Resilience-Oriented Management of Integrated Energy Systems: A Review of Characteristics, Quantification, Assessment and Enhancement Methods

Abstract

As the coupling and interaction among subsystems in integrated energy systems (IESs) increase, these systems are becoming increasingly vulnerable to failures under extreme weather events and natural disasters, which threatens overall operational security and stability. This paper reviews recent studies on the resilience-oriented management of IESs, with a focus on the characterization and assessment of energy system resilience covering various types of resilience-challenging extreme events, the related quantification metrics, methods, and resilience enhancement applications. Based on the reviewed studies, this paper attempts to figure out how internal and external adversities impact the systems, and identifies effective methods to detect, assess, and quantify system vulnerabilities and weak links under extreme events scenarios. Rather than confining the analysis to single-carrier systems, this study bridges cross-disciplinary perspectives to construct a resilience-oriented conceptual framework specifically designed for IESs, centering on the core logical, analytical, and technical strategies for both characterizing and advancing IESs’ resilience. Also, this review tries to reveal research opportunities to address significant gaps in the existing literature. Findings from the review can inform future research and help to develop scalable and effective ways to enhance IESs’ resilience.

1. Introduction

With the continuous growth of the global economy and the steady increase in population, energy and environmental issues have become the focus of attention around the world. An integrated energy system (IES) generally refers to a system that facilitates the coordinated optimization and complementary utilization of multiple energy forms, including electricity, gas, heat, cooling, and so on [1]. Through multi-energy coupling and synergetic optimization, such systems improve energy efficiency and operational economics, constituting a major mode of future energy supply, and provide an inevitable choice to achieve clean and low-carbon energy supply. Generally defined integrated energy systems encompass various types across different scales, such as residential small-scale IESs, district-level IESs, and urban-scale IESs. Energy systems of varying scales exhibit certain differences in size, renewable energy deployment, and technical measures [2], which may confront different technical issues when it comes to challenges. As one of the most critical infrastructures in modern society, an IES is of great significance for ensuring the normal functioning under various disturbance events. From this perspective, identifying the key concepts related to the quantification and evaluation of system performance under disturbances is of great significance.

As highly complex energy networks, IESs are experiencing a deeper coupling of heterogeneous energy flows, leading to a significant rise in the risk of system faults propagating across energy chains through coupling components [2]. Moreover, extreme weather, natural disasters, and sudden security incidents (such as cyberattacks) have shown an increasingly frequent trend with intensive global climate change [3] and persistent political instability [4]. Characterized by unpredictability and destructiveness, these events frequently trigger urban energy outages or demand surges, disrupting both essential appliances (e.g., plants, control devices, communication infrastructure) and critical building services (e.g., ventilation, heating, cooling) of IESs [5,6]. In the face of risks, it is vital to minimize vulnerability, improve flexibility, adapt to the surrounding environment, and increase the tolerance of IESs. This is referred to as system resilience enhancement.

Resilience theory finds application in diverse fields—from social sciences and public management to economics, engineering, as well as ecological and environmental sciences [7,8]. In 1973, Holling was the first to define resilience as the ability of a system to absorb or resist the effects of disturbances without altering its normal function [9]. As for the definition of energy system resilience, there is no consensus. Some organizations and researchers have attempted to define it from conceptual and empirical perspectives [7,10,11]. At an early stage, the energy resilience regularly combined with infrastructure resilience, which was defined as the ability to reduce the vulnerability of an element, absorb the effects of disruptive events, enhance its ability to respond and recover, and facilitate its adaptation to disruptive events [12,13]. With deeper coupling of heterogeneous energy flows and advancing interactive technologies of IESs, nowadays, the research focus has gradually shifted from infrastructure resilience to the quantification of overall resilient characteristics of systems [14]. That does not specify an isolated capability of any single infrastructure, but rather comprehensive ones that encompass all system components’ performance characteristics. For the sake of further research, it is also important to provide a clear, operational working definition of IES resilience.

The deep coupling of heterogeneous energy flows renders integrated IESs’ inherently complex. However, the existing literature shows ambiguity regarding the concepts of resilience, assessment indicators, evaluation frameworks, and enhancement methods, especially for IESs. This critical review, therefore, investigates comparatively representative definition and evaluation frameworks. The effective methodologies for identifying, assessing, and quantifying system vulnerabilities and weak links under adversity scenarios are also identified in this paper. With the particular contribution of exploring resilience metrics, evaluation approaches, and enhancement strategies from different perspectives (e.g., adverse event impacts, energy system heterogeneity, and application-specific characteristics), this paper tries to solve the following questions related to the IESs’ resilience:

(i)

How to define IESs’ resilience given its complexity, by linking adverse events to practical applications?

(ii)

What are the preferred metrics, frameworks, and methods to analyze and evaluate the IESs’ resilience?

(iii)

How to enhance IESs’ resilience, particularly with the advancement of emerging data and the accompanying technologies?

(iv)

What are the existing research gaps and future research perspectives?

In summary, we present a comprehensive literature review of IESs’ resilience, aiming to synthesize cross-disciplinary perspectives into an IES-tailored conceptual framework with a focus on core logical, analytical, and technical approaches for characterizing and improving resilience in IESs. With a specific focus on IESs rather than single-carrier systems, this study also advances a finer taxonomy of disturbances impacting system resilience, notably differentiating proximal from distal ones. Their features are examined within resilience assessment and modeling reviews, providing a basis for refined adverse-event characterization in subsequent work.

The subsequent sections are organized in the following manner: Section 2 introduces the preliminary methodologies and databases required to retrieve the IESs’ resilience-correlated literature. Section 3 delves into the definition of IESs’ resilience, particularly trying to figure out the relationship between the definition formation and adverse events. The classification of adverse events, as the origins of resilience-related incidents, is also expressed in Section 3. Section 4 explores the details of identifying and quantifying IESs’ resilience. Resilience assessment metrics, frameworks, and approaches are systematically analyzed in this part. While Section 5 represents various stages of resilience enhancement methods, it also summarizes state-of-the-art approaches, especially those driven by advances in artificial intelligence (AI) techniques, and discusses their corresponding applications. Finally, the conclusions, research gaps, and possible future directions are summarized in Section 6 and Section 7.

2. Research Methodology

The systematic literature review was conducted by performing three search runs in the ScienceDirect, Web of Science, and China National Knowledge Infrastructure (CNKI) databases with iteratively refined keywords. The selected run turns and related keywords are shown in

Table 1. Selected search runs.

Search Runs	Search Keywords
First run	“Integrated energy systems”
Second run	“Integrated energy systems” AND “resilience”
Third run	“Integrated energy systems” AND “resilience” AND “extreme events”

, and the further screening and cleaning query set used to retrieve research papers about IES resilience are outlined in

Table 2. Query set for the literature further screening and cleaning.

Query Set	Meaning
TS = “Energy system” NOT “Transportation, Supply chain, Social work, etc.”	The focus of the research is set around the energy system
TS = “Resilience” NOT “Flexibility”	Keywords are set for energy resilience
AND (“Power system” OR “Power grids” OR “Electricity and Natural gas system, etc.”	The focus of the research is set on the energy- and electricity-integrated system
AND (“Adversity” OR “Disaster” OR “Extreme weather”	Emphasizing specifying disturbance

.

The search for the literature was conducted in December 2025. As “resilience” has emerged as a relatively novel concept alongside the evolution of energy systems and the expansion of renewable energy deployment, no publication date restrictions were imposed in this study. The majority of the retrieved papers were published between 2016 and 2026. A total of 1341 research papers were identified after the first two runs in two English databases, comprising primarily peer-reviewed journal articles and conference proceedings. For the sake of comprehensiveness, reports and Chinese journals were also considered. The reports were excluded before screening, considering the credibility and authoritativeness. One Chinese journal paper was finally included. It was observed that some retrieved papers addressed the resilience of IESs without adequately considering the impact of extreme events. Consequently, a third search iteration was conducted. After screening and cleaning were performed, the corresponding query sets were presented in Table 2. After eliminating duplicate and tangential papers and screening the research topics based on their titles and abstracts, correlated articles were further evaluated according to their research boundaries and key problems. Finally, the total number of articles included in the review is 103. The screening process is shown in Figure 1.

The keywords cloud for reviewed articles is shown in Figure 2, where the text size indicates frequent use of keywords and closely related key terms in the article abstracts. Keywords ranked in the top 20 by frequency of occurrence are presented in Figure 3. As shown in the figure, the most frequent keywords in the IESs’ resilience research include energy, resilience, system, power, and others, indicating that these topics have received the most attention. The system integration, optimization, and operations research have received extensive attention. The frequent appearance of power and electricity indicates that power systems have been mainly studied.

3. Resilience in Integrated Energy Systems

For IESs, “resilience” is not an inherent system attribute, but rather a concept that has gradually emerged as extreme events become more frequent and systems face increasingly uncertain shocks from multiple sources. In general, resilience is defined as the ability to adapt to abrupt environmental changes, withstand disturbances, and rapidly recover into the original or new stable state [15]. It is an obvious dynamic process involving different phases, which can be summarized as preparedness, mitigation, response, and recovery. Other studies alternatively conceptualize these stages as vulnerability, resistance, robustness, and recovery, or as defense, mitigation, resistance, and recovery [5,10,15,16].

3.1. Differentiation of Key Concepts: The Difference Between Resilience and Reliability, Robustness, Flexibility, and Stability

Resilience is a complex concept encompassing multiple scenarios that require differentiation from reliability, robustness, flexibility, and stability [2,5,7,10]. Table 3 presents the comparative similarities and differences among them. Reliability refers to the system’s ability to demonstrate stable performance under frequent disturbances, which mainly refers to the high-probability and low-impact adversities [2,7,17,18]. Robustness emphasizes the system’s ability to maintain functional integrity under internal and external disturbances, with most studies focusing on “predictable” scenarios [2,7,17,18,19]. Flexibility concerns the system’s ability to endure frequent, low-impact disruptions such as routine equipment failures, scheduled maintenance, and peak-hour grid congestion. It entails adjusting demand to need [17,18,19]. Stability emphasizes the system’s maintenance of equilibrium, whereas resilience permits the system to proactively adapt to changes.

Table 3. Comparison of resilience and other conceptions [2,5,7,17,18,19,20]; both the similarities and differences are compared with resilience.

	Similarities	Differences
Reliability	Both focus on the energy supply loss scenarios	Focus on high-probability, low-impact adversities
Robustness	Both focus on the high-probability, high-impact scenarios	Emphasizes maintaining stable operation under disturbances, more focused on “predictable” scenarios
Flexibility	Both emphasize adaptation and guarantee supply–demand matching	Focus on frequent, low-impact disruptions
Stability	Both emphasize the ability to recover supply–demand matching	Stress the equilibrium operation rather than allowance to adapt to change

Table 3. Comparison of resilience and other conceptions [2,5,7,17,18,19,20]; both the similarities and differences are compared with resilience.

	Similarities	Differences
Reliability	Both focus on the energy supply loss scenarios	Focus on high-probability, low-impact adversities
Robustness	Both focus on the high-probability, high-impact scenarios	Emphasizes maintaining stable operation under disturbances, more focused on “predictable” scenarios
Flexibility	Both emphasize adaptation and guarantee supply–demand matching	Focus on frequent, low-impact disruptions
Stability	Both emphasize the ability to recover supply–demand matching	Stress the equilibrium operation rather than allowance to adapt to change

Clearly, the resilience of an IES differs from related concepts, such as robustness. Clarifying what resilience entails is essential for enhancing system resilience, necessitating an explicit definition grounded in the characteristics of both IESs and their associated perturbations.

Hereby, the definition of IESs’ resilience could be construed as follows: IESs’ resilience indicates a system capability to adapt to and tolerate shocks arising from unpredictable and high-impact disturbances. It also encompasses its ability to recover to the original or a new stable state following the disturbances.

3.2. Development of IESs’ Resilience

IESs are more complicated systems compared to the traditional energy systems. They are typically electricity-centered, enabling coordinated operations across multiple energy carriers, including electricity, natural gas, cooling, and heating, with intricate coupling among heterogeneous energy networks. The supply portfolio of an IES encompasses conventional thermal power plants (coal and gas-fired) and distributed energy resources such as photovoltaics, wind turbines, and combined heat and power (CHP) units. For energy conversion, heat pumps, boilers, and power-to-gas facilities are employed. Furthermore, energy storage is achieved through electrochemical and thermal storage systems [21]. Given its comprehensive connotation, IESs may suffer extreme event attacks from any of their subsystems. Meanwhile, the complementarity and backup effects among its multi-source, multi-network, multi-load, and multi-storage elements can enhance its resilience capabilities when facing extreme events [2]. Having a clear and consistent awareness of the causes and correlated influential factors of resilience is a necessity for the risk management of IESs.

Resilience perturbations denote adversities manifested through disruptive events. From the definition of IESs’ resilience, the unpredictability and uncertainty are the inherent characteristics along with the referred adversities, which concern the related occurrence and consequences [15]. System adversities are commonly understood to arise from dual sources: external environmental factors, inherent multi-energy coupling characteristics, and internal system dynamics [22]. Due to the change in global climate, there is an increasingly frequent occurrence of extreme weather events, which is one of the external factors. Another disturbance coming from the external environment for the IESs is natural hazards [23,24,25]. Attributing to progressive convergence of energy and information infrastructures, with advanced information and digital transformation in tandem, IESs are becoming increasingly intelligent and information-intensive. They exhibit high vulnerability to cyber-physical attacks. The adversities arising from internal system dynamics indicate the potential to induce outages ranging from localized disruptions to large-scale cascading failures [26,27]. Ultimately, internal system perturbations likewise originate from the impact of external extreme events. Both external and internal adversities have a significant impact on resilience perturbation. Above all, the adversities can be further specified as extreme events, which can be categorized according to the occurrence probability and consequence severity. They are divided into high-impact events, severe events, low-impact and high-probability events, and high-impact and low-probability events [24,28]. As for the IESs’ resilience, high-impact events should draw primary attention regardless of how often they occur. The way to quantify and identify resilience caused by these sorts of events is different.

Different types of adversities enable the system to exhibit distinct functionality states (different responses) throughout the resilience process. To this end, it is necessary to clarify their typology and dissect their specific relationships with the four phases of the resilience process [7,10]. Some of the extreme events seem to be acute, which happen abruptly with a well-defined onset time, brief duration, and limited impact influences, such as accidents or sabotage. While some are deemed as chronic adversities, with a substantial period. Based on whether a clear starting point exists, chronic adversities can be further categorized into distal onset and proximal onset. The former generally lacks a clear starting point, and its trigger is always a long-term mismatch between supply and demand, leading to functional or infrastructural incompatibility. On the contrary, the latter tends to have a clear start point (e.g., natural disaster, war, or extreme weather) [7]. It can be concluded that the distal-onset adversities hardly return to a “before” operative level. While the acute and proximal-onset adversities need a similar resilient recovery coping mechanism to help the system rapidly restore to normal operating conditions. The resilience process and resilient recovery coping mechanism should be discussed separately. For instance, the growing mismatch between infrastructure capacity and evolving energy production, exemplified by prolonged heatwaves that increase cooling loads, is a typical distal-onset adversity. In contrast, critical infrastructure damage caused by floods, earthquakes, or wars falls into the proximal-onset category.

To sum up, the resilience of an IES essentially denotes its capacity to activate emergency response mechanisms against perturbations induced by high-impact adversities, such as extreme weather, natural disasters, equipment malfunctions, and human errors. Based on the proposed definition, this capability is not an isolated performance, such as reliability, robustness, flexibility, or stability. It is more likely a comprehensive capability encompassing the aforementioned performance attributes that enables the system to swiftly revert to stable and reliable operational states [28].

4. Resilience Identification and Quantification

Resilience of IESs essentially refers to a system’s adoption of an emergency mechanism to cope with perturbations arising from extreme weather, natural disasters, equipment failures, and human operational errors, thereby restoring the system to a state of stable and reliable operation [28]. This mechanism encompasses the modeling of perturbations, their impact propagation through IESs, and the coping strategies employed for resilience enhancement, as is illustrated in Figure 4. Resilience identification and quantification of IESs demand more systematic and transparent assessment methods and categorization schemes. Drawing upon the reviewed papers, the following essential tools are the keys for this purpose: (1) the assessment framework of IESs, (2) the assessment metrics of IESs, (3) the modeling techniques for the resilience of IESs [15].

4.1. Assessment Framework of IESs

Serving as evaluation roadmaps, frameworks provide the structural layers, functional components, and procedural guidelines necessary for implementing and assessing resilience levels [29].

Some research condenses generic resilience assessment frameworks into typical operational steps. The generic conceptual framework raised from social-ecological systems was proposed as six key steps [30]: (i) defining and understanding the system, (ii) identifying the appropriate scale to evaluate resilience, (iii) identifying the system drivers and external and internal disturbance, (iv) identifying the key players in the system, (v) developing models for identifying necessary recovery activities, and (vi) proposing specific measures for recovery. The common evaluation steps for energy system resilience should include metric definitions, threat characterizations, scenario specifications, proactive management, system degradation definitions, recovery processes, and resilience evaluations [22]. Another framework focuses on system information (characteristics, topology, and operational constraints), vulnerability analysis, and resilience operation (system response, damage tolerance, and recovery) [31]. A four-part [32] and a five-part [33] approach was proposed, of which the recognition of extreme events, resilience metrics selection, and system performance evaluation are the critical translation procedures. Above all, the existing frameworks seem to be more maturely applied in the field of power systems. Given that resilience research of IESs is still in its infancy, there is still no consensus on a standardized framework for resilience quantification.

Irrespective of the specific framework employed, a cornerstone step involves developing an apt methodology or establishing a robust modeling framework to evaluate system resilience dynamics, which encompasses the entire process that a system deviates from nominal performance and returns to the nominal operating state. The ways to measure the states should be able to reflect uncertainty, consider recovery time, and be useful for following action, thus they generally fall into two categories: qualitative and quantitative [18,34,35,36]. Qualitative methods apply the latent drivers of a problem, its causes and motivations, rather than numerical values, formulas, or models, to assess the system’s resilience. This kind of method tries to elucidate resilience conditions based on the driving factors and correlated characteristics, which may require more information about operational states [15]. They typically function as an initial step toward detailed quantitative analysis [15,34,35,36]. Sometimes it tends to be semi-quantitative methods, which combine qualitative analysis with mathematical formulations [37]. Quantitative approaches employ performance measures to capture performance deviations, quantify damage extent, and assess function restoration states. The quantitative approaches can be further categorized into semi-quantitative, deterministic, and probabilistic methods. As the name implies, semi-quantitative methods refer to the methods that are not fully based on quantitative indicators, but partially rely on qualitative means, which may be surveys, expert judgment, and fuzzy logic methods. Deterministic methods are more precise for given sets of input, while probabilistic methods employ uncertainty methods for analysis. Specific prevalent methodologies include analytical, simulation-based methods and statistical methods. The systematic design and analysis framework for resilience is shown in Figure 5.

Considering its inherent limitations, the qualitative methods are widely applied in different resilience research domains in the early stages. Whereas quantitative approaches are more popular in the resilience assessment of IESs, having recently provided moderate accuracy. This trend can be clearly observed from

Table 4. The summary framework of existing papers according to the adversities’ type, dealing with various adversities with different characteristics.

Adversities’ Type	Characteristics	Framework Types	Reference No.
Not specified	-------	Qualitative: In-depth review	[40]
Not specified	-------	Qualitative: Semi-quantitative	[37]
Proximal-onset: High-impact rare events (Acute)	Moderate impact; Rapid recovery with targeted measures	Quantitative: Semi-quantitative	[41,42]
		Quantitative: Deterministic methods	[22,41,42,43,44,45,46,47,48]
		Quantitative: Probabilistic methods	[49,50]
Distal-onset events	Insidious, enduring, and deeply impactful; long-term and sustained adaptation measures	Quantitative: Deterministic methods	[51]

. Notably, deterministic methods represent the most widely adopted approach among these alternatives, and their application in power system resilience assessments has been relatively well-established. It is noteworthy that data-driven methods, such as the deep reinforcement learning method (DRL), have garnered significant attention, which should also be increasingly applied to the resilience evaluation of IESs. Among them, DL is extensively employed in load forecasting, equipment state prediction, renewable energy generation forecasting, and related tasks. While the optimal scheduling of the system can then be achieved through technologies such as ARIMA-DRL. DRL approaches address both external and internal disturbances. State-adversarial DRL (SA-DRL) has been applied to counter cyber-attacks and enhance system resilience [38], while other DRL-based strategies minimize disaster risks. For example, a transfer learning-enhanced hybrid convolutional neural network–gated recurrent unit (CNN-GRU) model estimates dynamic security limits under varying weather conditions [39].

Sometimes, ambiguous cases may arise in practical applications. For instance, when applying the ARIMA-DRL to the resilience analysis of power systems, this method could be regarded as both a probabilistic and data-driven method. Rather than clearly determining the exact method type, one only needs to confirm the method’s feasibility and the reliability of its specific operations.

In-depth research explicitly categorizing perturbations as distal or proximal is notably absent from the existing literature. On the basis of the resilience assessment framework and detailed methods, figuring out adversities’ characteristics and their influence on the system are the prerequisites for resilience evaluation and enhancement. Consequently, this paper tries to figure out the influences of distal and proximal-onset adversities on the resilience of IESs, separately. The retrieved papers were re-examined, and the results are demonstrated in Table 4.

Resilience induced by proximal and distal-onset extreme events is outlined in Table 4. It clearly demonstrates that the characteristics of the two kinds of events are notably different. The effects of proximal-onset extreme events are various, which may be infrastructure damage, as well as supply and demand mismatch, whereas the distal-onset ones collectively manifest as a stage that increases the infrastructural mismatch that has emerged. Namely, distal-onset chronic diversities are insidious, enduring, and deeply impactful due to changed energy production/consumption. While proximal-onset ones seem to be “moderate”. This, in turn, activates different types of response mechanisms (Figure 6). For distal-onset chronic diversities, long-term and sustained adaptation measures should be implemented. While proximal-onset events have a rapid recovery ability that can be achieved through targeted measures. The specific evaluation methods adopted in the obtained literature show no significant correlation with extreme event types; however, a pronounced inclination toward quantitative methods is evident.

4.1.1. Resilience Assessment Metrics

Resilience assessment metrics are a set or series of indicators that are able to measure specific properties linked with system resilience. They are typically used to identify weakness points of the system and plan preventive measures to attenuate adverse impacts [15]. Common features for resilience metrics are summarized as: (i) covering aspects with a reflection of systems’ recovery abilities, such as adaptive capacity, recoverability, and stability [24]; (ii) reflecting systems’ uncertainties [52]; (iii) prioritizing high-impact low-frequency events and their consequences [30]; and (iv) exhibiting spatiotemporal specificity, methodological coherence, and performance orientation [53]. Some researchers tried to propose resilience metrics according to the system itself [54], while others attempt to consider a detailed resilience process [55]. The former type of metrics is calculated in accordance with system performance, the data of which describes system’s operation states, particularly the fluctuating operational states of the system, for instance, energy source diversity. The latter metric typology concerns particular stages or partial aspects of resilience process. This kind of metric tries to depict the resilience by performance curves, where separate quantitative methods are jointly applied to detect the system capabilities [10,55].

This paper reviewed the papers based on different types of evaluation methods and adverse events. The results are displayed in

Table 5. Resilience assessment metrics found in the reviewed papers (the specific names vary across the different literature, and the main indicators summarized in the table are generalized based on the similarity of evaluation parameters found in the literature).

Metric	Metric Dimensions and Category	Evaluation Method	Strength and Limitations	Reference
Load shedding (Load curtailments)	Optimization-based	Probabilistic models	(i) Not suitable for HILF events; (ii) Need simplifying assumptions for easier computation	[56,57,58]
System recovery time (Recovery duration, recovery speed)	Reliability-based	Probabilistic models/deterministic methods	Neglect temporal dynamics and emergent resilience properties	[59,60,61]
Total functional service losses	Trend-based	Probabilistic models/deterministic methods	(i) Performance curves must be simplified (e.g., via trapezoidal modeling), which can result in computational errors. (ii) Absence of standardized metrics constrains comparative analysis	[45]
Fraction (Spare capacity factor)				[62]
Duration (Resilient operating time, time until failure)				[63]
System redundancy	Performance-based	Deterministic methods	More informative	[1]
Performance loss			Require data granularity and event modeling capabilities	[22,64]
Performance function			Inadequate for real-time dynamics under stress	[65]

.

Based on the literature summarized in Table 4 and Table 5, Figure 7 presents a parallel categorical visualization mapping the relationships between method category, detail method, scope, adversity types, framework types, and applied metrics. The segment size reflects the number of studies addressing each category, while color coding differentiates the resilience assessment metrics and framework types. It intuitively reveals interrelationships and shared characteristics of relevant review studies across dimensions, including research methodology, research subjects, and evaluation metrics.

From Table 5, retrieved resilience metrics cover eight aspects to varying extents, from economic costs to system-wide responsive capacities. Based on applied methods, the metrics are outlined, encompassing attribute-based and performance-based metrics. Attribute-based metrics elucidate the determinants of system resilience relative to its baseline state, which can be further divided into trend-based, reliability-based, and optimization-based metrics. For instance, key system attributes—including robustness, adaptability, resourcefulness, and recoverability—can be systematically quantified through such metrics [66]. Although parameters specific to each metric category differ and vary with the analytical scenario, no consensus has yet been reached [67]. As consistently recommended by numerous well-established studies, resilience assessment metrics should obey the following criteria [30,52,53]: (i) have a particular focus on high-impact, low-probability (HILP) events; (ii) be performance-based; (iii) reflect true intrinsic uncertainties; and (iv) be simple. From Table 5 and Figure 7, it is obvious that performance-based metrics are widely adopted, especially in the context of electric power systems.

4.1.2. Resilience Phases

As mentioned before, resilience comprises four sequential stages: preparedness, mitigation, response, and recovery. Prior to the exposure of the system to adverse perturbations, IESs operated in normal conditions and maintained preparedness for adverse events. This is the preparedness phase. Systems’ general operation states can be calculated according to historical data for the preparation to face upcoming shocks. Furthermore, strategies such as adding parallel units [68], adding storage modules [69], and using mobile technology to enhance system resilience is designed accordingly in this phase [70]. Subsequently, systems come to the mitigation phase with an obvious performance degradation; during this stage, resilience characteristics can be identified through system performance [71] and impact magnitude is contingent upon the extent of disruption absorption [35]. The ability to sustain the adverse impact of system is governed according to its resilience absorptive capacity in the face of disruptions. Then it comes to the response stage, which is also an adaptive stage. At this stage, system temporarily preserves stability, mobilizing supply- and demand-side resources to expedite recovery to normal operating conditions. The final is the recovery stage. Some researchers further subdivided it into recovery and restoration phases [10]. After the restoration actions, it may achieve full (100%) or partial recovery of energy system performance [14]. Two critical metrics need to be characterized during this phase: (i) the magnitude of performance degradation (the deviation between target and minimum levels), and (ii) the recovery duration to return to equilibrium [72,73]. The critical resilience process timeline is shown in Figure 8, which demonstrates the progression from inception (T₀) to completion (T₉). Three color lines represent distinct adversities. The blue one displays a general resilient reaction of an energy system, of which the mitigation stage begins at T₂. As for chronical adversities, system degradation happens earlier (T₁), often in the preparedness phase. It takes a while for the system to absorb the shocks; for distal- and proximal-onset events, that period is from T₁ to T₂. Response phases are unified for all conditions, which run over the period from T₃ to T₄. After T₄, the system endeavors to smooth and relieve performance reduction. The endurances of three different theoretical adversities may take a long time, the period intervals of which are [T₄, T₅], [T₄, T₆], [T₄, T₇], separately. After that, the system begins to restore a new balance, and the restoration time begins by implementing various short-term and real-time strategies. During the period after T₇, T₈ and T₉, the system returns to a new balance, indicating that a whole resilience process of the system has been experienced.

Figure 8 illustrates the temporal evolution of energy system performance, proxied by total energy supply, following a disruptive event. Observing the performance curves, temporal heterogeneity in the occurrence of various adversities is evident. Chronic adversities, whether proximal-onset ones or distal-onset ones, may exhibit suboptimal performance relative to pre-disturbance expectations, reflecting a cumulative degradation from sustained adverse conditions. This undermines overall system resilience, as persistent adversities—whether latent or immediate—intensify impacts, accelerate failure, and delay recovery [74].

4.2. Modeling Techniques for the Evaluation of Resilience of IESs

Energy system resilience modeling techniques are commonly organized around an analytical framework with the help of problem formulation and mathematical modeling. The initial step in the framework is to identify and model events and potential failures. Thereafter, resilience levels are quantified via selected metrics derived from the system’s performance under the impact of the events and their associated outcomes. Modeling techniques are the primary evaluation method. Based on the details of employed technologies, common modeling approaches can be classified as simulation modelling, optimization modeling, agent-based modeling, and stochastic modeling [14]. Some studies also find two other kinds of methods: indicator-based and fuzzy logic models. Modeling approaches of the reviewed papers are outlined in Table 6 with the specific application case and features.

Analysis carried out in Table 6 demonstrates the resilience evaluation methods and associated features drawn from reviewed papers. Simulation models are the most comprehensive and widespread methods, thanks to their high accuracy. They are especially suitable for the IESs. With the development of data-driven, deep learning, and AI technologies, future applications of simulation-based methods are promising [44]. Currently, optimization-based methods are the most popular, especially in the research of power system resilience. The reason for that is the early emergence of the topics and the urgent need to enhance energy resilience in power systems. Various models and frameworks have been proposed for this category of methods, which differ in their solution strategy. Some adopt a single-stage approach while others employ a multi-stage pipeline [14,75]. They also differ widely in their data requirements and computational efficiency, which range from minimal to substantial (Table 6). Agent-based methods are another kind of simulation-based method. They support the simulation-based modeling of autonomous, heterogeneous agents, thereby capturing emergent patterns and self-organizing behaviors arising from their interactions [76,77]. Stochastic-based approaches make every effort to capture uncertainties and system failures for resilience evaluation, such as the Bayesian network approach [78] and Monte Carlo simulation [79]. Other sorts of methods, including fuzzy logic models and index-based models, are more appropriate for large-scale analysis, which are infrequently employed in assessing energy resilience due to their computational coarseness.

Based on the detailed analytical methods, the aforementioned modeling approaches can be divided into three categories, which are top-down, bottom-up, and hybrid models [80,81]. Top-down models prioritize macro-economic aggregates over technological specifics, while bottom-up methods concentrate on granular technological details at each energy level, which is evident in specific applications throughout the literature. Also, divergent foci dictate that top-down models need to employ historically derived parameters to explicate aggregate energy system dynamics. The applicability and output of different models diverge contingent upon their features, which include respective computational precision, data demands, methodological approaches, computational duration, and sectoral scope, etc. From Table 6, it is clear that the most frequently applied methods (simulation-based and optimization-based ones) commonly acquire high data and long computational duration with relatively high precision. Moreover, owing to its earlier inception, research on power system resilience has reached a relatively advanced stage of maturity in both analytical methods (simulation, optimization, and stochastic-based methods) and results precision. The methods originating from other sectors, such as household, economy, and industry, are relatively more qualitative in methods (fuzzy logic and indicator-based methods) with lower result precision.

Table 6. The summarization of methods and modeling features based on the retrieved papers. (The symbol “√” denotes the features possessed by the corresponding methods under the checked categories.)

Model Description	Approach	Computational Precision		Data Demands			Computational Duration			Sectoral Scope	Reference
		Low	High	Low	Medium	High	Short	Medium	Long
Simulation models with a correlation matrix	Simulation-based: Bottom-up		√			√			√	Electrical distribution systems	[44]
Dynamic simulation			√		√			√		Integrated energy system (gas, heat, and power sector)	[46]
Performance-based model for characterizing and assessing resilience	Simulation-based: Top-down		√	√			√			Integrated energy system (power plant, heating and cooling systems, distributed generation systems)	[45]
Resilience assessment model and optimal power flow model	Optimization-based: Top-down		√	√			√			Integrated gas and power systems	[41]
Probabilistic modeling approach, including a hurricane hazard model, component fragility models, a power system performance model, and a system restoration model	Optimization-based: Bottom-up		√		√			√		Power system	[75]
Modeling the probability of transmission line failures			√	√			√			Integrated energy system	[48]
ORNL-PSerc-Alaska (OPA) and Crucitti-Latora-Marchiori (CLM) to analyze cascading failures			√		√			√		Power grids	[82]
Dynamic modeling of a power grid as a graph to analyze the cascading effects with a proposed robustness metric			√	√			√			Power grids	[83]
Resilience-driven multi-objective restoration model using mixed-integer programming			√		√		√			Interdependent infrastructure networks	[84]
Methods and techniques for simulating human systems	Agent-based: Top-down	√			√		√			-------	[76]
Modeling based on smart agent communication with sequential Monte Carlo simulation	Agent-based: Bottom-up		√		√			√		Integrated energy systems	[85]
Modeling by MG-based power system architecture		√			√			√		Power system	[86]
Modeling based on a resilience-driven multi-objective restoration model		√		√				√		Power system	[87]
Agent-based electricity market simulation		√			√		√			Power system	[88]
Modeling based on the interdependent critical infrastructure model (ICIM)		√			√		√		√	Power and water system	[89]
Modeling by the stochastic programming method	Stochastic-based: Bottom-up		√			√	√			Power and water system	[90]
Combine a stochastic model with a realistic cascading failure simulator for modeling			√			√	√			Power system	[91]
Integrated, dynamic modeling and simulation frameworks are applied			√			√	√			Power system	[92]
Modeling by a sequential Monte Carlo-based time-series simulation model			√			√	√			Power system	[79]
Fuzzy rule-based method of selecting between alternative infrastructure architectures	Fuzzy logic models	√		√			√			Connected infrastructure system (focus on general system infrastructures)	[93]
Indices from both the system level and the component level	Indicator-based: Top-down		√	√		√				Island city-integrated energy systems (IC-IESs)	[47]
Integrated framework with an energy security metric to assess energy security		√		√				√		Energy system	[94]

Table 6. The summarization of methods and modeling features based on the retrieved papers. (The symbol “√” denotes the features possessed by the corresponding methods under the checked categories.)

Model Description	Approach	Computational Precision		Data Demands			Computational Duration			Sectoral Scope	Reference
		Low	High	Low	Medium	High	Short	Medium	Long
Simulation models with a correlation matrix	Simulation-based: Bottom-up		√			√			√	Electrical distribution systems	[44]
Dynamic simulation			√		√			√		Integrated energy system (gas, heat, and power sector)	[46]
Performance-based model for characterizing and assessing resilience	Simulation-based: Top-down		√	√			√			Integrated energy system (power plant, heating and cooling systems, distributed generation systems)	[45]
Resilience assessment model and optimal power flow model	Optimization-based: Top-down		√	√			√			Integrated gas and power systems	[41]
Probabilistic modeling approach, including a hurricane hazard model, component fragility models, a power system performance model, and a system restoration model	Optimization-based: Bottom-up		√		√			√		Power system	[75]
Modeling the probability of transmission line failures			√	√			√			Integrated energy system	[48]
ORNL-PSerc-Alaska (OPA) and Crucitti-Latora-Marchiori (CLM) to analyze cascading failures			√		√			√		Power grids	[82]
Dynamic modeling of a power grid as a graph to analyze the cascading effects with a proposed robustness metric			√	√			√			Power grids	[83]
Resilience-driven multi-objective restoration model using mixed-integer programming			√		√		√			Interdependent infrastructure networks	[84]
Methods and techniques for simulating human systems	Agent-based: Top-down	√			√		√			-------	[76]
Modeling based on smart agent communication with sequential Monte Carlo simulation	Agent-based: Bottom-up		√		√			√		Integrated energy systems	[85]
Modeling by MG-based power system architecture		√			√			√		Power system	[86]
Modeling based on a resilience-driven multi-objective restoration model		√		√				√		Power system	[87]
Agent-based electricity market simulation		√			√		√			Power system	[88]
Modeling based on the interdependent critical infrastructure model (ICIM)		√			√		√		√	Power and water system	[89]
Modeling by the stochastic programming method	Stochastic-based: Bottom-up		√			√	√			Power and water system	[90]
Combine a stochastic model with a realistic cascading failure simulator for modeling			√			√	√			Power system	[91]
Integrated, dynamic modeling and simulation frameworks are applied			√			√	√			Power system	[92]
Modeling by a sequential Monte Carlo-based time-series simulation model			√			√	√			Power system	[79]
Fuzzy rule-based method of selecting between alternative infrastructure architectures	Fuzzy logic models	√		√			√			Connected infrastructure system (focus on general system infrastructures)	[93]
Indices from both the system level and the component level	Indicator-based: Top-down		√	√		√				Island city-integrated energy systems (IC-IESs)	[47]
Integrated framework with an energy security metric to assess energy security		√		√				√		Energy system	[94]

5. Resilience Enhancement

How to improve IESs’ resilience has been an increasingly critical issue in recent research. Enhancing resilience entails strengthening the IESs’ ability to withstand, recover from, and adapt to the impacts of adversities. The research focus should extend from “pre-event planning” and “predicting and mitigating risks” to encompass “post-event recovery” [1,95]. As a result, IESs’ resilient enhancement concrete work processes should assess failure probabilities under extreme events, identify cascading failure risks, and issue dynamic alerts [96] before, during, and after adversities, as depicted in Figure 9 [97]. The three-stage framework demonstrates a systematic enhancement process for system resilience, enabling critical infrastructure systems to effectively predict (prep, reinforcement), adapt to (adapt, repair, response), and recover from extreme events (recover, reconstruct, restore), thereby ensuring operational safety and reliability.

For the resilience enhancement of IESs, effective coordination between planning and operational dispatch is critical. From the planning perspective, the existing literature has examined resilience enhancement strategies for both individual systems and conceptual frameworks. Methodological approaches for bolstering resilience through multi-agent interconnection (spanning diversified energy sources, multiple regions, and multi-dimensional systems) across IESs of varying scales have occurred. From the operational dispatch perspective, analysis focuses on resilience enhancement measures across four phases of the dynamic process, namely preparedness, mitigation, response, and recovery, encompassing three phases: pre-disturbance preparedness, real-time disturbance response, and post-disturbance recovery. The framework with the specific work that may be involved is demonstrated in Figure 9.

The resilience enhancement strategies can be drawn based on resilience assessment results. Guided by resilience evaluation, energy systems can enhance their performance throughout disruptive events by harnessing resilience characteristics through various enhancement methods. The main enhancement efficacy is to reduce performance degradation and boost restoration. Regarding the specific methodologies employed, the methods can be categorized into two main categories: long-term planning/physical-based models and short-term operational methods [15,97,98]. Long-term strategies indicate physical strategies to improve the systems’ infrastructure reliability, such as distribution systems, deploying sensors, improving network redundancies, and relocating facilities [99]. Whereas, short-term strategies generally necessitate immediate temporary measures amid and after disruptions, such as demand response, microgrids [100], distributed energy [101], risk prediction, and distributed control methods [102].

Detailed approaches to enhancing energy system resilience typically encompass the integration of energy storage with smart grid communication and control infrastructures [98,103,104,105], the utilization of electric vehicles as flexible demand-side resources [69,98,106,107,108,109,110], the establishment of equitable electricity market mechanisms [44,88,111], the deployment of peer-to-peer energy trading architectures [112], and other strategies. A significant mismatch often arises between energy supply availability and user demand patterns when facing the shocks of adversities. As energy storage technologies offer effective means of tackling this issue through improved renewable integration and reduced supply–demand disparities, integration of energy storage as one of the measures has attracted considerable research interest. The most common storage device is batteries [88,103,105], and thermal storage devices are also applied for industrial or livelihood purposes [113]. Electric vehicles are deemed as mobile energy storage devices, which frequently appear in scenarios equipped with energy hubs [10]. The electricity market mechanisms and peer-to-peer energy trading methods are more likely to analyze economic, environmental, and risk factors to improve the energy systems’ reliability [88,111,112].

In tandem with AI technological advances, big data and AI techniques for energy resilience enhancement have progressively garnered academic focus. With respect to enhancing urban energy resilience, Zhou et al. [24,25,108,114,115,116] conducted studies on energy resilience enhancement with big data and AI techniques. The methods for city-scale information modeling included ML-based system design, data communication, and GenAI. The specific application scenarios were clarified, and the correlations between AI and energy resilience were investigated in their studies with potential proposed trade-off solutions. Those achievements demonstrate significant application potential of AI-driven methods in energy system resilience research.

6. Conclusions

This paper presents a broad literature review focusing on IESs’ resilience assessment indicators, frameworks, methodologies, and enhancement strategies. Correlated key points of logical, analytical, and technical methods are systematically summarized and identified based on the retrieved papers. In addition, the conception of IESs’ resilience and its specifying characteristics under challenging conditions, mitigating adverse event consequences, and facilitating recovery, are comprehensively dissected. The research on IES resilience has achieved certain progress, but the way to evaluate and enhance the resilience is still in the early stages. There is no consensus on the definition of IESs’ resilience. The following concrete conclusions are discussed for the sake of the standardized definition, assessment, and enhancement frameworks.

(i)

Resilience is a complex concept involving multiple scenarios, which needs to be clearly distinguished from reliability, robustness, flexibility, and stability. In this paper, an IESs’ definition was attempted to be proposed, which is summarized as follows: IESs’ resilience indicates a system’s capability to adapt to and tolerate shocks arising from unpredictable and high-impact disturbances. It also encompasses its ability to recover to the original or a new stable state following the disturbances.

(ii)

The extreme events causing systems’ resilient process can be categorized based on their occurrence probability and consequence severity. From the perspective of IES resilience, high-impact events should draw primary attention irrespective of their occurrence frequency.

(iii)

Different types of adversities lead to distinct functional states (different responses) of the system during the resilience process. Considering the specific action points of extreme events, the adversities are grouped into acute, distal-onset chronic, and proximal-onset chronic. Chronic adversities can reflect the cumulative degradation from sustained adverse conditions. The impact of such events on system resilience is often hidden, yet it warrants focused attention. Nevertheless, a considerable portion of existing research has yet to address this in a targeted manner.

(iv)

Inspired by well-developed analytical approaches in power systems, an evaluation framework with superior performance should exhibit a critical ability of extreme event recognition, resilience metrics selection, and system performance evaluation.

(v)

Most appropriate resilience assessment metrics should have common features in reflecting systems’ recovery abilities, reflecting systems’ uncertainties, prioritizing different events and their consequences, and exhibiting spatiotemporal specificity. According to review results, a total of eight metrics, including load shedding, system recovery time, and fraction, etc., which are categorized into attribute-based and performance-based, are the most robust and suitable for evaluation purposes.

(vi)

IESs’ resilience evaluation approaches are reviewed from the perspective of their modeling characteristics and methodologies. Key features examined include the computational precision (low or high), data demands (low, medium, high), computational duration (short-term or long-term), and sectoral coverage. The results indicate that quantitative analytical methods (including simulation, optimization, and stochastic approaches) are well-developed and have high-precision results.

(vii)

Key resilience enhancement methods include storage-grid integration, vehicle-to-grid utilization, market mechanism reform, and decentralized trading. Furthermore, the significant application potential of AI-driven methods in energy system resilience research has progressively garnered academic focus.

7. Resilience Research Gaps and Future Research Fields

Resilience is inherently a broad research concept. Originating in economics and sociology, resilience studies have gradually expanded to infrastructure, then to power systems, and finally to energy systems. At present, both the definition of integrated energy system resilience and its associated technologies remain in the early stages. Most technical methods mentioned in this paper are relatively mature in the study of power system resilience. For IESs, resilience evaluation indicators, assessment frameworks, evaluation methods, and enhancement strategies still require further investigation. Given the complexity of IESs and the coupling of heterogeneous energy flows within them, a critical and urgent issue in current resilience research on IESs is how to analyze the specific response processes of key system components and critical energy conversion nodes under extreme conditions, as well as the corresponding system recovery process. Furthermore, with the advancement of big data and artificial intelligence technologies, an in-depth exploration should be done on acquiring the data relevant to resilience assessment and leveraging these emerging technologies to enhance the resilience of IESs.

As a system in the infant stage, the existing limitations across studies are concentrated in three aspects: (i) the concept definition and evaluation framework remain incomplete, (ii) multi-energy coupling and dynamic modeling are insufficient, and (iii) extreme event scenarios are limited.

Considering potential research issues, several themes appear highly promising:

(i)

From the perspective of adversities for IESs, the proper way to comprehensively estimate the adverse influence on each and overall components of IESs should be necessary. More advanced approaches should be employed to conduct risk analysis for the sake of guiding IESs’ resilience management.

(ii)

Given the complexity of IESs, how to more effectively model the system is an important issue to be explored. Optimal methods that have the ability to extract a large amount of information and knowledge hidden behind the complex system, as well as to model the system characteristics, should be developed. Particular attention should be paid to the data-driven and AI technologies.

(iii)

With respect to big data-driven technologies and AI for enhancing IESs’ resilience, potential research directions center on reliable data acquisition in long-term and short-term frameworks, including the construction of data platforms, data scarcity under extreme scenarios, data aggregation scales, and data reliability validation, among others.