Emergency Decision Support Techniques for Nuclear Power Plants: Current State, Challenges, and Future Trends

Abstract

Emergency decision support techniques play an important role in complex and safety-critical systems such as nuclear power plants (NPPs). Emergency decision-making is not a single method but a framework comprising a combination of various technologies. This paper presents a review of various methods for emergency decision support systems in NPPs. We first discuss the theoretical foundations of nuclear power plant emergency decision support technologies. Based on this exposition, the key technologies of emergency decision support systems in NPPs are presented, including training operators in emergency management, risk assessment, fault detection and diagnosis, multi-criteria decision support, and accident consequence assessment. The principles, application, and comparative analysis of these methods are systematically described. Additionally, we present an overview of emergency decision support systems in NPPs across different countries and feature profiles of prominent systems like the Real-Time Online Decision Support System for Nuclear Emergencies (RODOS), the Accident Reporting and Guiding Operational System (ARGOS), and the Decision Support Tool for Severe Accidents (Severa). Then, the existing challenges and issues in this field are summarized, including the need for better integration of risk assessment, methods to enhance education and training, the acceleration of simulation calculations, the application of large language models, and international cooperation. Finally, we propose a new decision support system that integrates Level 1, 2, and 3 probabilistic safety assessment for emergency management in NPPs.

1. Introduction

The issue of sustainable energy development is a focal point and challenge for the world’s future progress. Since the birth of the first electricity-generating nuclear reactor in the United States in 1951 [1], nuclear power, as a high-quality clean energy source, has been instrumental in mitigating the energy crisis and optimizing the energy structure. However, the development of nuclear energy is a double-edged sword. On the one hand, it has the potential to alleviate the energy crisis, achieve “zero carbon” emissions, and contribute to the sustainable development of global energy. On the other hand, its potential risks cannot be ignored. Accidents in nuclear power plants (NPPs) have severe consequences, including the risk of widespread radioactive contamination, the induction of genetic mutations and chromosomal abnormalities in organisms, long latency periods of harm, direct injury to emergency response personnel, and the likelihood of causing poisoning among individuals [2]. Moreover, severe NPP incidents usually involve explosions that cause irreversible damage to ecosystems, resulting in the death of wildlife and even affecting the food chain and ecological balance [3]. They also impose a heavy psychological burden on local populations, leading to decreased social stability [4]. The losses caused by the three major NPP accidents—the 1979 Three Mile Island accident in the United States [5], the 1986 Chernobyl accident in the former Soviet Union [6], and the 2011 Fukushima nuclear plant accident in Japan—are difficult to estimate [7]. These incidents led countries like Germany, Switzerland, and Italy to announce a permanent move away from nuclear power [8] and also prompted a polarized attitude towards nuclear energy among some scholars. Furthermore, the conflict between Russia and Ukraine has cast a shadow of a nuclear crisis over the world [9].

Nuclear power plant emergency decision support technology is a critical tool in NPP accidents [10], serving as an indispensable component of NPP safety management. It always covers a broad range of systems, hardware, and communication technologies, including fault detection and diagnosis, decision-making, and other subsystems. By providing scientific and accurate decision support, it aids in mitigating the impacts of accidents in NPPs, thus protecting the safety of people’s lives and property and ensuring environmental health. Firstly, nuclear power plant emergency decision support technology offers rapid and precise information analysis, assisting decision-makers in understanding the potential development paths of an accident and its potential impacts on individuals, the environment, and facilities [11]. Secondly, it enhances the coordination and efficiency of accident response efforts [12]. In a nuclear power plant emergency, multiple agencies and departments need to work closely together to address the situation. Through integrating communication, resource allocation, and task coordination functionalities, decision support systems (DSSs) facilitate efficient collaboration among different teams, ensuring resources are utilized effectively and avoiding redundant efforts and response delays. Thirdly, it also improves public communication and trust [13]. Transparent and timely information dissemination is an essential part of managing emergencies. Decision support technology enables official agencies to provide accurate information in an easily understandable format which can promptly address public concerns and thereby enhance public trust and cooperation. Finally, with technological advancements, nuclear power plant emergency decision support technology is continually evolving, with modern technologies such as artificial intelligence (AI) and big data analysis being incorporated to further improve the scientific rigor and accuracy of decision-making [14]. The development and application of this technology are of significant importance for enhancing safety levels and fostering a culture for NPPs.

Many countries are advocating for the establishment of an emergency DSS for NPPs. There are some projects that have studied emergency DSSs (for details, please refer to Section 4.2 of this paper). However, there is no systematic compendium of the field. This paper aims to comprehensively review the progress of techniques and establish a new framework for emergency decision support systems for NPPs. Two of the purposes of this paper are to introduce existing nuclear power plant emergency decision support problems to the research group and to underscore probabilistic safety assessment (PSA) techniques in the nuclear power industry to make the combination of PSA and decision support systems better. Note that this paper will not discuss all the techniques in the field of emergency decision support systems, such as emergency preparedness plans. The reader may refer to [15,16] for more information on these.

This paper is organized as follows: Section 2 introduces the theoretical foundations of nuclear power plant emergency decision support technologies. Section 3 introduces the key technologies in nuclear power plant emergency decision support systems. Section 4 introduces some case studies of nuclear power plant emergency decision support systems. Section 5 discusses the existing challenges and issues in emergency management of NPPs. Section 6 is a hybrid probability safety assessment and decision support system for emergency management in NPPs. Section 7 provides the conclusions of this paper.

2. Theoretical Foundations of Emergency Decision Support Technologies for Nuclear Power Plants

It is important to identify the concepts of DSSs and the history of NPP emergency DSSs to understand the theoretical foundations of emergency decision support technologies for NPPs.

2.1. Fundamental Knowledge of Decision Support Systems and Applications in Nuclear Power Plant Emergency Management

DSSs are part of the area of the information systems discipline that is focused on supporting and improving managerial decision-making. The term DSS was introduced by Keen and Scott Morton (1978) [17] to describe a facet of information processing that specifically supports management decision-making. Despite the challenges associated with defining DSSs, Sprague and Carlson (1982) found it feasible to outline a set of characteristics indicative of a DSS. These characteristics encompass problem definition, solution methodologies, and user interaction. In a different vein, King (1983) [18] proposed identifying DSSs through a list of the integral components that any comprehensive DSS should possess, including (a) decision models, (b) interactive computer hardware and software, (c) a database, (d) a database management system, (e) graphical and other advanced display capabilities, and (f) a modeling language designed to be user friendly. Eom et al. (1988) [19] contended that DSSs can be categorized into the following four types:

• Passive Support: This category provides decision-makers with DSS tools that are familiar and user friendly, enabling them to make decisions autonomously;
• Traditional Support: DSS tools are integrated into the decision-making process, enhancing and refining decisions;
• Extended Support: The DSS actively presents alternative decisions to the decision-makers;
• Normative Support: This represents the most involved level of support, where the DSS essentially leads the decision-making process.

Arnott et al. [20] provide a comprehensive overview of the history of DSSs, delineating the evolution across several distinct research and practice subdomains. These include personal DSSs, group DSSs, negotiation DSSs, intelligent DSSs, knowledge-management-based DSSs, executive information systems/business intelligence, and data warehousing. The applications of DSSs are extensive and include areas such as corporate financial planning, marketing analysis, real estate investments, mineralogical exploration, transportation routing, and portfolio analysis [19]. In this paper, we specifically focus on the applications in NPPs. Below is an evaluation of the emergency responses to three accidents in NPPs:

• The chaos at the onset of the Three Mile Island accident (1979) [5] and initial misjudgments about the issue led to delayed and ineffective response measures;
• Following the Chernobyl accident (1986) [6], the lack of transparency and immediate international communication contributed to sluggish emergency response. Effective emergency plans were not adhered to initially in addressing the accident, and the severity of the situation was initially underestimated;
• The Fukushima [7] accident (2011) was triggered by the Great East Japan Earthquake and the subsequent tsunami. The scale and complexity of the accident exceeded expectations, making it difficult to implement some planning measures. The emergency response following the Fukushima incident faced criticism, particularly regarding the disclosure of information and evacuation planning.

VOSviewer 1.6.17 is a software tool for constructing and visualizing bibliometric networks [21]. It is tailored for crafting and visualizing bibliometric networks, including co-word, co-citation, and co-authorship networks. As shown in Figure 1, we used VOSviewer software to conduct statistical and cluster analyses of the literature to obtain the hot topics and frontier trends in the field of “emergency decision support for nuclear power plants”. Different colors identify distinct clusters or thematic groups, grouping frequently co-occurring keywords under the same color. The linkage of “software tool” and “reliability” signifies the reliability of the software tools employed. The decision support studies span multiple disciplines, including engineering, computer science, and environmental science. For instance, “software tool” could refer to computer programs for simulating and evaluating emergencies. The emphasis on terms like “emergency preparedness”, “implementation”, and “decision-making process” in the map implies that the outcomes of these studies have direct implications for real-world emergency preparedness and policy formulation. The keyword “complexity” indicates that the studies acknowledge the complexity of emergency response and decision-making processes which involve multiple variables and uncertainties.

2.2. A Brief History of Nuclear Power Plant Emergency Decision Support Systems

The history of emergency DSSs for NPPs can be divided into three distinct phases of development. In Figure 2, the diagram presents a timeline depicting the historical evolution. The line within the graph represents the technological growth trajectory, delineating three distinct phases of development: the early history, the exponential growth phase, and steady development.

2.2.1. Early History

As early as 1961, Mayo et al. [22] introduced the concept of aiding decision-making in NPPs. During this period, research on decision support primarily focused on site selection for NPPs [23,24,25,26] and the choice of reactor types [27], with only a minority of studies addressing emergency response plans for nuclear facilities [28]. Before 1965, the creation of large-scale information systems required significant financial investment. The origins of the early DSSs can be attributed to management information systems (MIS), which were predominantly developed for gathering and processing company data, thus aiding managerial decision-making processes. However, during the 1960s and 1970s, the MIS in NPPs did not experience significant advancement. This period saw the emergence of a variety of data management initiatives [29], service systems [30], and life-support systems [25].

In the 1980s, the concept of the DSS in NPPs underwent significant refinement. Wallace et al. [31] concentrated their research on DSSs, distinguishing DSSs from the more traditional MIS. They articulated that DSSs are specifically tailored to support complex decision-making processes. Bonczek et al. [32] focused on constructing a DSS. Brill et al. [33] discovered that a human–machine decision-making system performed better when the human was presented with a diverse set of alternatives as opposed to a homogeneous set of options. Rasmussen [34] posited that it is essential to focus on hierarchical knowledge representation in decision-making and system management. Silver et al. [35] proposed a three-tiered approach to describe DSSs.

2.2.2. Exponential Growth Phase

In the late 1980s, expert systems began to flourish. Nelson first introduced the concept of an expert system for nuclear reactor accidents [36]. Subsequently, after a few years, there was a significant surge in research related to expert systems for NPPs in 1987 and 1988. Gallanti and Guida (1986) [37] dedicated their work to elucidating the influence of expert system technology on the development of intelligent decision aids for process environments. Numerous scholars focused on proposing approaches to constructing decision support systems based on expert systems methodologies [38,39,40,41,42].

Meanwhile, during this period, the graphical user interface (GUI) emerged [43,44,45,46,47,48], enabling non-technical users to utilize DSSs more easily. Furthermore, during this period, NPP emergency decision support systems began to integrate external data sources, enhancing decision relevance and efficiency [19,44,49,50,51]. Meanwhile, during this era, the advancement of internet technology enabled DSSs to be accessible online, supporting decentralized and remote decision-making processes. Consequently, numerous studies emerged based on web-based technologies [12,52,53,54]. These decision support systems focused on model-driven DSSs constructed utilizing decision analysis, optimization, and simulation technologies [55] and incorporated spatio-temporal data [49,56]. Lastly, many countries also developed their national emergency decision support systems for NPPs during this period, as detailed in Section 4.1.

2.2.3. Steady Development

In this era, with the expansion of industrial networks, NPPs’ vast arrays of sensors have been generating extensive datasets for status monitoring. This development aids operators in making rapid and accurate decisions, mitigating the influence of subjective human judgment throughout the decision-making process. However, there has been limited adoption of advanced algorithms and artificial intelligence technologies. Conversely, in the fault diagnosis component of nuclear power plant emergency decision systems, a considerable amount of research incorporating deep learning [57,58,59], reinforcement learning [60], and machine learning [61,62,63,64] has surfaced.

Furthermore, entering the 2020s, the surge of artificial general intelligence and computing (AIGC) has ushered in unprecedented transformations across various industries. The advent of GPTs has altered daily life, with numerous applications of large language models (LLMs) emerging in specialized fields such as medicine [65] and law [66], showcasing their extensive impact and utility. Therefore, future research could consider integrating AIGC to explore related studies further. This approach may offer new insights and methodologies in the field.

3. Key Technologies in Emergency Decision Support Systems for Nuclear Power Plants

The U.S. Nuclear Regulatory Commission (NRC) and the Federal Emergency Management Agency (FEMA) have established four emergency action levels for nuclear power plants: (1) unusual event, (2) alert, (3) site area emergency, and (4) general emergency. These levels are strategically structured to escalate in response to the increasing potential impact on NPP safety and public health. The unusual event represents the lowest tier of operational irregularities, where anomalies could potentially escalate to affect plant safety. However, the deviations at this level are not expected to have any immediate impact on public safety. At the second level, an alert will be triggered by a failure in primary safety systems. However, it still remains under control and without any off-site risk. At the third stage, there could be potential radiological releases within regulatory limits but significant enough to warrant rigorous on-site emergency responses. The general emergency is the highest emergency classification, reserved for extreme situations where radiological controls have failed. There is a likelihood of substantial radioactive releases that could have widespread environmental and health impacts at this stage. Whereas most incidents never become site area or general emergencies, a major incident might progress through each of the four stages. As shown in Figure 3, this study provides a conceptual framework for various technological strategies in emergency management for NPPs. It is systematically categorized into two main phases: pre-accident and accident. In the first phase, the emphasis is on mitigation and preparedness. It starts with training operators and emergency management, highlighting the importance of human resource development through education and drills, which enhances the capacity of responders and managers. Risk assessment technology refers to the evaluation of potential risks and vulnerabilities. The next level is planning. It denotes the formulation of strategies and protocols to address potential accidents effectively. Note that this paper will not discuss emergency planning. The reader may refer to [67] for more information on the history of and methods for diagnosis techniques. Transitioning to the accident phase, the focus shifts to response, which involves immediate actions to be taken. This phase is further detailed as three components for NPPs: fault detection and diagnosis, multi-criteria decision support, and accident consequence assessment and calculation. In addition to key technologies, some studies focus on the visualization of interfaces [68].

3.1. Training Operators and Emergency Management

Safety culture was first introduced in 1986 after the Chernobyl accident, and it has received renewed attention recently in safety-critical industries including nuclear power [69]. Many statistical results from safety reports suggest that human-related errors are the dominant influencing factor in the safe operation of NPPs. For instance, Miranda et al. [70] demonstrated that insufficient knowledge acquisition compromises the safety of NPPs. Fortunately, training operators in technical and non-technical skills can prevent many types of human errors. Training operators no longer involves operators just simply taking courses. In recent years, it has become more in depth and detailed. Certain studies have quantitatively assessed the impact of human perceptions on the risks associated with NPPs [71,72,73]. Some studies focused on managing workers’ physiological health, emphasizing the importance of health control in enhancing overall workplace productivity and safety [74,75]. Furthermore, as early as 2001, Corcoran et al. [76] proposed the principles of nuclear power plant operational safety. Several studies also aimed to mitigate the impact of human factors through management strategies [77]. Some studies have explored enhancing the immersion and realism of training simulations to improve learning outcomes. Mendonça et al. [78] introduced gaming simulations for assessing decision aids. Moreover, there are some research report results obtained with the use of a game engine as a tool to create and navigate in virtual environments [79], or with the visualization of 3D radiation dose [80] to perform simulations and training of workers in risky areas for safety purposes.

It is worth noting that some new advancements have made immersive virtual environments (IVE) such as virtual reality (VR) a viable tool for operator training. VR is an important visualization technique increasingly applied for plant safety purposes [81]. In safety-critical industries, training is costly, and operator error must be avoided. For example, VR is routinely used to teach manual and technical skills through simulation before pilots assume flight responsibilities. Additionally, VR and IVE have been used in diverse applications related to training and education such as engineering applications [82,83,84,85,86,87,88], medical surgery [89,90,91,92,93,94,95,96], and fire [97]. However, it is worth noting that VR for training operators has not been widely utilized in NPPs.

3.2. Risk Assessment

Quantitative risk analysis constitutes a pivotal component for NPPs. Several studies have been conducted in the past decade to support quantitative risk analysis. These studies are based on some commonly used techniques such as risk matrix, decision trees, Monte Carlo, and sensitivity analysis. In addition, most of the studies combine these methods with fuzzy theory [77,98]. Furthermore, it is noteworthy that, in research, the risk assessment methods for multi-unit NPPs often differ from those for single-unit facilities, necessitating separate studies for accurate evaluation [99]. Additionally, developing living risk detection technologies represents a new trend [100]. For instance, Shalev et al. [101] propose condition-based fault tree analysis not only during the design phase of a system but also throughout its operational life cycle. In conclusion, the current risk assessment methods in NPPs can generally be categorized into two types: traditional risk assessment methods and domain-specific mathematical equation risk modeling.

3.2.1. Traditional Risk Assessment Methods

Nuclear power plant risk assessment refers to the systematic analysis and evaluation of potential hazards and risks across various stages of construction, operation, and decommissioning. The assessment encompasses multiple factors including the architectural structure, system equipment, operational management, personnel quality, and environmental impacts. The primary objective is to ascertain the overall safety level of the NPP, reduce the probability of accidents, minimize the losses resulting from accidents, and ensure the safe and reliable operation of the NPP. There are two basic risk assessment methods: deterministic safety analysis (DSA) and probabilistic safety assessment (PSA). These approaches are somewhat complementary to each other. Some researchers integrate PSA with DSA to conduct risk analysis [14,102,103,104,105]. This hybrid approach synergizes the strengths of both methodologies, enhancing the comprehensiveness and accuracy of risk assessments in NPP operations.

Deterministic Safety Analyses. Deterministic safety analysis can predict the response to postulated initiating events. A specific set of rules and acceptance criteria is applied. Typically, these studies concentrate on the neutronic, thermohydraulic, radiological, thermomechanical, and structural aspects. Furthermore, they are frequently analyzed using various computational tools. Deterministic safety analyses for design purposes are characterized by their conservative assumptions and bounding analysis. This is achieved by an iterative process in the design phase where the limiting cases in terms of the minimum margin to the acceptance criteria are determined for each group of postulated initiating events and sequences. To determine the limiting case for a given transient or set of transients, the consequential failures caused by the initiating event (internal or external) are taken into account.

Probabilistic Safety Assessment. In the domain of NPPs, PSA is the most extensively employed method for risk assessment. In 1975, the U.S. NRC officially published the first formal report, known as WASH-1400, on PSA [106]. Subsequently, in 1981, the International Atomic Energy Agency (IAEA) released the fault tree handbook [107], marking the widespread adoption of PSA for safety evaluations. PSA is structured into three levels: Level 1 PSA focuses on evaluating the core damage frequency; Level 2 PSA assesses the containment performance and potential radioactive release; and Level 3 PSA examines the consequences of radioactive release for the environment and public health. In addition, Geographic Information System (GIS) technology is often applied in conjunction with Level 3 PSA [108,109,110].

The analytical methodologies of PSA encompass event trees, fault trees, dynamic PSA, Bayesian networks, Monte Carlo simulations, hazard and operability study (HAZOP) [111], and others. They offer a robust toolkit for risk evaluation. In practical analysis, fault tree analysis (FTA), event tree analysis (ETA), and failure mode and effects analysis (FMEA) are often employed in conjunction [112]. As shown in Figure 4, this study illustrates a comprehensive procedural synergy of three pivotal risk analysis methodologies, FMEA, ETA, and FTA, through a simple case example. FMEA is employed to identify the initial event, which is input into the event tree to determine the top event. Subsequently, a fault tree is constructed to conduct the risk assessment. This article provides detailed explanations of these three critical analysis methods.

Failure Mode and Effects Analysis. FMEA, a particularly prevalent method, was initially delineated in the U.S. Military Procedure MIL-P-1629. It was later refined in MIL-STD-1629A by the U.S. Department of Defense in 1980. As an inductive analysis technique, FMEA meticulously examines all conceivable permutations of the impacts resulting from single-component failure modes. Furthermore, this method facilitates the execution of probabilistic analysis, enabling the determination of the criticality associated with various failure modes. Through this comprehensive approach, FMEA provides a robust framework for identifying and mitigating potential risks in system design and operation [113]. This section briefly touches on FMEA. For more detailed exploration, reference [114] offers extensive insights.

Event Tree Analysis. Event tree analysis is a forward logic model employed to analyze the potential pathways and outcomes stemming from an initial event, such as a system failure or operational error. By evaluating different paths of success and failure, ETA assesses the probabilities of the system reaching various final states following a specific initial event. This analysis is graphically represented as an event tree, where each branch depicts a possible sequence of events ultimately leading to diverse outcomes or system states. Through this structured approach, ETA facilitates a comprehensive understanding of the potential impacts of initial events, aiding in critical control points and effective risk management strategies. Research on event trees and fault trees is closely aligned, with a significant focus on the integration of fuzzy theory [115] and the exploration of dynamic event trees [116]. For instance, Brian et al. [117] integrated safety and security analysis of NPPs using dynamic event trees. The INL has developed its tool to perform dynamic PRA, the Reactor Analysis and Virtual Control Environment (RAVEN) [118], modeling the interface of manual fire protection actions with fire progression in fire PSA for NPPs.

Fault Tree Analysis. Fault tree analysis [119] is a well-established and well-understood technique widely used to determine system dependability. In fault trees, the logical connections between faults and their causes are represented graphically. FTA is deductive, meaning that the analysis starts with a top event (a system failure) and works backward from the top of the tree towards the leaves of the tree to determine the root causes of the top event. The results of the analysis show how different component failures or certain environmental conditions can combine to cause a system failure. There are different forms of fault trees, including static, dynamic, and non-coherent fault trees [120]. Advanced dynamic FTA [121,122,123,124,125] extends traditional FTA by incorporating time-dependent behaviors and sequences of events that can affect the system’s reliability and safety. Some fault tree studies also incorporate expert knowledge bases [126]. Additionally, several studies have applied enhancements to FTA by integrating fuzzy theory [127], aiming to address the inherent uncertainties and subjective judgments involved in risk assessment processes. This approach is particularly useful in scenarios where quantitative data on parameter fluctuations are limited [128,129,130,131].

3.2.2. Domain-Specific Mathematical Equation Risk Modeling

The calculation methods for external risks to NPPs, such as earthquakes [132], evacuations, traffic dispersal, and distribution of materials and fires, often differ from those applied to internal risks. Some studies have integrated knowledge from these specific domains to construct corresponding risk assessment models for NPPs. For instance, Zografos et al. [133] proposed a decision-making approach regarding the routing of hazardous materials based on the associated transportation risk. Sakurahara et al. [134] developed a methodology for modeling the interface between manual fire protection (i.e., manual fire detection and suppression) and a computational fluid dynamics (CFD) fire progression model utilizing the fire dynamics simulator (FDS). The core function of the risk assessment subsystem developed by Jensen et al. [110] involves the processing of anomalies. Upon detection of an anomaly, detailed information regarding the failure’s characteristics (such as the onset speed, type of material, and concentration released), its origin, and estimated intensity is relayed to the analytical component of the risk assessment system. This subsystem is tasked with identifying populations at risk and integrating these data points to assess potential hazards and their impacts accurately. Meng et al. [135] segmented the urban emergency development process into discrete time slices, each depicting a cascading event sequence. By employing population distribution heat maps, they assessed the impact on individuals across varying scenario areas throughout these sequences to gauge severity. Concurrently, they introduced a network-based risk analysis method utilizing network indices to evaluate scenario risks from a structural perspective. Integrating these assessments, they prioritized scenarios requiring immediate attention, streamlining the nuclear power plant emergency response strategy. Zhao et al. [136] developed a nuclear emergency partition evacuation framework by integrating nuclear radiation risk, predicated on radionuclide concentration, with traffic risk, informed by vulnerability assessments.

3.3. Fault Detection and Diagnosis

Fault detection and diagnosis involve the steps of monitoring, detecting, and identifying system faults. Fault diagnosis includes fault isolation and fault identification [137]. Fault detection and diagnosis methods can be applied to monitor a system continuously during operation, which is often referred to as online monitoring.

3.3.1. Monitoring and Detection

Monitoring the electrical and mechanical components of systems is directly associated with the performance and safety of NPPs. Online monitoring systems have been applied to reactor vessel internals, pumps, safety and relief valves, and turbine generators. The monitoring techniques include noise analysis, vibration analysis, and loose parts detection [138]. The online monitoring also extends to sensor surveillance, including response time, and encompasses fatigue monitoring of components, diagnostics of rotating machinery (notably bearing degradation and shaft crack detection), and critical safety parameter monitoring [139,140,141], such as subcooling margin determination. Additionally, chemical monitoring is integral to maintaining the plant’s operational safety and efficiency [142]. It is worth noting that extensive research exists on sensor calibration [143,144].

Data from NPPs can be categorized into two types: static and dynamic. Within the emergency decision support system of an NPP, the following data require monitoring, as highlighted by French et al. [145]: (a) plant status data, which encompasses the readings from various sensors within the facility, such as temperature, pressure, humidity, and radiation monitoring; (b) current and predicted meteorological information; (c) on-site stack and periphery monitoring data, which include off-site fixed and mobile data; (d) hydrological data, which pertain to flow rates, depths, and contamination levels; (e) population data, detailing the demographics of those potentially exposed to risks, such as villages, towns, and cities around the site [146]; (f) agricultural, economic, and land use data; and (g) data concerning the adherence to and effectiveness of implemented countermeasures. It is worth noting that meteorological data used in simulations usually come from two sources, the European Centre for Medium-Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction (NCEP). However, the study of the Fukushima accident has shown that the agreement of simulation results with NCEP data is better than that with ECMWF data [147]. The detection data prepare the groundwork for subsequent aging monitoring systems [148], situation awareness [149], and fault diagnosis [150].

3.3.2. Fault Diagnosis

This section primarily provides an overview of the methods used in fault diagnosis. Fault diagnosis is classified into three main categories: model-based methods, signal-based methods, and data-driven methods, as referenced in [151]. The following sections will detail each of these methods.

Model-Based Methods. Model-based methods are divided into two main categories: the consistency-based approach and abductive diagnosis [152]. In a consistency-based approach, a mathematical model is used to represent the normal system behavior. Faults in the system are detected and diagnosed by checking the consistency between the observed behavior and the predicted behavior through the model. This process is illustrated in the flowchart presented in Figure 5, summarizing the logical reasoning behind fault diagnosis within the formal model-based diagnosis (MBD) and fault detection and isolation (FDI) frameworks. Figure 5a describes the MBD framework, which starts by defining residuals using available analytical redundancy relations (ARRs). Conflicts are then identified from non-zero residuals, leading to minimal diagnoses derived from these conflicts. Figure 5b explains the FDI framework, which also begins by defining residuals using ARRs. However, it includes an additional step where zero residuals are used to isolate relevant faults, reducing conflicts before obtaining minimal diagnoses. In contrast, abductive diagnosis relies on models depicting the behavior under abnormal conditions. These methods are based on, e.g., parameter estimation, Kalman filters, parity equations, or state observers. However, the practical application of model-based fault detection and diagnosis methods faces significant challenges due to the difficulty of obtaining accurate models in real-world scenarios [151].

Data-Driven Methods. Data-driven methods for fault diagnosis in NPPs have attracted increasing interest in recent years. They include artificial neural networks (ANNs) [150,153,154], support vector machines (SVMs), decision trees (DTs), principal component analysis (PCA), and clustering [155]. Certainly, numerous studies have also opted for hybrid algorithms [155]. In data-driven fault diagnosis for NPPs, a common research paradigm is depicted in Figure 6. Figure 6 illustrates the process starting with the collection of real-time data from nuclear power plants. These data are then fed into a data-driven fault diagnosis model which processes the data to detect anomalies or potential faults. The model’s output is used to provide early warnings, enabling timely interventions and preventing potential issues from escalating. It is worth noting that these models need a large number of data. Under the current wave of global industrial intelligence, numerous countries have researched the integration of AI, especially through the development of digital instrumentation and control systems, which can collect large numbers of operational data. Our previous work presents a first-of-its-kind open dataset created using PCTRAN, a pre-developed and widely used simulator for NPPs [156].

Signal-Based Methods. Signal-based methods operate in the time domain and employ techniques such as wavelet analysis, time–frequency analysis, and spectral analysis. Signal-based fault diagnosis methods can be thus classified into time-domain signal-based approaches, frequency-domain signal-based approaches, and time–frequency signal-based methods [157]. As shown in Figure 7, the process begins with the monitoring of signals generated from the process. These signals are then subjected to feature extraction, where relevant characteristics or patterns are identified. The extracted features are evaluated to determine their significance and relevance to potential faults. Based on this evaluation, symptoms are generated which are then compared to baseline measurements and existing knowledge when a diagnostic decision is being made [158]. Although this scheme is simple, it can only be applied to simple processes with the aid of experienced operators for fault isolation. In fact, for signal-based fault diagnosis, fault signatures need to be extracted from representative fault data. However, they are difficult to obtain in most situations. Additionally, in a real process, the measured signals always contain noise [159], so traditional signal-based methods cannot adapt to the complexity of modern mechanical systems [160].

In conclusion, due to the limitation in practical applications of model-based methods, various data-driven methods [161,162,163] and signal-based methods [157,164,165] have been applied for monitoring key subsystems in NPPs. The advantages and disadvantages of different methods and the common associated algorithms are summarized in

Table 1. Merit and demerit of fault diagnosis approaches.

Method Classification	Advantages	Disadvantages	Common Algorithms
Model-based methods	High reliability and high interpretability	Need to establish systematic mathematical modeling, and it is very difficult to achieve	Parity equations, observers, Kalman filters, parameter estimation
Data-driven methods	No need for an accurate analytical model, and the modeling process is relatively simple and universal	Sample data are difficult to obtain, sensitive to data changes, not interpretable, and difficult to convince operators	ANN, SVM, decision tree, PCA, clustering, multivariate state estimate technique (MSET), partial least squares (PLS), auto associative kernel regression (AAKR)
Signal-based methods	It possesses inherent interpretability and does not require excessive data	Cannot fully utilize data information	Spectrum analysis, time–frequency analysis (TFA), wavelet transform (WT), autoregressive (AR) signal model, control charts

[166]. Model-based methods are noted for their high reliability and interpretability. However, they require systematic mathematical modeling, which is challenging to achieve. Data-driven methods are lauded for not requiring an accurate analytical model, with the modeling process being relatively simple and universal. The disadvantages include difficulty in obtaining sample data, sensitivity to data changes, and a lack of interpretability that often leads to skepticism from operators. Signal-based methods are appreciated for their inherent interpretability and the lack of a need for excessive data. Nevertheless, their inability to fully utilize data information marks a significant drawback. Furthermore, it should be noted that some studies also integrate mechanical simulation models with data-driven approaches [167,168].

3.4. Multi-Criteria Decision Support

Decision-making in NPPs during emergencies involves a diverse array of stakeholders, each with their unique perspectives, responsibilities, and interests, necessitating a consensus. French et al. [169] highlights considering multiple attributes in supporting decision-making following a nuclear power plant accident, as illustrated in Figure 8. It shows the hierarchy of attributes considered in the decision-making process, emphasizing the balance between effects and resources. Under the effects category, health impacts are further divided into radiation-related and stress-related issues. Public acceptability encompasses the affected region, the rest of the U.S.S.R. This comprehensive approach ensures that all critical factors are taken into account, providing a holistic view of the consequences. The adoption of multi-criteria decision-making (MCDM) enhances the transparency and traceability of decisions by incorporating subjective preferences, making it an ideal approach for engaging various stakeholders and expert groups. MCDM facilitates the inclusion of all relevant parties with varied background knowledge and differing views in the decision-making process, ensuring that the final decision represents the best possible compromise.

MCDM techniques are particularly suitable for decisions in NPPs due to the multiple-criteria nature of these types of problems, and they have been widely applied in this area [170]. Applications of MCDM in nuclear power plants encompass a variety of critical tasks. These tasks include, but are not limited to, site selection [171,172,173,174,175], determination of nuclear power plant types [176,177,178], impact on the living standard [179], emergency and remediation management [180], intervention strategies [181], and environmental risk management [182,183]. Additionally, MCDM finds application in energy policy assessments [184,185,186,187], risk evaluation, assessment of nuclear safety culture [188], and the techno-economic analysis and optimization of NPP projects [189]. It is noteworthy that the projects discussed in Section 4.2, namely, NARSIS [190] and RODOS [10], have employed MCDM tools.

The MCDM process is systematically divided into four sequential phases, as illustrated in Figure 9. In Phase I, a comprehensive criteria system is developed through synthesizing relevant studies and expert opinions. Phase II focuses on identifying suitable alternatives that align with the objectives of the issue at hand. Phase III involves a detailed evaluation of these alternatives, structured into three steps to ensure thorough analysis. Phase IV entails making the final decision, incorporating an aggregation of preference orders to reflect a collective decision-making approach [191]. Additionally, conducting a sensitivity analysis is deemed crucial in this phase to assess the stability and reliability of the decision, ensuring that the chosen solution stands resilient under various scenarios.

Specifically, the MCDM technique is classified into multi-attribute decision-making (MADM) and multi-objective decision-making (MODM). MODM is applicable for decision spaces in continuous space such as problems that require mathematical programming with multiple objective functions, while MCDM is utilized for problems where their decision space is discrete [192]. Therefore, in this study, a comparative analysis is performed by describing and implementing seven prominent MCDM methods in the field of NPPs: basic methods AHP, TOPSIS, PROMETHEE, SAW, VIKOR, and COPRAS and advanced method MAUT [183]. The strengths and weaknesses of these methods [193] are detailed in

Table 2. A summary of the strengths and weaknesses of the MCDM methods used in the present studies.

Method	Strengths	Weaknesses
AHP	The AHP method is advantageous for its straightforward application and adaptable hierarchy structure, suitable for diverse issue sizes.	The subjectivity and the requirement for comparative rather than independent option grading.
TOPSIS	The TOPSIS method excels in delivering stable results across varying data due to its simple, programmable approach and direct evaluation without needing data transformation, preserving data integrity.	The use of Euclidean distance neglects attribute correlation, making weighting and consistent judgment difficult, particularly with extra attributes.
PROMETHEE	The PROMETHEE method is valued for its ease of use, bypassing variable minimization and preserving data integrity without distortion.	This tool does not provide a clear framework for assigning the weights.
SAW	It features a simple algorithm that can be executed manually or with basic software, offering variable compensation and demonstrating versatility and user-friendliness.	The estimates yielded do not always reflect the real status. The result may not be consistent in terms of logic, with the measures of one particular variable widely differing from one of other variables.
VIKOR	The VIKOR method introduces stability intervals in weights. The result of ranking is a list of alternatives after special compromise ranking and the solution with an advantage rate.	This tool needs initial weights.
COPRAS	It allows the final results of measuring to be easily compared and checked.	COPRAS is less stable compared to SAW or TOPSIS methods in data variation cases.
MAUT	It incorporates uncertainty, providing a thorough evaluation of all consequences and preferences at each calculation step.	The MAUT method requires extensive data at every step to reflect decision-makers’ preferences, rendering it data heavy and subjective.

This table incorporates a range of MCDA techniques, each abbreviated as follows: analytic hierarchy process (AHP), technique for order preference by similarity to ideal solution (TOPSIS), preference ranking organization method for enrichment evaluations (PROMETHEE), simple additive weighting (SAW), visekriterijumsko kompromisno rangiranje (VIKOR), complex proportional assessment (COPRAS), and multi-attribute utility theory (MAUT).

, with specific methodological details available for reference [174,194,195,196,197,198,199]. Notably, the MCDM method integrated with fuzzy theory [171,177], strengths, weaknesses, opportunities, and threats (SWOT) analysis [171], and GIS [172,174,175,200] has gained popularity and visibility.

3.5. Accident Consequence Assessment and Calculation for Nuclear Power Plants

Assessing and calculating accident consequences are tools for decision-making in an NPP emergency.

3.5.1. The Principles of Radiation Release from Nuclear Power Plants

Some basic information, including the various types of radiation such as alpha particles and the basic units of radiation dosage, along with their calculation methods, is detailed in references [201,202]. Several studies have simulated dose calculations for a specific nuclear power plant under accident conditions, such as Xianning NPP [203], Sanmen NPP [204], Krško NPP [205], and Bushehr NPP [146,206]. Other research has specifically recreated the radiation dispersion scenarios following the Fukushima [207,208,209,210,211,212,213] and Chernobyl nuclear accidents [214,215]. Additionally, the decommissioning of facilities in NPPs requires a rapid dose estimation method [216]. In the past, dose estimation methods included the deterministic and stochastic methods [217]. It also plays an important role in Level 3 PSA of NPPs [218].

The research extensively recognizes the link between radiation exposure and cancer risk [219,220]. Nowadays, strategies to protect the population in the early phase of a nuclear power plant crisis consist of three main actions: sheltering, evacuation, and iodine pill ingestion. The first chain of events is associated with the environmental consequences of the release, and the second chain of events involves the social and organizational response to that release. Yumashev et al. [221] advocate for long-term responses within the context of flexible decision-making following large-scale nuclear emergencies. Miao et al. [222] propose a dynamic dose-based emergency evacuation model for enhancing NPP emergency response strategies. Malizia et al. [223] propose free license codes as a DSS for emergency planning to simulate radioactive releases in case of accidents in NPPs.

3.5.2. Off-Site Consequence Calculation Model

The calculation of off-site consequences for NPPs typically employs models to assess the impact of radioactive material leakage on human health and the environment in the event of an accident. The RASCAL (Radiological Assessment System for Consequence Analysis) software [224], developed by the U.S. NRC, offers rapid assessments of radioactive release quantities and their impact zones following NPP accidents. It incorporates atmospheric dispersion models and health impact assessments. Similarly, the MACCS (MELCOR Accident Consequence Code System) [225] evaluates the effects of radioactive releases on the surrounding environment and population after severe NPP accidents, covering aspects like atmospheric dispersion, ground deposition, and early and late health impact assessments. In Europe, the collaborative model system COSYMA (Code System from Maria) [226] is used for assessing radiation impacts after NPP accidents, integrating modules for atmospheric dispersion, food chain contamination, and radiation dosage evaluations among the population.

The primary parameters considered in the accident consequence assessment and calculation include meteorological conditions, terrain, and population distribution. With these monitoring data, the ability to obtain the best estimates of release rates, radiation dose maps, and plume movements is improved. For instance, the dose projection software developed for the Krško NPP located in Slovenia can be used for quick emergency evaluation in the case of a hypothetical pressurized water reactor accident and for emergency exercises. It was developed to estimate reactor core damage, the status of fission product barriers, potential releases, atmospheric dispersion, and, finally, dose calculation. The software comprises several key components: measurements and software modules, an air pollution dispersion calculation module, the radioactivity of the core, the radiological importance of the source term, the assessment of the core damage, and dose projection.

3.5.3. Dosimetric Calculations for Humans and the Environment

Dosimetric quantities are needed to assess human radiation exposures quantitatively. In practical environmental impact assessments, three aspects are commonly considered: the radiological impact of airborne effluents, the radiological impact of liquid effluents, and the triad of key analyses. The exposure pathways for airborne effluents include immersion external irradiation, ground deposition external irradiation, inhalation internal irradiation, and ingestion internal irradiation. For liquid effluents, exposure pathways encompass shore deposition external irradiation, underwater immersion external irradiation, and ingestion of marine products internal irradiation. The triad of key analyses focuses on critical population groups, critical exposure pathways, and critical radionuclides.

As for the calculation, there are two classic methods: the Monte Carlo method and the point kernel integration (PKI) method. The Monte Carlo method is preferred for transport calculation due to its powerful capability to accurately model the physics of particles in scientific research, such as in Super MC, which was developed by fusion digital simulation (FDS). However, the application of the Monte Carlo method for generating massive 3D fine-mesh dose rate results of the whole shielded hall for visualization is still constrained for large-scale nuclear plants. Further research has computed radioactivity concentrations and dose assessment for soil samples from NPPs [227]. Additionally, certain studies leverage data to forecast doses for unobserved times [228]. Several studies have employed Gaussian processes for predicting radiation doses in NPPs. This approach facilitates accurate forecasting, essential for ensuring safety and efficiency. Researchers prioritize precision in dose estimation to optimize reactor performance and mitigate risks [229]. The International Commission on Radiological Protection (ICRP) offers models, dose coefficients, and reference levels that facilitate human dose assessments based on radioactive materials.

4. Case Studies of Nuclear Emergency Decision Support Systems

In this section, we present an overview of the emergency DSSs for NPPs in various countries, along with profiles of prominent systems such as RODOS, ARGOS, and NARSIS.

4.1. Emergency Decision Support Systems for Nuclear Power Plants in Some Countries

In this section, we endeavor to explore NPP emergency decision support systems at the national level. Due to limitations in searching for resources, this paper only presents the cases of France, Korea, the United Kingdom, Belgium, and China.

France. The coordination among various organizations is highlighted in the DSS for French pressurized water reactor off-site emergency management [230]. Aid for decision-making is supplied at a national level by an EDF emergency team and emergency teams from the Safety Authority (DSIN), the Radiological Protection Authority (DGS/OPRI), and the Civil Defense Authority (DSC), all of which are located in Paris. These teams are in constant liaison throughout a nuclear power plant emergency. Furthermore, the 3D/3P approach (triple diagnostic/triple prognostic), developed with the operating organization, has been implemented. Two computer systems support this system: SESAME, which addresses questions about the installation, and CONRAD, which calculates environmental radiological consequences.

Korea. In 1997, the Korean Institute of Nuclear Safety developed the computerized technical advisory system for radiological emergency (CARE) [231]. It is a pioneering system designed to manage accidents during radiological emergencies by collecting emergency data, analyzing it, estimating projected doses, and recommending public safety measures. Significantly, Korea has integrated the application of big data into its NPP safety and emergency management frameworks to enhance the effectiveness of such systems. CARE comprises six key modules: SIDS for safety data collection from NPPs; REMDAS for automatic acquisition of local weather data; IERNet for national environmental radiation monitoring; FADAS for assessing atmospheric dispersion and radiation doses; a technical support module guiding protective actions and emergency preparedness; and an information management module responsible for data handling, database management, graphical interfaces, information control, and maintaining communication networks with emergency organizations.

United Kingdom. In 1997, the United Kingdom unveiled a DSS specifically designed to bolster emergency response capabilities during NPP incidents [232]. This comprehensive system offers a suite of functionalities tailored to effectively manage crises streamlined into a coherent process that includes site control, accident consequence assessment, departmental notification, countermeasure strategizing against environmental pollution, accident reconstruction efforts, and efficient information dissemination. It is crafted with a user-friendly interface to ensure accessibility and ease of use for responders. At its core, the system integrates plant status data, radiation monitoring readings, and meteorological information to conduct sophisticated analyses, thereby supporting the strategic deployment of emergency measures and facilitating informed decision-making for mitigating the impacts of nuclear incidents.

Belgium. In 2004, Belgium significantly underscored the importance of teamwork in nuclear power plant emergency response, specifically spotlighting the deployment of communication tools and the nuclear radiological emergency response room (NREE) [233]. This focus was aimed at ensuring comprehensive coverage during emergencies, incorporating detailed information on the unfolding of the event, data from radiological surveys, forecasts of radiological effects tailored to the specific type of incident, meteorological updates, casualty figures, and insights into deployment strategies and the resources at hand. Additionally, Belgium introduced a meticulously structured data exchange format, the EURDEP format, designed to facilitate the exchange of crucial information. Characterized by its remarkable extendibility and flexibility, the EURDEP format has gained widespread acceptance across a variety of measurement and modeling systems. Its implementation signifies Belgium’s dedication to improving the efficiency and effectiveness of data communication and analysis in managing radiological emergencies.

China. In 1993, the Chinese State Council issued regulations on emergency management of accidents at NPPs [234], establishing the national policy for emergency management. This seminal document delineated the primary responsibilities and tasks of relevant governmental departments in managing nuclear emergencies. It further outlined essential elements for consideration in emergency preparation, including the execution of protective actions and the formulation of emergency response procedures. Moreover, it set forth a series of regulations concerning the siting, construction, commissioning, operation, and decommissioning of NPPs. In a significant move towards enhancing emergency management, the Chinese competent authority resolved in 1995 to develop a Chinese decision support system tailored for this purpose. Consequently, the RODOS was adopted as the foundational platform by the National Nuclear Emergency Management Agency (NNEMA) in the 1990s, marking a pivotal step in the Chinese approach to managing emergencies for NPPs.

Beyond the aforementioned nations, the development of emergency DSSs for NPPs has seen the emergence of key institutions. There are two such entities in accident management and radiological data exchange: CEDIM (https://www.cedim.kit.edu/english/index.php (accessed on 15 May 2024) [235]) and EURDEP (https://remon.jrc.ec.europa.eu/About/Rad-Data-Exchange (accessed on 15 May 2024) [236]). CEDIM, the Center for Disaster Management and Risk Reduction Technology, prioritizes the advancement of technologies for disaster management and risk mitigation. Conversely, EURDEP, the European Radiological Data Exchange Platform, established in 1986 following the Chernobyl incident, enhances international collaboration in radiological monitoring and data exchange. It now encompasses 39 countries and over 5500 automatic monitoring systems, guaranteeing rapid and dependable access to radiological data in nuclear emergencies.

The European Union supports numerous projects aimed at developing emergency DSSs for NPPs. A concise overview of significant projects related to emergency management and safety improvements for NPPs across Europe from 1996 to 2021 is illustrated in

Table 3. Comprehensive overview of emergency decision support projects in nuclear power plants.

Abbreviations	Project Duration	Full Names	Objectives
RODOS https://resy5.ites.kit.edu/, accessed on 15 May 2024RODOS [237]	1996–1999	Real-Time Online Decision Support System for Nuclear Emergencies	This project aimed to develop a comprehensive real-time online decision support system for emergency management in NPPs across Europe [237].
DAONEM https://cordis.europa.eu, accessed on 15 May 2024 [238]	2000–2004	Data assimilation for off-site nuclear emergency management	This initiative aimed to enhance the RODOS by developing and integrating a data assimilation capability for managing nuclear emergencies off-site [238].
MODEM	2001–2005	Monitoring data and information exchange among decision support systems	The MODEM project used XML technology to stimulate communication between scientific experts from different countries and institutes by facilitating the exchange of information used in decision support models to assess the impact of a release of radioactive material in the environment [233].
EVATECH https://cordis.europa.eu, accessed on 15 May 2024 [239]	2001–2005	Information requirements and countermeasure evaluation techniques in nuclear emergency management	It focused on information requirements and countermeasure evaluation techniques in nuclear emergency management [239].
EURANOS https://resy5.ites.kit.edu/, accessed on 15 May 2024EURANOS [240]	2004–2009	European approach to nuclear and radiological emergency management and rehabilitation strategies	This project aimed to improve nuclear and radiological emergency management and rehabilitation strategies across Europe [240].
FP7	2007–2013	European Union’s Seventh Framework Programme for Research and Technological Development	It aimed to strengthen the scientific and technological base of European industry and to encourage its international competitiveness.
PREPARE	2013–2016	Innovative integrated tools and platforms for radiological emergency preparedness and post-accident response in Europe	The project aimed to improve nuclear and radiological preparedness in Europe, especially considering the lessons learned from the Fukushima accident [240].
HARMONE	2015–2017	Harmonising Modeling Strategies of European Decision Support Systems for Nuclear Emergencies	This project sought to improve the environmental modeling and human dose assessment capabilities of JRODOS [241].
NARSIS https://www.narsis.eu/, accessed on 15 May 2024 [242]	2017–2021	The New Approach to Reactor Safety Improvements	This project aimed to improve safety assessment for NPPs in the event of external natural hazards [242].

. It enumerates various initiatives like RODOS, DAONEM, MODEM, EVATECH, EURANOS, FP7, PREPARE, HARMONE, and NARSIS, detailing their operational periods, core objectives, and contributions to the nuclear safety domain. For instance, RODOS, operational between 1996 and 1999, focused on real-time online decision support for nuclear emergencies. HARMONE, running from 2015 to 2017, aimed to enhance environmental modeling and human dose assessments. The table systematically presents these projects, offering insights into their goals, timelines, and impacts on European nuclear safety standards.

4.2. Profiles of Prominent Emergency Decision Systems for Nuclear Power Plants

Based on the introduction to the EU projects in Section 4.1, this section will provide a detailed discussion of several related systems, including RODOS, ARGOS, and Severa.

4.2.1. RODOS: The Real-Time Online Decision Support System for Nuclear Power Plant Emergencies

RODOS (a real-time online decision support system for off-site nuclear emergencies in Europe) [243] was commissioned within the radiation protection research action of the European Commission’s nuclear fission safety program. It is a common platform for incorporating the best features of existing decision support systems and future developments. This represents the first coordinated universal platform involving multiple countries. It facilitates improved communication between countries of monitoring data, as well as sound prediction of consequences, also covering the countermeasure strategy effectiveness and other relevant aspects of accidents for NPPs.

RODOS is a typical DSS for emergency response in NPPs. As shown in Figure 10, it comprises the analysis subsystem (ASY), the countermeasure subsystem (CSY), and the evaluation subsystem (ESY). By 2005, Germany had published a study focusing on RODOS in off-site emergency management scenarios [244]. RODOS has been re-engineered in the last decade as multiplatform software JRODOS (Version 2.1) in a Java environment. The software architecture of JRODOS organizes the data flow between different sources and recipients, e.g., databases, numerical models, and user interfaces, via unified data objects. These objects (data items) are organized in an expandable hierarchical tree of Java classes using the benefits of object-oriented programming principles. Numerical model integration is carried out by distributed wrapper objects (DWO), which provides logical, visual, and technical integration of computational models and the system core, even if models use different programming languages such as FORTRAN, C, and Java.

4.2.2. ARGOS: The Accident Reporting and Guiding Operational System

The ARGOS [245] decision support system is used for consequence assessment and decision support following NPP emergencies. The ARGOS system is in operation or being commissioned in eight European countries, as well as in Canada, Brazil, and Australia. Since the initial version was released 15 years ago, the system has developed from a simple data presentation application to a sophisticated platform that integrates radiological monitoring data, atmospheric dispersion models, and calculation of doses in the food chain and the urban environment.

ARGOS’ interfaces with an SQL server are shown in Figure 11. It facilitates GIS visualization and automates the display of monitoring data. The system encompasses atmospheric dispersion assessment, incorporating short and long-range models to estimate low-altitude contaminant spread in urban environments. Additionally, ARGOS integrates the source modeling of reactors and unconventional explosive devices, food chain and dose modeling, MCDA modules, GIS, ERMIN (European Model for Inhabited Areas), and the modeling of chemical and biological releases for comprehensive risk assessment. As for communication, ARGOS provides the option to publish results, including animations or data visualizations, online with regular updates through an auto prognosis internet service. This sophisticated approach allows ARGOS to stand as a robust framework for NPP emergency preparedness and response.

4.2.3. Severa: The Decision Support Tool for Severe Accidents

Severa is the main product of the H2020 project “New Approach to Reactor Safety Improvements” (NARSIS, 2017–2021), which aimed to propose some improvements to be integrated into existing PSA procedures for NPPs considering single, cascade, and combined external natural hazards (earthquakes, flooding, extreme weather, tsunamis) [246]. The project led to the release of various tools together with recommendations and guidelines for use in nuclear safety assessment, including a Bayesian-based multi-risk framework able to account for causes and consequences of technical, social/organizational, and human aspects, as well as a supporting severe accident management decision-making tool for demonstration purposes.

Figure 12 delineates the eight-step cyclical process model employed by Severa. Commencing with monitoring operating parameters, the model advocates for a systematic assessment of barrier status. This is followed by the prognostication of accident progression, subsequently informing the identification of pertinent recovery actions. The fifth step involves the practicability of the proposed actions. Advancing further, the model prescribes forecasting potential outcomes resulting from the actions. The penultimate stage entails a comprehensive analysis of the actions, leading to the recommendation of the most efficacious one. The cycle culminates with the implementation of the selected action, after which the process recommences, perpetuating the decision-making cycle in intervals ranging from 10 to 20 min in length.

5. Existing Challenges and Issues in Nuclear Emergency Management

Although the current emergency DSSs for NPPs have been developed over approximately 30 years, there still are some existing challenges and issues for emergency management. These challenges include, but are not limited to, the need for more comprehensive data integration, overcoming methodological limitations, and enhancing the scalability of current models [10,247,248]. Based on the aforementioned insights, the current challenges in this field are summarized as follows:

• The integration between the current NPP emergency DSSs and risk assessment is insufficient. There has been a considerable underutilization of fault trees, event trees, and other PSA results;
• The emergency decision support system for current NPPs lacks comprehensiveness in accident response, notably missing a resource allocation module. Despite incorporating evacuation studies related to traffic and fire scenarios, the system has yet to integrate these elements;
• The methods for education, training, and the popularization of nuclear safety culture also require enhancement. Despite the availability of 3D training technologies, practical applications in the workplace are limited due to economic and technical challenges. The integration of digital humans, embodied intelligence, and robotics technologies could offer significant improvements [249,250,251];
• The simulation calculation module is notably time consuming. Conducting consequence analysis is time intensive and complex, indicating a significant need for optimization. To address this, the introduction of AI methods is an excellent approach. Furthermore, the advanced merging capabilities of popular LLMs, which mimic human behavior to the extent of passing the Turing test [252], indicate their potential to substitute human roles in certain contexts, thereby reducing the risks associated with human errors;
• Despite the widespread acknowledgment of the importance of sharing experiences related to NPP incidents, the gap between this recognition and reality is substantial. The significant risks posed by NPPs, exemplified by the Japanese wastewater discharge incident [253], highlight a challenge too vast for any nation to tackle alone. There is an urgent demand for enhanced cooperation and communication among nations, which would ensure a collective capacity to address and mitigate the risks associated with NPPs.

6. From Review to Innovation: Proposing a Hybrid Probabilistic Safety Assessment and Decision Support System for Emergency Management in Nuclear Power Plants

In NPPs, risk assessment is a crucial process during the design phase involving extensive work and analysis of potential faults or accidents. However, fault trees and event trees built in this analysis are not currently utilized in the decision support phase. Therefore, we propose integrating the fault tree and event tree analyses from the PSA into the decision support process to better leverage prior knowledge. This integration will enhance the interpretability and responsiveness of current DSS strategies. Additionally, our framework can specifically incorporate PSA into other processes such as operational training and emergency planning.

Specifically, the comprehensive framework built in this study uniquely synthesizes existing PSA insights to inform decision-making during emergencies, as illustrated in Figure 13. It is delineated into pre-accident and accident stages. In the pre-accident segment, the framework places a strong emphasis on training/education, risk assessment, and planning, all grounded in a PSA knowledge base encompassing Level 1 (incident frequency and consequence analysis), Level 2 (accident progression analysis), and Level 3 (off-site consequence analysis) PSA. The risk assessment stage generates PSA and provides insights across these levels to devise mitigation strategies which can guide the development of emergency response plans. Consequently, planning informs and guides the educational and training modules, which inherently reduces risks and affects the risk assessment results.

Transitioning into the accident phase, an initiating event triggers the accident response sequence, encompassing operational monitoring, barrier assessments, warning issuances, progression predictions, recovery action identifications, and the execution of pre-planned mitigation strategies. It is important to note that steps 4, 5, 6, and 7 are highlighted in gray-pink, indicating a critical integration with prior PSA knowledge. Specifically, following fault diagnosis in NPPs which confirms the initial event, it is noted that the consequences of accidents are often uncertain. To address this, we propose the reuse of event trees and fault trees from the prior PSA knowledge. Given that the initiating event is identified, certain event trees can be eliminated, transforming large-scale event trees into more manageable, small-scale ones. This simplification facilitates the recalculation of probabilities for various accident outcomes. Consequently, it becomes possible to assess the probabilities of different accident progressions. Furthermore, by integrating Level 2 and Level 3 PSA, a comprehensive analysis of both on-site and off-site radiological consequences of accidents can be performed. Additionally, the mitigation strategies have also been analyzed in prior PSA. This reuse of prior PSA knowledge could provide critical decision support and action recommendations for the accident management process.

7. Conclusions

This paper reviews emergency decision support techniques for nuclear power plants. Emergency decision-making technology is not a single method but a framework comprising a combination of various technologies. This integration includes training operators in emergency management, utilizing risk assessment models, fault detection and diagnosis, multi-criteria decision support, and accident consequence assessment. The principles, application, and comparative analysis of these methods are systematically described. Additionally, we present an overview of emergency decision support systems in nuclear power plants across different countries and feature profiles of prominent systems like RODOS, ARGOS, and Severa. This review highlights the critical infrastructure and methodologies used globally to manage emergencies. Then, the existing challenges and issues in this field are summarized, including the need for better integration of risk assessment, methods to enhance education and training, the acceleration of simulation calculations, the application of large language models, and international cooperation. Lastly, a new decision support system that integrates PSA for emergency management in NPPs is established for future research and development.