The NPPs big data could be gathered from mathematical model. The NPPs model is developed using the first principle approach based on fundamental physical laws and assumptions (Vajpayee et al., 2020). It is a valuable and easily accessible data source during the lack of real observations of the NPPs. Big data incorporates diverse data formats and types, including structured, semi-structured, and unstructured data. The dataset is then stored in the database and enough amount of data should be extracted for the intended applications. The PWR model is utilized to generate the necessary amount of data for the prediction of transients under reactivity and inlet coolant temperature perturbations (Ejigu and Liu, 2023). The NPPs system dynamics model is established for studying the transients and designing a risk assessment platform (El-Sefy et al., 2019). Further, the data produced during simulation is employed for ANN training to estimate the NPP behavior and demonstrate the potential of AI in risk mitigation strategies (El-Sefy et al., 2021).

The mathematical model development for the NPPs also assists the plant operator in understanding transients and achieving the necessary safe operation. The solution of the model is obtained by system simulation through MATLAB and Python. The simulation offers several advantages including gaining fundamental concepts of the NPPs dynamics during transients, analyzing the transient of the NPP under normal maneuvering and accident situations, and plant operator training (Kerlin and Upadhyaya, 2019b). However, due to its nonlinearity characteristic, an accurate mathematical model development of the NPPs is a challenging task. The mathematical model of the reactors should be reasonably accurate and simple to accomplish the objectives.

The big data for NPPs is also collected from different software platforms. These sources can offer information for estimation, prediction, safety analysis, and maintenance in the NPPs. However, the availability and accessibility of NPPs data from software sources may vary depending on several factors including security restrictions, regulations, and user permissions. The NPPs data could be collected from RELAP5 thermal-hydraulics codes. It is applied to model the coupled behavior of the primary and secondary systems under various operational conditions. This modeling tool is also used to study the transients of the NPPs (Li R. et al., 2022). It is also employed to model natural circulation flow in the PWR fuel (Ni et al., 2021), estimation of the countercurrent flow in the downcomer (Li et al., 2023a), analysis of loss of flow accidents (Corzo et al., 2023), and study rod ejection accidents (El-Sahlamy et al., 2022). Further, the NPPs data is generated from the STAR-CCM+ CFD simulation tool for several applications (Marfaing et al., 2018; Benavides et al., 2020; Zhang et al., 2021; Yang et al., 2023). Besides, the big data of the NPPs could be collected from education and training simulators. These simulators offer an easy and effective means of examining the physics and engineering designs of multiple kinds of NPPs. Furthermore, the simulators are useful for both technical and non-technical individuals as introductory instructional tools. The IAEA provides several kinds of NPP simulators (Cabellos et al., 2018; Developing a Systematic Education and Training Approach, 2018).

The big data of the NPPs could be collected by conducting experiments. It is carried out in the laboratory to collect comprehensive and robust datasets with the help of apparatus. Experimental data sources can vary widely depending on research goals, and available resources. Experiments in nuclear engineering are performed for different applications (Geslot et al., 2023; Guillen et al., 2023; Zhang et al., 2023).

Sensor data is produced when an instrument recognizes and reacts to some form of physical input. These are the real sources of data (Schokker et al., 2022). The NPPs consist of numerous sensors and a tremendous volume of data could be collected by sensors from the plant site which is recorded over time and continuously. This data provides valuable insights for several applications in the estimation and control of reactor variables. The sensor is employed to regulate core outlet temperature in NPPs (Hyer et al., 2023) and measure the position of the control rod in a nuclear reactor (Hu et al., 2020).

Data mining includes data collection, preparation, and analysis. The purpose of data mining is to extract new information and relationships from the existing raw data. The data collected from the NPPs need preprocessing to ensure accurate, and efficient application. Data preprocessing techniques prepare the raw dataset for suitable model building and AI algorithm training to provide the desired output. Numerous kinds of statistical approaches are employed for preprocessing the NPPs big data. The common data preprocessing techniques include smoothing, cleaning, normalization, and removing (Morrisset et al., 2022). Figure 1 illustrates the procedures undertaken from NPPs big data collection until the required services are obtained in a flowchart. Further, Table 1 presents some big data mining methods.

The key roles of big data are big data engineering and analytics. Figure 2 presents a summary of these functions. Big data engineering incorporates essential steps for data collection, management, and regularization. The key components of big data engineering are acquisition, processing, storage, databases, and pipeline. Big data analytics are used to categorize, characterize, consolidate, predict, infer, and classify data to provide meaningful information. Prediction, classification, clustering, inference, and optimization can be summarized as the core operations, while statistics, data mining, expert knowledge, machine learning, and deep learning are frequently applied techniques (Li F. et al., 2023).

Big data can impact NPPs (Lorenz and Schmidt, 1986). The relationship between NPPs and big data lies in the potential application of big data computing techniques to enhance the performance, safety, and efficiency of NPPs. In order to address the intended application, traditional big data analysis tools face limitations. These drawbacks could be overcome through AI algorithms, that can process and handle the NPPs big data with fast computations. Scholars are creating and employing AI algorithms by considering the potential of big data collected from NPPs.

GA is an evolutionary mechanism that works based on natural selection. The basic idea of the GA starts with the population for the potential solution of a complex problem that evolves iteratively over generations. The GA is applied for the optimization of load and reloading of fuel assemblies in the nuclear reactor core (Sobolev et al., 2017). It is also employed for designing and simulating safe and effective fuel-loading patterns in nuclear reactors (Zhao et al., 1998). Further, the GA is utilized for designing efficient radiation shielding in SMR (Bagheri and Khalafi, 2023), development of an optimized thermodynamic model in a VVER-1200 reactor (Khan et al., 2022), optimal energy management in the HTGR (Sun J. et al., 2022), optimization of fuel loading pattern in the experimental fast reactor (Lima-Reinaldo and François, 2023), in-core fuel management in the PWR (Rodrigues et al., 2022) and optimal design of a core in VVER-1000 nuclear reactor (Kianpour et al., 2020). Yet, the GA shows limitations in both too small and large-scale population sizes to converge into optimal solutions (Cavallaro et al., 2024). Figure 4 demonstrates the overall steps of the GA in a flowchart.

PSO is a metaheuristic algorithm inspired by a group of animals. The concept of swarm intelligence is based on the social collaboration of individuals to learn with their own experience in a group. It is applied to optimize fuel reloading problems in PWR (Meneses et al., 2009), optimization of fuel core loading pattern in a VVER nuclear reactor (Babazadeh et al., 2009), optimization of secondary circuit system in marine NPP (Zhao et al., 2023), optimize the design parameters of radiation shielding system material (Lei et al., 2023), optimization of control drum operation for a microreactor under normal and transient conditions (Price et al., 2022), and designing of space nuclear reactor fuel (Rafiei and Ansarifar, 2022). Besides, the PSO mechanism is employed in NPPs for fault diagnosis (Wang H. et al., 2021), designing maintenance and safety systems (Wang J. et al., 2021), and control system development (Coban, 2011; Safarzadeh and Noori-kalkhoran, 2021; Ejigu and Liu, 2022a; Ayele Ejigu and Liu, 2022; Muthuraj et al., 2023). However, in a high-dimensional search space, the PSO tactic converges slowly toward the optimal solution and produces poor results (Bucz et al., 2018). In order to overcome this limitation, the PSO is combined with the GA (Rahnama and Ansarifar, 2021) and GD (Ejigu and Liu, 2022a) algorithms. Figure 5 displays the overall procedures of the PSO algorithm in a flowchart.

The BAS is a metaheuristic algorithm that works with the foraging principle of the beetles using two antennae. The two antennae of the beetle explore the food odor randomly in the nearby area. The beetle takes a step towards the strong odor concentration using the two antennae. The searching performance of the beetles using two antennae could be used to formulate an optimization algorithm (Jiang and Li, 2017). It is employed for estimation (Xie et al., 2019; Zivkovic et al., 2021), fault detection (Huang et al., 2020), control system optimization (Fan et al., 2019), and cooperative and constrained control design (Ejigu and Liu, 2022b). Recently, the GA, PSO, and BAS algorithms have given more attention to training ANN algorithms for different applications (Li, 2020; Vasumathi and Moorthi, 2012; da Silva Veloso et al., 2020; Yadav and Anubhav, 2020; Jamali et al., 2019). However, the BAS algorithm faces several shortcomings as reported in Ref. (He et al., 2022). Figure 6 summarizes the working principle of the BAS optimization algorithm in a flowchart.

ML is a subfield of AI algorithm that builds a mathematical model based on the data for prediction and making decisions. ML is a powerful data-based modeling mechanism by processing a massive volume of data (Manley et al., 2022). In nuclear engineering, the ML algorithm is employed in NPP to model the surveillance test data (Lee et al., 2021), crack fault diagnosis (Zhong and Ban, 2022), probabilistic safety assessment for fire hazard model (Worrell et al., 2019), seismic fragile analysis (Wang Y. et al., 2023), and equivalence assessment between the simulation and operation data (Li X. et al., 2021). Yet, the ML shows limitations as reported in a review article in Ref (Xu et al., 2024). Figure 7 presents the workflow of ML that comprises different steps from loading the data to integration of the best-trained model into a production system.

ANNs are the most efficient nonlinear modeling and data processing units based on the functioning of a human brain. The ANN designing process involves defining the structure. The building blocks of the ANN are the layers (input, hidden, and output), neurons, and connection weights as shown in Figure 8. The input and output layers are connected by the hidden layer. Successive layers of the ANN are linked by weights. Each layer of the network consists of various amount of processing elements, called neurons. The dataset enters into the network through the input layer. The hidden neurons receive the weighted dataset and process it using the activation function. The output neurons then receive the processed dataset and send it to the users. More, connection weights are used to measure the data and transfer it into the next layer.

Designing the structure of ANNs and selecting an efficient training algorithm are challenging tasks. These issues are open problems for designers. More, the accuracy of training algorithms varies and is affected by the training data points (Zhou et al., 2022). Once the ANNs are trained with a necessary amount of representative quality data, they could be applied to estimate the response under new inputs.

ANNs give attention in nuclear engineering research fields to help plant operators in decision-making to take corrective actions during failure. These intelligence tools are recommended to detect faults in the resistance temperature detector sensors based on the fuel rod temperature profile through modeling (Messai et al., 2015). The ANNs are also suggested to estimate the PWR core state variables online in order to detect faults caused by measurement noise and sensor faults (Kumar and Devakumar, 2022). Further, these modeling mechanisms are used to design the core fuel assembly of the research reactor automatically (Kim et al., 2020). More, they are employed for optimization and burnup calculations of the reactor core (Afzali et al., 2022) as well as for the NPPs fault supervision (Khentout and Magrotti, 2023). The ANNs are suitable and effective mechanisms to diagnose transients of a nuclear reactor during operation and to improve safety (Santosh et al., 2007). Moreover, These potential technologies are employed to predict the state of the nuclear reactor, improve reactor assets as well as empower fast emergency response of nuclear power plants (El-Sefy et al., 2021).

Several types of ANN models are considered and applied for different purposes. The backpropagation neural network is one category of ANN. It is utilized to estimate the PWR core parameters for optimal fuel reloading patterns in order to overcome the restrictions of traditional fuel reloading problems in high-temperature gas-cooled reactors in a short time (Kim et al., 1993). The recurrent multilayer perception ANN model based on the backpropagation algorithm is implemented to model the core neutronics of the NPPs (Adali et al., 1997). Further, the RBFNN is a kind of ANN that has numerous advantages such as simple to design, strong tolerance to disturbance, good generalization, and efficient learning capabilities. Due to these characteristics, the RBFNN model is employed for different applications such as fault assessment, optimization (Sun M. et al., 2022), and adaptive control development (Feng et al., 2022). In nuclear engineering, the RBFNN is deployed to control the core power distribution and rebuild measurements of the core information of the reactor detector (Li W. et al., 2022). Overall, the ANNs seek effective training algorithms. Population-based tactics received more attention for ANN training recently. Figure 9 highlights the possible input and target variables of a reactor to train the ANN through population-based optimization algorithms in a block diagram.

DL is a kind of ML and powerful modeling approach designed by using the DNN model. The DNN model is an intelligent algorithm that works based on the ANN to transform the data into amenable outputs for various applications. The structure of the DNN model comprises numerous hidden layers between input and output layers (Wang J-C. et al., 2023; Yassir et al., 2023), as depicted in Figure 10. As indicated in the block diagram, the workflow in the DNN model starts in the input layer and ends in the output layer. The size of neurons in the input and output layers relies on the input and target variables. However, the design of hidden layers and the corresponding neurons are challenging tasks and an open issue for engineers. These DNN components should be nominated carefully to remove computational challenges such as overfitting and underfitting. The hidden layers and hidden neurons of the DNN model should be simple enough to avoid complexity and reduce computational time. In general, the minimum size of the DNN model is necessary to incorporate good design.

The DNN algorithm is used to model complex systems by creating nonlinear relationships. The accuracy of the developed DNN model output relies on its structure and amount of training data. The advancements in computer systems initiate the use of the DNN model in different architectures in several research areas for various applications. In nuclear engineering, the DNN model is utilized for solving numerous problems such as fault diagnosis (Qian and Liu, 2022a), safety assessments (Bae et al., 2022), internal state prediction (Koo et al., 2021), and control system development (Ejigu and Liu, 2022a). Several DNN models including convolutional neural network (CNN), long short-term memory (LSTM), and multi-layer neural network (MLANN) are reported by (Arji et al., 2023; Sun et al., 2023). However, as presented in Ref. (He et al., 2023), the DNN model shows shortcomings such as overfitting and underfitting. Overall, Figure 11 demonstrates the framework of AI algorithms. The framework also presents the main applications of these AI techniques.

It is a difficult task and AI needs trusted data to capture real information efficiently (Momota and Morshed, 2022). Reliability of the data is an important aspect of science and engineering for making informed decisions, reaching valid findings, and producing credible outcomes. Establishing data reliability in AI is an ongoing and challenging process that necessitates regular monitoring, improvement, and adaptation.

AI and big data applications in NPPs necessitate access to massive amounts of sensitive and vital data. It is critical to protect this data against unwanted access, cyber-attacks, and hacking (Ayodeji et al., 2023). Nuclear data needs a strong cybersecurity framework to safeguard the privacy and security of information.

AI algorithms and big data analysis primarily rely on quality data for precise decision-making. The quality of data also directly impacts the performance generalization and decision-making capability of the AI models (Qi et al., 2022). Ensuring honest data sources and maintaining data quality over time is a significant challenge, especially considering the long operational lifetimes of NPPs.

NPPs are governed by strict rules and safety standards. The integration of AI technologies and big data analytics necessitates modifications to existing regulations and the development of new guidelines to assure compliance while maintaining safety and reliability.

AI models are complex and difficult to comprehend (Balasubramaniam et al., 2023). Transparency in AI decision-making processes is vital in safety-sensitive applications such as NPPs. Operators should understand the judgments of AI to trust and verify its behavior.

Introducing AI and big data into NPPs necessitates a transition in human responsibilities, from direct manual operation to supervisory and decision-support roles (Hiroshi et al., 2021). Effective collaboration between human operators and AI systems is important to ensure safe and optimal plant operation.

The adoption of AI technologies and big data analysis includes substantial costs such as infrastructure investment, personnel training, and ongoing maintenance (OECD and Nuclear Energy Agency, 2020). It might be difficult for NPPs operators to ensure the balance of the benefits with the expenses.

The nuclear industry needs trained and educated personnel who can use AI technologies and big data analytics (International Atomic Energy Agency, 1996). This can be accomplished by enrolling experts from other more mature industries and training specialists in big data approaches relevant to nuclear energy and AI. Combining these experts with other energy domain knowledge experts is recommended. The nuclear industry should also make investments in personnel training and reskilling to manage and operate the systems and assets of NPPs.

These are common issues encountered in the development of models using AI techniques, particularly in ML. Hence, understanding these concepts is vital for developing effective and reliable AI-based models (Rattan et al., 2022). Overfitting and underfitting issues could be overcome by data augmentation, regularization, adjustment of the model, and k-fold cross-validation methods (Mutasa et al., 2020).

Overall, these challenges could be overcome through cooperation among nuclear experts, data scientists, and AI developers. By efficiently managing these difficulties, AI and big data computing can significantly improve the safety, efficiency, and security of NPPs.