M. Neural Network Based Predictive Emission Monitoring Module for a GE LM2500 Gas Turbine. Proceedings of the International Pipeline Conference; Calgary, AB, Canada, 27 September–1 October 2010.

43. Blackburn, L.D.; Tuttle, J.F.; Andersson, K.; Fry, A.; Powell, K.M. Development of novel dynamic machine learning-based optimization of a coal-fired power plant. Comput. Chem. Eng.; 2022; 163, 107848. [DOI: https://dx.doi.org/10.1016/j.compchemeng.2022.107848]

44. Chen, C.P.; Tiong, S.K.; Albert, F.Y.C.; Koh, S.P. A Support Vector Based CO₂ Gas Emission Prediction System for Generation Power Plant. Adv. Sci. Lett.; 2017; 23, pp. 4518-4522. [DOI: https://dx.doi.org/10.1166/asl.2017.8875]

45. Lv, Y.; Liu, J.; Yang, T.; Zeng, D. A novel least squares support vector machine ensemble model for NOx emission prediction of a coal-fired boiler. Energy; 2013; 55, pp. 319-329. [DOI: https://dx.doi.org/10.1016/j.energy.2013.02.062]

46. Wang, F.; Ma, S.; Wang, H.; Li, Y.; Zhang, J. Prediction of NO_X emission for coal-fired boilers based on deep belief network. Control Eng. Pract.; 2018; 80, pp. 26-35. [DOI: https://dx.doi.org/10.1016/j.conengprac.2018.08.003]

47. Zheng, L.; Zhou, H.; Wang, C.; Cen, K. Combining Support Vector Regression and Ant Colony Optimization to Reduce NO_x Emissions in Coal-Fired Utility Boilers. Energy Fuels; 2008; 22, pp. 1034-1040. [DOI: https://dx.doi.org/10.1021/ef700451v]

48. Tan, P.; Xia, J.; Zhang, C.; Fang, Q.; Chen, G. Modeling and Optimization of NOX Emission in a Coal-fired Power Plant using Advanced Machine Learning Methods. Energy Procedia; 2014; 61, pp. 377-380. [DOI: https://dx.doi.org/10.1016/j.egypro.2014.11.1129]

49. Bai, Z.; Li, F.; Zhang, J.; Oko, E.; Wang, M.; Xiong, Z.; Huang, D. Modelling of a Post-combustion CO₂ Capture Process Using Bootstrap Aggregated Extreme Learning Machines. Computer Aided Chemical Engineering; Kravanja, Z.; Bogataj, M. Elsevier: Amsterdam, The Netherlands, 2016; Volume 38, pp. 2007-2012.

50. Rahmanda, A.A.; Soeprijanto, A.; Muhammad, A.; Syaiin, M.; Adhitya, R.Y.; Herijono, B.; Endrasmono, J.; Singgih, A.; Zuliari, E.A.; Haryudo, S.I. et al. Control and monitoring system optimalization of combustion in furnace boiler prototype at industrial steam power plant with comparison of Neural Network (NN) and Extreme Learning Machine (ELM) method. Proceedings of the 2017 International Symposium on Electronics and Smart Devices (ISESD); Yogyakarta, Indonesia, 17–19 October 2017; pp. 123-128. [DOI: https://dx.doi.org/10.1109/ISESD.2017.8253317]

51. Tan, P.; Xia, J.; Zhang, C.; Fang, Q.; Chen, G. Modeling and reduction of NOX emissions for a 700 MW coal-fired boiler with the advanced machine learning method. Energy; 2016; 94, pp. 672-679. [DOI: https://dx.doi.org/10.1016/j.energy.2015.11.020]

52. Liu, X.; Bansal, R.C. Integrating multi-objective optimization with computational fluid dynamics to optimize boiler combustion process of a coal fired power plant. Appl. Energy; 2014; 130, pp. 658-669. [DOI: https://dx.doi.org/10.1016/j.apenergy.2014.02.069]

53. Yang, G.; Wang, Y.; Li, X. Prediction of the NOx emissions from thermal power plant using long-short term memory neural network. Energy; 2020; 192, 116597. [DOI: https://dx.doi.org/10.1016/j.energy.2019.116597]

54. Tuttle, J.F.; Vesel, R.; Alagarsamy, S.; Blackburn, L.D.; Powell, K. Sustainable NOx emission reduction at a coal-fired power station through the use of online neural network modeling and particle swarm optimization. Control Eng. Pract.; 2019; 93, 104167. [DOI: https://dx.doi.org/10.1016/j.conengprac.2019.104167]

55. Cuccu, G.; Danafar, S.; Cudré-Mauroux, P.; Gassner, M.; Bernero, S.; Kryszczuk, K. A data-driven approach to predict NOx-emissions of gas turbines. Proceedings of the 2017 IEEE International Conference on Big Data (Big Data); Boston, MA, USA, 11–14 December 2017; pp. 1283-1288. [DOI: https://dx.doi.org/10.1109/BigData.2017.8258056]

56. AlKheder, S.; Almusalam, A. Forecasting of carbon dioxide emissions from power plants in Kuwait using United States Environmental Protection Agency, Intergovernmental panel on climate change, and machine learning methods. Renew. Energy; 2022; 191, pp. 819-827. [DOI: https://dx.doi.org/10.1016/j.renene.2022.04.023]

57. Ye, T.; Dong, M.; Liang, Y.; Long, J.; Li, W.; Lu, J. Modeling and optimization of the NO_X generation characteristics of the coal-fired boiler based on interpretable machine learning algorithm. Int. J. Green Energy; 2022; 19, pp. 529-543. [DOI: https://dx.doi.org/10.1080/15435075.2021.1947827]

58. Wei, Z.; Li, X.; Xu, L.; Cheng, Y. Comparative study of computational intelligence approaches for NO_x reduction of coal-fired boiler. Energy; 2013; 55, pp. 683-692. [DOI: https://dx.doi.org/10.1016/j.energy.2013.04.007]

59. Zhang, W.; Zhang, Y.; Bai, X.; Liu, J.; Zeng, D.; Qiu, T. A robust fuzzy tree method with outlier detection for combustion models and optimization. Chemom. Intell. Lab. Syst.; 2016; 158, pp. 130-137. [DOI: https://dx.doi.org/10.1016/j.chemolab.2016.09.003]

60. Chen, L.-Y.; Hong, W.-C.; Panigrahi, B.K.; Wei, S.Y. SVR with Chaotic Genetic Algorithm in Taiwanese 3G Phone Demand Forecasting. Swarm, Evolutionary, and Memetic Computing: Proceedings of the Second International Conference, SEMCCO 2011, Visakhapatnam, Andhra Pradesh, India, 19–21 December 2011; Panigrahi, B.K.; Suganthan, P.N.; Das, S.; Satapathy, S.C. Springer: Berlin/Heidelberg, Germany, 2011; pp. 248-256.

61. Guo, M.; Li, D.; Du, C.; Jia, Z.; Qin, X.; Chen, L.; Sheng, L.; Li, H. Prediction of the Busy Traffic in Holidays Based on GA-SVR. Advances in Computer Science and Information Engineering; Springer: Berlin/Heidelberg, Germany, 2012.

62. Lu, Y.; Roychowdhury, V.P. Parallel randomized sampling for support vector machine (SVM) and support vector regression (SVR). Knowl. Inf. Syst.; 2008; 14, pp. 233-247. [DOI: https://dx.doi.org/10.1007/s10115-007-0082-6]

63. Wang, Y.; Xie, R.; Liu, W.; Yang, G.; Li, X. Modeling and Optimization of NO_x Emission from a 660 MW Coal-Fired Boiler Based on the Deep Learning Algorithm. J. Chem. Eng. Jpn.; 2021; 54, pp. 566-575. [DOI: https://dx.doi.org/10.1252/jcej.21we004]

64. Botheju, D.; Glarborg, P.; Tokheim, L.-A. NOx reduction using amine reclaimer wastes (ARW) generated in post combustion CO₂ capture. Int. J. Greenh. Gas Control; 2012; 10, pp. 33-45. [DOI: https://dx.doi.org/10.1016/j.ijggc.2012.05.015]

65. Yang, F.; Du, M.; Hu, X. Evaluating Explanation without Ground Truth in Interpretable Machine Learning. arXiv; 2019; arXiv: 1907.06831

66. Zawawi, M.H.; Saleha, A.; Salwa, A.; Hassan, N.H.; Zahari, N.M.; Ramli, M.Z.; Muda, Z.C. A review: Fundamentals of computational fluid dynamics (CFD). AIP Conf. Proc.; 2018; 2030, 020252. [DOI: https://dx.doi.org/10.1063/1.5066893]

67. Lu, J.; Behbood, V.; Hao, P.; Zuo, H.; Xue, S.; Zhang, G. Transfer learning using computational intelligence: A survey. Knowl. -Based Syst.; 2015; 80, pp. 14-23. [DOI: https://dx.doi.org/10.1016/j.knosys.2015.01.010]

68. Ali, S.; Abuhmed, T.; El-Sappagh, S.; Muhammad, K.; Alonso-Moral, J.M.; Confalonieri, R.; Guidotti, R.; Del Ser, J.; Díaz-Rodríguez, N.; Herrera, F. Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence. Inf. Fusion; 2023; 99, 101805. [DOI: https://dx.doi.org/10.1016/j.inffus.2023.101805]

69. Dwivedi, R.; Dave, D.; Naik, H.; Singhal, S.; Omer, R.; Patel, P.; Qian, B.; Wen, Z.; Shah, T.; Morgan, G. et al. Explainable AI (XAI): Core Ideas, Techniques, and Solutions. ACM Comput. Surv.; 2023; 55, 194. [DOI: https://dx.doi.org/10.1145/3561048]