A stochastic bilevel model for facility location-protection problem with the most likely interdiction strategy
Introduction
Critical infrastructure facilities in the supply networks, such as power grid, biomedical devices, and post offices, are threatened by the rapidly growing terrorism in recent years. After being interdicted, the losses of the supply network function have happened. To reduce the threat of terrorist attacks, there is a growing interest in locating and/or protecting facilities against imminent attacks. The problem is regarded as facility interdiction problems which have been widely modeled by game theory [1], [2], [3], [4], [5]. However, the existing studies usually consider location and protection decisions separately, and most of them provide solutions based on assumptions that fail to address real-life problems. Such assumptions include (1) facilities are uncapacitated and can provide unlimited services and resources to the demand centers (uncapacitated facility), (2) the attacker is a perfectly rational player who prefers conducting the worst case of interdiction against facilities (perfectly rational attacker), (3) facilities are completely protected if fortified and fail completely once exposed to an interdiction (perfect fortification and interdiction), (4) after an interdiction, facilities are either operational with full functionality or non-operational with the losses of function (binary-state capacity). However, in reality, there are a set of constraints on the capacities of facilities that could limit the services of facilities. And the assumption of perfect rationality would provide non-robust strategies for the defender because the attacker has biases that affect his decision-making processes. Further, the fortification does not provide perfect defense against intentional attacks. Similarly, the interdiction also has no ability to disrupt the facility completely. Thereby, the facilities can be partially operational after interdiction, depending on the fortification level and the total exerted interdiction amount. Thus, the existing studies based on these assumptions are likely to provide inferior decisions about the location/fortification of facilities in response to intentional attacks. In this regard, we relax these assumptions and extend the literature by taking the capacitated facility into account, considering the joint location and fortification decisions, incorporating a boundedly rational attacker. Furthermore, probabilistic fortification and interdiction, multiple levels of fortification, and multiple types of interdiction are allowed to make the uncertainty in the post-interdiction capacity of facilities.
In terms of facility interdiction models, location and protection are two mitigation strategies to decrease the impact of intentional disruption risks. Location strategy is implemented in the design phase of the supply networks. O’Hanley and Church [2] were one of the pioneers who formulated a bilevel programming to decide where facilities to locate for hedging against worst-case facility losses. Meng et al. [6] proposed a mixed integer programming to determine the optimal location of terror response facilities when subject to both unintentional and intentional disruptions. This work was extended by Li et al. [7] where hidden information about facilities is taken into account. Ghaffarinasab and Motallebzadeh [8] investigated three different variants of hub interdiction problem in a multiple allocation hub-and-spoke, which was further extended by Ullmert et al. [9]. Xiang and Wei [10] studied an integrated emergency facility location and network interdiction problem where the attacker is allowed to use an interdicted road segment.
The concept of protection strategy allocates resources across targets to fortify against terrorism threats [11]. Church and Scaparra [12] first introduced the r-interdiction median problem with fortification to optimally allocate defensive resources to mitigate disruption risks stemming from the attack. According to this fortification model, extensive previous efforts such as uncertain number of attacks [13], probabilistic protection [14] and defensive budget constraint [15] were developed. Moreover, multiple allocation hub interdiction problems in hub-and-spoke networks have been conducted [16], [17], [18]. Khanduzi and Maleki [19] offered a multi-period interdiction problem with fortification that involves a dynamic Stackelberg game based on capacity recovery. In each period, the defender provides services to customers and allocates fortification resources to facilities that are exposed to possible interdiction exerted by the attacker. Parajuli et al. [20] formulated a tri-level mixed integer programming model for the protection problem that seeks to secure and backup facilities with appropriate capacity levels and response speeds under disruptions. A thread of literature that concerns on protection strategy at different systems consist of literature [21], [22], [23], [24], [25].
Note that, all of the above-mentioned models deal with the protection and location strategies separately that fail to provide reliable decisions against attacks [26]. Thus, the joint consideration of the facility location and protection strategies is necessary. Mahmoodjanloo et al. [27] and Akbari-Jafarabadi et al. [28] proposed an r-interdiction median problem combining location and protection decisions, in which complete interdiction and probabilistic protection were modeled. Jalali et al. [29] formulated a convex risk-averse location-protection problem that a risk measure is incorporated into the decision-making. Zhang et al. [30] presented a multi-objective optimization for location and protection of facilities simultaneously, where secrecy is used into a simultaneous game so that each player moves without knowing their adversary’s strategies. In these location and/or protection models, the uncapacitated facility has been widely investigated. Although there exist some studies on the capacitated facility [19], [27], [28], [31], most of them assumed that the capacity of facilities could be decreased to zero when exposed to an interdiction, i.e., the interdiction is perfect. Despite Hausken and Levitin [32] have investigated multi-state system survivability theory, each element in the system is still assumed to have two states: perfect functioning and total failure. However, the element may have another state, such as partial operation. Thus, we assume that the capacitated facility can remain partially operable after interdiction, depending on the fortification level and the total exerted interdiction amount. In this regard, we allow probabilistic fortification and interdiction to make uncertainty in the post-interdiction capacities of facilities for improving the robustness of the model.
According to the literature above, a defender–attacker sequential game has been widely applied in the field of the facility interdiction problem. However, all of these game models assume that the attacker is perfectly rational and would always choose a pure strategy to maximize his expected utility or maximize the defender’s loss. In reality, the attacker’s behavior would be influenced by many factors, such as the degree of rationality, risk preference, and intelligence information [33], [34]. Thus, within the defender–attacker game, the behavioral model is worth studying. Nikoofal and Gumüs̈ [35] studied a game-theoretic model by considering information asymmetry and discussed how the attacker’s degree of rationality and target preference affect defense equilibrium. Zhang et al. [36] used a quantal response model to model the attacker’s bounded rationality for multi-target defensive resource allocation in defender–attacker game. Cheung and Bell [37] also used the concept of quantal response to allow for biases when defining attack probabilities in defending critical assets. Zhang et al. [38] studied the risk preferences of both defender and attacker in a sequential game. They analyzed the effect of risk preferences on the players’ behavior in equilibrium. Lei et al. [39] utilized conditional value-at-risk to deal with a generic maximum flow interdiction problem, where the performance of solutions by the leader and the follower is measured in the left-tail and right-tail CVaR values, respectively. The risk measure was also used by Jalali et al. [29] accounting for facility location and protection problem under interdiction. There is little literature applying the behavioral model within the defender–attacker game to facility interdiction problem. In this paper, we introduce a new approach, namely the minimum information (MI) approach inspired by [40], to model the attacker’s decision-making behavior. The prior information is given as the probabilities for attacking each facility by different interdiction types, which may be determined by considering some attributes, including but not limited to the degree of rationality and risk preference, taken into account by multiple attributes decision-making procedures.
As described by Hausken and Levitin [41], systems defense and attack models were methodically classified according to the system structure, defense measures, and attack tactics and circumstances. Theoretically, within the classifications proposed in Hausken and Levitin [41], the system structure, defense measures, and attack tactics and circumstances considered in this paper fall under the category of multiple elements, protection, and attack against multiple elements, respectively. Furthermore, by scrutinizing the theoretical literature, this paper makes analyses similar to the analysis in [32], [42], [43], [44]. In detail, this paper is devoted to studying facility interdiction problem jointly considering location and fortification decisions and involving the capacitated facility, probabilistic fortification and interdiction, as well as bounded rationality. The defender decides the location and fortification of facilities. Whereas the attacker is a boundedly rational player who exerts a certain amount of interdiction for established facilities according to his most likely interdiction strategy. Note that the probabilistic fortification and interdiction make the uncertain post-interdiction capacity of facilities constructing scenario set. To this end, we first propose a two-stage stochastic bilevel programming model (TSSBP) with decision-dependent uncertainty, where the first stage is modeled as a bilevel game formulation that the defender and attacker make decisions sequentially at the outer and inner-level, respectively. The defender-level objective function is devoted to minimizing the total cost of locating and fortifying facilities and serving demand centers. While the attacker level problem is modeled by the MI approach to determine the most likely interdiction strategy. After realizing a specif configuration of location, fortification and interdiction strategies, the post-interdiction capacities of the facilities can be obtained, and a set of scenarios is constructed. Each scenario is constructed by a particular combination of the post-interdiction capacities of the facilities. The total number of scenarios is equal to the number of possible combinations of the facilities’ post-interdiction capacities. A scenario-based minimization problem for the user is presented in the second stage of the model. The minimization problem seeks to minimize the network cost by optimally assigning demand centers to facilities bases on existing information.
The resulting TSSBP model is a nonlinear and NP-hard problem, which is acknowledged as a challenging problem to solve. A similar structure of the model that can be found in the literature is [45], where decision-dependent uncertainty is not involved in comparison to our TSSBP model. Thus, the multi-objective solution method proposed in [45] cannot be used in this paper. In this regard, we develop a novel hybrid genetic algorithm that is used to iteratively solve the defender and attacker level problem in the first stage of the model. In detail, the outer-level GA proceeds to obtain the optimal defense strategies. And the inner-level GA is designed to generate an approximation to the attacker’s most likely interdiction strategy, given a defense strategy. Moreover, an analytic hierarchy process(AHP) and a multinomial logistic regression are embedded into the hybrid GA to estimate the probability of interdiction exertion and post-interdiction capacity state, respectively.
Compared with previous literature, this paper has the following significant contributions: (1) We develop facility interdiction problem jointly considering location and protection decisions under the capacitated facility and bounded rationality for practical implications. (2) Involving probabilistic fortification and interdiction makes the decision-dependent uncertainty in the post-interdiction capacity states of the facilities. The probability of post-interdiction capacity states is estimated by a predictive modeling technique. (3) The most likely interdiction strategy of a boundedly rational attacker is determined by the minimum information approach. The probability of exerting interdiction for facilities is estimated by the analytic hierarchy process. (4) A hybrid genetic algorithm is proposed for dealing with the intractable two-stage stochastic bilevel programming model.
For ease of reference, we provide the necessary notation used in this paper in Table 1.
The remainder of this paper is organized as follows. In Section 2, the MI approach for determining the most likely interdiction strategy, decision-dependent scenario probability as well as mathematical programming formulation are stated. In Section 3, a hybrid genetic algorithm is introduced to solve the TSSBP model. Moreover, the probabilities of interdiction exertion and capacity states are estimated. In Section 4, several computational experiments are provided for the purpose of illustration. In the end, Section 5 summarizes the paper and presents some possible works for future research.
Section snippets
Problem description and mathematical formulation
The supply network is represented with a set of potential facilities transferring requirements to a set of demand centers (i.e., user) . Every facility is established with a finite supply capacity and fortified at a certain level associating with effort criterion . The cost of establishing facility at fortification level is . Taking the attacker’s behavior into account, we must treat the attacker as a boundedly rational player who tries to disrupt the supply network by
Solution methodology
Obviously, the proposed TSSBP model is complex, nonlinear and NP-hard. In this regard, we develop a hybrid genetic algorithm(hybrid GA) for solving the TSSBP model, in which the analytic hierarchy process and predictive modeling technique are implemented to estimate the probability of interdiction exertion and capacity states, respectively. The similar framework of hybrid GA algorithm has been effectively applied in [32], [49].
Computational experiments
In this section, numerical experiments are generated to illustrate the performance of the proposed TSSBP model. Considering that there are no benchmark problems for the proposed model in the literature, some experiments have been conducted randomly using the approach introduced by Aksen et al. [31]. The generated numerical experiments with random parameter values are shown in Table 2. For each value of , two random examples are conduced in any size. All of them are numbered as they are given
Conclusions
This paper researches a capacitated facility location-protection problem under terrorist attacks in which the attacker is boundedly rational, and probabilistic fortification and interdiction are involved to make the decision-dependent uncertainty in the post-interdiction capacities of the facilities. The problem is formulated as a two-stage stochastic bilevel model. The defender and attacker make decisions sequentially in the first stage of the model. The second stage determines the optimal
CRediT authorship contribution statement
Qing Li: Conceptualization, Methodology, Supervision, Writing – original draft, Writing – review & editing. Mingchu Li: Software, Resources, Project administration, Funding acquisition, Supervision. Runfa Zhang: Data curation, Visualization, Formal analysis. Jianyuan Gan: Validation, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors gratefully acknowledge the support from the National Science Foundation of China under grant no. 61877007, the Fundamental Research Funds for the Central Universities, China under no. DUT20GJ205.
References (61)
- et al.
Designing robust coverage networks to hedge against worst-case facility losses
European J Oper Res
(2011) - et al.
A bilevel fixed charge location model for facilities under imminent attack
Comput Oper Res
(2012) - et al.
Reliable facility location design under disruptions
Comput Oper Res
(2013) - et al.
Metaheuristics for protecting critical components in a service system
Expert Syst Appl
(2016) - et al.
On the p-hub interdiction problem
Comput Oper Res
(2020) - et al.
Joint optimizing network interdiction and emergency facility location in terrorist attacks
Comput Ind Eng
(2020) - et al.
Analysis of facility protection strategies against an uncertain number of attacks: The stochastic r-interdiction median problem with fortification
Comput Oper Res
(2011) - et al.
The r-interdiction median problem with probabilistic protection and its solution algorithm
Comput Oper Res
(2013) - et al.
Multiple allocation hub interdiction and protection problems: Model formulations and solution approaches
European J Oper Res
(2018) - et al.
A branch-and-cut algorithm for the multiple allocation r-hub interdiction median problem with fortification
Expert Syst Appl
(2018)
The impact of congestion on protection decisions in supply networks under disruptions
Transp Res Part E-Logistics and Transp Rev
Metaheuristics for protecting critical components in a service system: A computational study
Expert Syst Appl
A trilevel r-interdiction selective multi-depot vehicle routing problem with depot protection
Comput Oper Res
Responsive contingency planning of capacitated supply networks under disruption risks
Transp Res Part E-Logistics and Transp Rev
A fast genetic algorithm for a critical protection problem in biomedical supply chain networks
Appl Soft Comput
A bi-objective partial interdiction problem considering different defensive systems with capacity expansion of facilities under imminent attacks
Appl Soft Comput
A tri-level covering fortification model for facility protection against disturbance in r-interdiction median problem
Comput Ind Eng
A tri-level r-interdiction median model for a facility location problem under imminent attack
Comput Ind Eng
Locating and protecting facilities from intentional attacks using secrecy
Reliab Eng Syst Saf
A bilevel partial interdiction problem with capacitated facilities and demand outsourcing
Comput Oper Res
Minmax defense strategy for complex multi-state systems
Reliab Eng Syst Saf
Risk-attitude-based defense strategy considering proactive strike, preventive strike and imperfect false targets
Reliab Eng Syst Saf
Modeling multi-target defender-attacker games with quantal response attack strategies
Reliab Eng Syst Saf
Attacker–defender model against quantal response adversaries for cyber security in logistics management: An introductory study
European J Oper Res
The role of risk preferences in a multi-target defender-attacker resource allocation game
Reliab Eng Syst Saf
Stochastic maximum flow interdiction problems under heterogeneous risk preferences
Comput Oper Res
A bi-level capacitated p-median facility location problem with the most likely allocation solution
Transp Res B
Defense and attack of complex and dependent systems
Reliab Eng Syst Saf
A set of efficient heuristics and metaheuristics to solve a two-stage stochastic bi-level decision-making model for the distribution network problem
Comput Ind Eng
A minimum information principle : Theory and practice
Reg Sci Urban Econ
Cited by (11)
Metaheuristics for bilevel optimization: A comprehensive review
2024, Computers and Operations ResearchMinimizing the maximal reliable path with a nodal interdiction model considering resource sharing
2023, Reliability Engineering and System SafetyA risk-averse tri-level stochastic model for locating and recovering facilities against attacks in an uncertain environment
2023, Reliability Engineering and System SafetyCitation Excerpt :On the other hand, the recovery is a post-disruption mitigation strategy, focusing on recovering these disrupted facilities via reactive measures. Both of these strategies provide good risk mitigation methods, and some researchers have highlighted the necessity of implementing preventive or recovery strategies to stabilize the supply networks [9–14]. Bao et al. [15] further demonstrated that integrating preventive strategy with the disruption recovery model can efficiently enhance network reliability.
The leader multipurpose shopping location problem
2022, European Journal of Operational ResearchCitation Excerpt :Bi-level programming has been extensively researched (see, e.g., Wen and Hsu, 1991; Vicente and Calamai, 1994; Colson et al., 2005 and 2007; Angelo and Barbosa, 2015; Caramia & Mari, 2015; Sinha et al., 2018; Kalashnikov, Dempe, Pérez-Valdés, Kalashnykova & Camacho-Vallejo, 2015) and also applied to location problems. Some recent examples are Casas-Ramirez, Camacho-Vallejo, Diaz & Luna (2020) for the maximal covering problem; Labbé, Leal and Puerto (2019) for controversial facilities; Alekseeva & Kochetov (2013) for the discrete (r|p)-centroid problem; Aksen, Aras & Piyade (2013), Ushakov, Klimentova & Vasilyev (2018) and Abareseshi & Zaferanieh (2019) for p-median type problems; Calvete, Galé, Iranzo, Camacho-Vallejo & Casas-Ramirez (2020) for a facility location problem with cardinality constrains and preferences, and Li, Li, Zhang & Gan (2021) for a stochastic facility location-protection problem. The hub location problem is closely related to the facility location problem and includes routing decisions.
Locating and protecting interdependent facilities to hedge against multiple non-cooperative limited choice attackers
2022, Reliability Engineering and System SafetyCitation Excerpt :In their models, the location and protection strategies as two effective defensive strategies are considered separately, which fails to provide solutions for mitigating disruption risks. Following [12], we incorporate the joint location and protection decisions in the facility interdiction problem to make the supply network efficient and reliable. Besides, from the perspective of the attacker, a common assumption is that the attacker allocates his attack resources among all potential targets in response to the defensive strategies.
Nonlinear optimization in bi-level selective maintenance allocation problem
2022, Journal of King Saud University - ScienceCitation Excerpt :Most recently, Ghasemi et al. (2021) formulated a bi-level mathematical model in response to the COVID-19 pandemic for logistic management considering the evolutionary game with environmental feedback. Similarly, Li et al. (2021) proposed a stochastic bi-level model for the facility location and protection problem having a probabilistic interdiction strategy. Candler et al. (1982) formulated a Stackelberg game as a linear two-level programming problem where the first player optimizes his objective function.