AI-Assisted Regional Collaborative Game of an Emergency Supplies Reserve Supply Chain

Abstract

This study is devoted to the analysis of regional collaboration in emergency supply chain reserves. To address this critical research issue, we have developed an AI-assisted tripartite evolutionary game model involving governments, manufacturers, and suppliers across different regions under demand uncertainties and resource disparities. In this study, we employ replicator dynamic equations to derive strategic evolution paths and utilize numerical simulations enhanced by AI-powered global sensitivity analysis for subsequent parameter sensitivity analysis, enabling a systematic examination of equilibrium conditions and stability strategies. Our research findings demonstrate that when government incentive mechanisms provide greater benefits than speculative gains then supply chain enterprises evolve toward collaborative strategies, with the system achieving optimal stability at the equilibrium where collaboration benefits outweigh costs. Our AI-enhanced analysis results also reveal that while higher subsidies accelerate enterprise participation, they may reduce government motivation, necessitating carefully balanced penalty scales to sustain long-term cooperation—findings validated through sensitivity analyses of key parameters. The study’s integration of game theory with AI techniques offers both theoretical innovation in multi-agent decision modeling and practical value for strengthening national emergency management frameworks.

1. Introduction

A series of unprecedented emergencies, spanning pandemics due to SARS and COVID-19, major geopolitical conflicts, and intensifying climate-related disasters, have posed severe and continuous threats to human society. Based on a report from the International Emergency Disaster Database (IEDDB), the world has experienced more than 22,000 major emergencies, resulting in trillions of dollars in direct economic losses. The increasing frequency and severity of public emergencies—such as sudden-onset natural disasters, pandemics and public health crises, large-scale accidents, and social security incidents—have underscored the critical importance of robust and responsive emergency supply chains. Emergency supplies are essential for responding to such emergencies. They also cover a wide range of tools, items, equipment, and devices used in the process of prevention and rescue. Effective reserve and distribution systems for emergency supplies are essential to mitigate risks, save lives, and maintain social stability. However, traditional models of emergency supplies reserves often operate in a fragmented and regionally isolated manner, leading to inefficiencies, resource imbalances, and delayed responses when cross-regional coordination is required. In this context, regional collaboration has emerged as a vital strategy to enhance the resilience and adaptability of emergency supply networks, particularly in contexts characterized by demand uncertainty and uneven resource endowments.

Regional collaboration in emergency supply reserves operates on the principle of optimizing resource allocation across administrative boundaries through coordinated planning, shared information, and joint response mechanisms. By integrating government policies, enterprise participation, and cross-regional coordination, this approach ensures efficient distribution of emergency supplies during crises, minimizing redundancy in well-resourced areas and addressing shortages in vulnerable regions. The system relies on incentive mechanisms (e.g., subsidies and penalties) to align stakeholders’ interests. In order to overcome geographical barriers and achieve a win-win outcome and sharing, regional collaboration should be taken as a pivotal role. In the modern era, regional collaboration development is crucial for boosting a nation’s overall competitiveness, encouraging the modernization and optimization of its economic structure.

Despite widespread recognition of its importance, the implementation of regional collaborative mechanisms faces significant challenges. Key stakeholders—including governments, manufacturers, and suppliers—often operate with divergent interests, information asymmetry, and varying incentive structures. While governments aim to maximize social welfare and public safety, enterprises may prioritize economic gains, creating potential conflicts of interest. Moreover, the dynamic and multi-dimensional nature of inter-regional cooperation complicates the design of sustainable and stable collaborative mechanisms. Traditional analytical approaches have struggled to fully capture the strategic interactions and evolutionary behaviors of multiple agents under uncertain conditions. We can fully utilize each region’s comparative advantages, encourage industry optimization, and upgrade resource utilization efficiency by fostering regional collaboration. Therefore, in order to prevent and respond to potential crises promptly, there is an urgent need to place a greater focus on the development of an effective and powerful emergency supplies reserve system. In light of the background information provided above, in this paper, we examine the regional coordinated reserve of emergency supplies.

To address the above challenges, we developed an AI-assisted tripartite evolutionary game model involving governments, manufacturers, and suppliers across different regions under demand uncertainty and resource disparities. Moreover, we also explored the strategies of regional collaboration under different circumstances. The key theoretical contribution lies in the following: (i) developing a novel AI-assisted tripartite evolutionary game model that uniquely integrates regional collaboration dynamics into emergency supply chain reserves; (ii) establishing a systematic linkage between government incentive mechanisms and enterprise collaboration behaviors through rigorous game-theoretic modeling, supported by AI-enhanced sensitivity analysis of critical parameters; and (iii) by focusing on concepts such as subsidy allocation coefficients and region-specific penalty scales within an evolutionary game context, we offer a new analytical lens that bridges the traditional disconnect between macroeconomic policy instruments and micro-level supply chain decision-making in crisis situations. From a practical standpoint, the findings presented in this study help to improve the efficiency and effectiveness of emergency supply reserves by providing governments with direct practical guidance for developing effective and scientific emergency supply policies and assisting businesses in making wise supply decisions.

The remainder of this paper is organized as follows: In Section 2, we present a brief literature review; in Section 3, we focus on game model construction; in Section 4, we provide a calculated case analysis; lastly, the conclusions are presented in Section 5.

2. Literature Review

The aim of this study is to investigate AI-assisted regional collaboration in the emergency supplies reserve supply chain. Considering related and similar works, we divided the current section into the following three subsections: (i) literature related to emergency supply reserve; (ii) literature related to emergency supplies supply chain collaboration; and (iii) literature related to the role of AI in supply chain management.

2.1. Emergency Supply Reserve Mechanisms

The issue of emergency supply reserve has drawn increasing attention from researchers over recent decades. Liu et al. (2016) examined the negotiation principles of government requisitioning of emergency supplies in the event of a disaster. They analyzed the relevant influencing factors and guidelines to which the government should adhere when negotiating during the requisitioning process to guarantee efficient acquisition [1]. In order to guarantee the efficacy and availability of medical supplies during emergency responses, Zhou and Olsen (2017) investigated the issue of medical supply inventory rotation. The authors analyzed their results and provided suggestions on how to optimize medical supply inventory rotation strategies [2]. Liu et al. (2022) proposed an emergency reserve strategy using an option contract mechanism, which lessens the pressure on the nation’s reserves while also promoting the social implementation of emergency reserve and improving the cooperative relationship between the government and businesses [3]. By examining strategic interactions between participating entities and evolutionarily stable strategies, Wang (2024) developed a tripartite evolutionary game model to investigate methods to improve the quality of emergency supplies [4], allowing them to uncover the internal mechanisms and influencing factors. Huang et al. (2024) constructed a new two-stage stochastic optimization methodology to address contingency REM capacity planning and dynamic inventory operation problems while improving the resilience of energy-constrained supply chain management for perishable products [5]. The authors of existing studies emphasize vertical coordination (e.g., government–enterprise contracts) or single-region contexts but overlook horizontal cross-regional collaboration among upstream and downstream supply chain actors. This gap is critical given regional disparities in reserve capacities and resource endowments, which we address in our study by modeling inter-regional manufacturer-supplier coordination.

2.2. Emergency Supply Chain Collaboration

Regarding emergency supplies supply chain collaboration, studies featured in the relevant literature mostly focus on supply chain collaboration mechanisms and the collaboration effect. In the literature review by Wankmüller and Reiner (2020), the authors noted significant scholarly activity related to supply chain coordination, cooperation, and collaboration between 1996 and 2019 [6]. Kadiyala et al. (2020) examined the inventory management problems faced by upstream suppliers (e.g., Vendor Managed Inventory (VMI)) that have entered into co-operation agreements with retailers and proposed a dynamic inventory mechanism for suppliers to efficiently manage their inventories and information in the supply chain, which significantly improves the expected returns to the suppliers [7]. Zhang et al. (2024) proposed optimization strategies to improve the efficiency of rescue cooperation. Their bi-objective robust optimization study, which considers secondary disasters, illustrates the distribution of these strategies [8]. Xia et al. (2025) investigated the problem of scheduling emergency medical supplies. The authors focused on public health emergencies, and algorithms were designed based on artificial intelligence techniques to optimize the scheduling arrangement of emergency medical supplies [9]. Adhikari et al. (2025) examined how a multi-layered AI-enabled healthcare supply chain collaborates and coordinates during disaster situations; they also investigated how to implement the appropriate tactics to increase the healthcare supply chain’s operational effectiveness and responsiveness during emergencies [10]. Zhang (2025) focused on designing collaborative supply chain networks in the face of demand; they used efficient optimization techniques to investigate methods to build supply chain networks to tackle issues induced variations in demand [11]. Through methodical modeling and wise resource allocation, the authors seek to improve supply chain stability and adaptability. Bimpikis et al. (2019) investigated Cournot competition between businesses in networked markets. They built a model to study enterprises’ production decisions while considering the interdependence between their goods and those of other firms through network interactions [12]. Regarding the regional collaborative management of emergency supplies during pandemic outbreaks, Zhang (2023) investigated methods for creating the best funding reserve and inventory policies in emergency medical supplies [13]. Katsaliaki et al. (2024) conducted a systematic literature review analyzing supply chain cooperation through the PRISMA framework and SCOR model, identifying key structures and operational mechanisms while concurrently developing a novel thematic framework to guide academic and practical SCC management [14]. In order to achieve efficient resource allocation and crisis management, in their study, the authors use models and analysis to strike a balance between the demands of pandemic response and the costs of supply reserves. Through the optimization of readiness and reaction processes, the study findings offer a basis for decision-making that ensures public health security. The study authors calculated the best production strategies for enterprises and market equilibrium results, while simultaneously demonstrating the influence of network topology and other factors on competition and market equilibrium. While collaboration mechanisms have been thoroughly researched, the authors of most studies assume symmetric regional development or centralized decision-making. Few models account for demand uncertainty and resource asymmetry across regions, which we incorporate in the present study through an evolutionary game framework with heterogeneous agents.

2.3. AI’s Role in Supply Chain Management

Recent research demonstrates AI’s transformative impact on supply chain management (SCM), with studies highlighting three key dimensions: technological convergence, resilience enhancement, and emerging implementation challenges. Scholars observe increasing integration of AI with blockchain (Charles et al., 2023) and generative technologies (Jackson et al., 2024), enabling advanced capabilities such as real-time transparency and adaptive decision-making [15,16]. Empirical work by Belhadi et al. (2024) and Ivanov (2023) reveals AI’s role in strengthening supply chain resilience through predictive analytics and digital twins, though effectiveness varies by operational context [17,18]. However, adoption barriers persist, including data quality issues (Grover et al., 2022) and ROI uncertainties (Wamba et al., 2024), prompting calls for hybrid human–AI systems and modular deployment strategies [19,20]. The literature collectively underscores AI’s potential to redefine SCM paradigms while emphasizing the need for ethical frameworks and sector-specific solutions to bridge theoretical advancements with practical implementation. AI applications in supply chains remain focused on technical optimization rather than strategic inter-organizational behaviors. In our study, we bridge this gap by using AI-assisted sensitivity analysis to simulate dynamic stakeholder strategies under bounded rationality, directly linking AI capabilities to policy design.

The examination of the current literature reveals that supply chain collaboration and emergency supplies reserve have generated several fruitful discussions both domestically and internationally; in the emergency supplies reserve supply chain, however, collaboration is largely for intergovernmental collaboration or vertical binding for the implementation. Regarding the issue of significant differences in emergency supply reserve levels due to unbalanced regional development, there are still relatively few studies on collaboration between upstream and downstream enterprises in the supply chain located in different regions. Our study is based on the characteristics of the emergency supplies reserve supply chain, including research on the regional collaboration management of emergency supplies, in order to provide corresponding practical strategy suggestions. Unlike prior work focusing primarily on vertical government–enterprise relationships or single-region contexts, we explicitly model cross-regional coordination between manufacturers and suppliers across different administrative areas, capturing the critical challenges of demand uncertainty and resource disparities that characterize real-world emergency scenarios. The framework introduces an innovative equilibrium analysis method that quantifies the stability conditions for collaborative strategies under asymmetric regional development levels, addressing a significant gap in the literature where such spatial and organizational complexities are often oversimplified.

While traditional game theory provides foundational tools for analyzing strategic interactions—such as Nash equilibrium and replicator dynamics—its static frameworks often struggle with three critical challenges in emergency supply chain contexts: (i) real-time adaptation to dynamic regional disparities; (ii) quantification of bounded rationality effects among heterogeneous agents; and (iii) high-dimensional parameter optimization under uncertainty. Using our AI-assisted approach, we aim to address these gaps.

3. Game Model Construction

3.1. Problem Description

The emergency supply reserves are characterized by uncertain demand, the need for rapid response, and inherent weak economic returns. The current reserve system primarily comprises contributions from governments, businesses, markets, households, and non-governmental organizations. Among these, government-held reserves of emergency supplies form the foundational source for prevention initiatives, emergency drills, and initial disaster response. They play a central role in both disaster prevention and relief efforts, with central and local governments acting as the primary entities responsible for maintaining these stocks. Presently, a widespread construction of emergency supply depots is underway across provinces, cities, and counties nationwide. This effort has led to the significant accumulation of local emergency resources and marks the initial establishment of a national emergency supply reserve system. However, this system faces a critical challenge: significant regional disparities in reserve capacities. Empirical data reveals that coastal provinces maintain reserve levels 30–50% higher than those in inland regions, while disaster-prone areas (e.g., the Sichuan earthquake belt) allocate 2–3 times more resources to reserves than more geologically stable regions. These disparities often lead to simultaneous situations of shortage in some areas and surplus or redundant storage in others, highlighting an urgent need for optimized, coordinated reserve strategies. Beyond government reserves, the emergency supply system incorporates multiple complementary models to enhance resilience. Enterprise reserve refers to the practice where operational firms pre-stock a certain percentage of supplies, which can be promptly utilized by the government during disasters, under its guidance, oversight, and policy support. Studies indicate that enterprises in developed regions demonstrate 25% faster response times but also face 20% higher opportunity costs for participating in collaborative reserves, creating complex coordination challenges [21].

Market reserve serves an additional function, as commodities available on the market can swiftly and directly meet emergency demand when government and enterprise reserves are insufficient to handle crises [21]. At the community level, family reserves have gained prominence; for instance, following the COVID-19 pandemic, China promoted household stockpiling to cope with daily life and emergencies. Maintaining appropriate reserves, such as fire extinguishers, drinking water, non-perishable food, basic medications, and hygiene supplies, can significantly mitigate disaster impact during the initial response. Furthermore, non-governmental organizations (NGOs), exemplified by the Red Cross and other private charitable entities in China, play a vital role in the emergency supply reserve ecosystem.

Recognizing the limitations of single-entity reserve models, the Chinese Ministry of Emergency Management has actively promoted a multi-entity cooperative reserve model. A representative example is the government-enterprise joint reserve mode, where the government collaborates with businesses to accumulate supplies through policy support such as cash subsidies and tax exemptions. This approach facilitates resource sharing and leverages mutual advantages, thereby not only overcoming the shortcomings of isolated stockpiling but also improving the overall efficiency and reliability of emergency supply reserves, providing stronger assurance for emergency response operations.

Regional collaboration within the emergency supply chain is examined in this paper. Under the joint reserve mode of government and enterprises, the relevant enterprises store emergency supplies to handle emergencies, and the government offers certain subsidies for the enterprises. However, when considering the various levels of development in the different regions, the reserve capacity gap is large, and each region stores emergency supplies according to its own needs and resources, which will result in the duplication of supplies in a particular location, i.e., there will be a lack of supplies in certain areas and duplication and waste in others. In light of the above, in this study, we developed a government–manufacturer–supplier tripartite game model, assuming that the supply chain’s suppliers and manufacturers are located in different administrative regions [22]. Through regional coordination, each region can work together to create plans for the emergency supplies reserve, share resources, complement one another’s strengths, and assist the emergency supplies reserve supply chain in creating a stable, sustainable, and highly efficient operation mechanism that will provide strong support for emergency responses [23].

3.2. Research Hypothesis and Model Construction

Hypothesis 1.

It is assumed that the government, manufacturer (in region a), and supplier (in region b) are boundedly rational actors. Practice evidence from interregional supply chain conflicts during COVID-19 reveals that their strategic choices reflect real-world trade-offs between regional self-interest and collaborative incentives (Ivanov et al., 2024) [23]. Due to information asymmetry and regional disparities, each party prioritizes its own interests but may occasionally deviate due to social norms or reputational concerns (Simon, 1957) [24]. The manufacturer chooses between “regional collaboration” (probability x) and “no regional collaboration” (1 − x); the supplier chooses between “regional collaboration” (probability y) and “no regional collaboration” (1 − y); and the government chooses between “positive collaboration” (probability z) and “negative collaboration” (1 − z), where x, y, z ∈ [0,1].

Hypothesis 2.

The government’s baseline benefit during emergency operations is R, with a reserve cost C_g. If the government adopts a positive collaboration strategy and observes that manufacturers and suppliers are actively collaborating, it provides a subsidy W to the enterprises, with a subsidy allocation coefficient α(α ∈ [0,1]). The government also gains implicit benefits M (e.g., improved public trust and governance image). If enterprises are found to be non-collaborative while the government is actively collaborating, a fine, F, is imposed. If the government itself chooses non-collaboration, it faces a penalty T from higher-level authorities and suffers a reputational loss Q_g. The government’s implicit benefits (e.g., public trust gains) align with empirical evidence from disaster management cases where proactive governance improved crisis outcomes (e.g., post-earthquake public confidence boosts in Sichuan, China). Subsidies and penalties mirror real-world policy tools, such as China’s intergovernmental accountability mechanisms, where higher-level authorities penalize local inaction (Zhang et al., 2023) [13].

Hypothesis 3.

If supply chain enterprises do not collaborate, they gain idle resource returns V₁ and V₂ but incur average losses Q₁ and Q₂ (e.g., reputational damage from public scrutiny). If one party collaborates while the other does not, the non-collaborator gains a “free-rider” benefit H₁ or H₂. Collaboration entails costs C₁ and C₂ but brings implicit benefits P₁ and P₂ (e.g., enhanced corporate image and brand value) and potential government subsidies. Speculative behavior detected by the government results in a penalty F. Reputational dynamics (e.g., brand value loss from non-collaboration) reflect empirical patterns from corporate crisis responses, such as firms suffering market share declines after perceived inaction during floods (Meena, 2023) [21]. The “free-rider” problem is empirically observed in supply chain partnerships where asymmetric efforts reduce long-term collaboration (Katsaliaki et al., 2024) [14].

Hypothesis 4.

We assume that collaboration is economically viable: government collaboration cost is less than its implicit benefits (C_g < M); enterprise collaboration cost is less than its total gains (C < P); and idle resource returns are lower than collaboration benefits (V < P). Free-rider gains are smaller than non-collaboration losses or subsidies (H < Q, H < W), ensuring that enterprises lean toward collaboration. Government subsidies are smaller than penalties for non-collaboration (W < T), incentivizing active governance. Cost–benefit assumptions are calibrated using empirical data from emergency reserve programs. For example, subsidy thresholds (W) correspond to actual government compensation schemes in regional joint-reserve pilots (e.g., Guangdong’s subsidy benchmarks for medical supply reserves), while penalty magnitudes (F) reflect contractual penalties in government–enterprise agreements (Liu et al., 2022) [3].

The relevant parameter settings of the regional collaboration evolution game model of the emergency supplies reserve supply chain are shown in

Table 1. Model parameters.

Parameter Symbol	Definition
$\alpha$	Government Subsidy Allocation Factor
$C_g$	Government Collaboration Costs
$C_1$	Manufacturer Collaboration Costs
$C_2$	Supplier Collaboration Costs
$R$	Government Basic Benefits
$M$	Government Collaboration Benefits
$V_1$	Return on Investment of Manufacturers’ Idle Resources
$V_2$	Return on Investment of Supplier Idle Resources
$P_1$	Manufacturer Collaboration Benefits
$P_2$	Supplier Collaboration Benefits
$H_1$	Hitchhiking by Manufacturer
$H_2$	Hitchhiking by Supplier
$Q_g$	Losses from Negative Government Collaboration
$Q_1$	Losses from Manufacturer Non-collaboration
$Q_2$	Losses from Supplier Non-collaboration
$W$	Government Subsidies
$T$	Government Negative Collaboration Penalty
$F$	Enterprise Penalties

. The links between the stakeholders and key parameters are shown in Figure 1.

According to the game model assumptions presented above and the setting of parameters, the payment benefit matrix is shown in

Table 2. Matrix of payment benefits.

Manufacturer	Supplier	Government
		$\mathbf{Positive} \mathbf{Collaboration} \left(\mathit{z}\right)$	$\mathbf{Negative} \mathbf{Collaboration} \left(1-\mathit{z}\right)$
Regional collaboration $\left(x\right)$	Regional collaboration $\left(y\right)$	$P_1-C_1+\alpha W$ $P_2-C_2+\left(1-\alpha \right)W$ $R+M-C_g-W$	$P_1-C_1$ $P_2-C_2$ $R-C_g-T-Q_g$
	No regional collaboration $\left(1-y\right)$	$P_1-C_1+\alpha W$ $V_2-Q_2+H_2-F$ $R+F+M-C_g-\alpha W$	$P_1-C_1$ $V_2-Q_2+H_2$ $R-C_g-T-Q_g$
No regional collaboration $\left(1-x\right)$	Regional collaboration $\left(y\right)$	$V_1-Q_1+H_1-F$ $P_2-C_2+\left(1-\alpha \right)W$ $R+F+M-C_g-\left(1-\alpha \right)W$	$V_1-Q_1+H_1$ $P_2-C_2$ $R-C_g-T-Q_g$
	No regional collaboration $\left(1-y\right)$	$V_1-Q_1-F$ $V_2-Q_2-F$ $R+2F+M-C_g$	$V_1-Q_1$ $V_2-Q_2$ $R-C_g-T-Q_g$

.

3.3. Strategic Evolution Stability Analysis of the Participating Subjects

(1) Strategic stability analysis of the manufacturers

The expected returns obtained by the manufacturer choosing the regional collaboration and no regional collaboration strategies are $U_{m1}\mathrm{and}{U}_{m2}$, as shown in Equations (1) and (2), respectively, and the average expected return is U_m (Equation (3)).

$$\begin{array}{c}{U}_{m1}=yz\left[P_1-C_1+\alpha W\right]+y\left(1-z\right)\left[P_1-C_1\right]+\left(1-y\right)z\left[P_1-C_1+\alpha W\right]\\ +\left(1-y\right)\left(1-z\right)\left(P_1-C_1\right)\end{array}$$

(1)

$$\begin{array}{c}Um2=yz\left[ V_1-Q_1+H_1-F\right]+y\left(1-z\right)\left[V_1-Q_1+H_1\right]+\left(1-y\right)z\left[V_1-Q_1-F\right]\\ +\left(1-y\right)\left(1-z\right)\left[V_1-Q_1\right]\end{array}$$

(2)

$$Um=xUm1+\left(1-x\right)Um2$$

(3)

The replicator dynamic equation for the manufacturer’s strategy selection construct is as follows:

$$F\left(x\right)=dx/dt=x\left(U_{m1}-U_m\right)=-x\left(x-1\right)\left(P_1-C_1+Q_1-V_1+zF-H_{1}y+\alpha zW\right)$$

The first order derivative of $F\left(x\right)$ with respect to $x$ and set $K(z)$ function is as follows:

$$dF \left(x\right) /dx = \left(1-2x\right)[P_1-C_1+Q_1-V_1-H_{1}y+z\left(F+\alpha W\right)]$$

$$K\left(z\right)=P_1-C_1+Q_1-V_1-H_{1}y+z\left(F+\alpha W\right)$$

According to the stability theorem for differential equations, when $F\left(x\right) = 0$ and $dF\left(x\right)/dx < 0$, the stabilization point is obtained. Since $\partial K \left(z\right) /\partial z > 0$, $K\left(z\right)$ is an increasing function about $z$.

The manufacturer’s strategy evolution phase diagram is shown in Figure 2. As shown in the first image in Figure 2, when $z=z*=\left({\mathrm{C}}_1-{\mathrm{P}}_1-{\mathrm{Q}}_1+{\mathrm{V}}_1+{\mathrm{H}}_1\mathrm{y}\right)/\left(\mathrm{F}+\mathsf{\alpha }\mathrm{W}\right)$, $dF\left(x\right)/dx = 0$, all x reach an evolutionary steady state; when $z\ne z*$, two specific evolutionary state equilibrium will be obtained, $x=0,x=1$, and then two scenarios are analyzed.

When $z<z*$, $dF\left(x\right)/dx |x=0<0,x=0$ is the manufacturer’s evolutionary stabilization strategy, meaning that, under the influence of government subsidies, fines, and other influences, if the probability of positive regional positive collaboration is lower than $\mathrm{z}*$, the manufacturer firm enterprise will choose the non-regional collaboration strategy.

When $z>z*$, $dF\left(x\right)/dx |x=1<0,x=1$ is the manufacturer’s evolutionary stabilization strategy, that is, the government’s positive regional collaboration strategy pushes manufacturing firms to choose a regional collaboration strategy when it is greater than $z*$.

(2) Strategic stability analysis of suppliers

The expected returns obtained by suppliers choosing the regional collaboration and non-regional collaboration strategies are U_r1 and U_r2, respectively, with an average expected return of U_r. U_r1, U_r2, and U_r, as shown in Equations (4)–(6):

$$\begin{array}{c}{U}_{r1}=xz\left[P_2-C_2+\left(1-\alpha \right)W\right]+x\left(1-z\right)(P_2-C_2)+\left(1-x\right)z\left[P_2-C_2+\left(1-\alpha \right)W\right]\\ \left(1-x\right)\left(1-z\right)\left(P_2-C_2\right)\end{array}$$

(4)

$$\begin{array}{c}{U}_{r2}=xz\left(V_2-Q_2+H_2-F\right)+x\left(1-z\right)(V_2-Q_2+H_2)+\left(1-x\right)z\left(V_2-Q_2-F\right)\\ +\left(1-x\right)\left(1-z\right)(V_2-Q_2)\end{array}$$

(5)

$$U_r=y{U}_{r1}+\left(1-y\right)U_{r2}$$

(6)

The replicator dynamic equation for the supplier strategy selection construct is as follows:

$$F\left(y\right)=dx/dt=y\left(Ur1-Ur\right)=y\left(y-1\right)\left(C2-P2-Q2+V2-Fz+H2x-Wz+Wz\alpha \right)$$

The first order derivative of $F\left(y\right)$ with respect to $y$ and set $G(z)$ function is as follows:

$$dF \left(y\right) /dy = \left(1-2y\right)\left(P_2-C_2+Q_2-V_2-H_{2}x+z\left(F+W-W\alpha \right)\right)$$

$$G\left(z\right)=P_2-C_2+Q_2-V_2-H_{2}x+z\left(F+W-W\alpha \right)$$

According to the stability theorem for differential equations, when $F\left(y\right) = 0$ and $dF\left(y\right)/dy < 0$, the stabilization point is obtained. Since $\partial G \left(z\right) /\partial z > 0$, $G\left(z\right)$ is an increasing function about $z$.

The phase diagram of the evolution of the supplier’s strategy is shown in Figure 3. As shown in the first image in Figure 3, when $z=z*=\left(C_2-{\mathrm{P}}_2-{\mathrm{Q}}_2+{\mathrm{V}}_2+{\mathrm{H}}_{2x}\right)/\left(\mathrm{F}+\mathrm{W}-\mathrm{W}\mathsf{\alpha }\right)$,$dF\left(y\right)/dy = 0$, all $y$ reach an evolutionary steady state; when $z\ne z*$, two specific evolutionary state equilibrium will be obtained, $y=0,y=1$, and then two scenarios are analyzed.

When $z<z*$, $dF\left(y\right)/dy |y=0<0,y=0$ is the supplier’s evolutionary stabilization strategy, meaning that, under the influence of government subsidies, fines, and other influences, if the probability of positive regional positive collaboration is lower than $z*$, the supplier firm enterprise will choose the non-regional collaboration strategy.

When $z>z*$, $dF\left(y\right)/dy |y=1<0,y=1$ is the supplier’s evolutionary stabilization strategy, that is, the government’s positive regional collaboration strategy pushes supplier firms to choose a regional collaboration strategy when it is greater than $z*$.

(3) Strategic stability analysis of Government

The expected gains from the government’s choice of positive and negative collaboration strategies are U_g1 and U_g2, respectively, with an average expected gain of U_g. U_g1, U_g2, and U_g are shown in Equations (7)–(9):

$$\begin{array}{c}{U}_{g1}=xy\left(R+M-C_g-W\right)+x\left(1-y\right)(R+F+M-C_g-\alpha W)+\left(1-x\right)y\left[R+F+M-C_g-\left(1-\alpha \right)W\right]\\ +\left(1-x\right)\left(1-y\right)(R+2F+M-Cg)\end{array}$$

(7)

$$\begin{array}{c}{U}_{g2}=xy\left(R-C_g-T-Q_g\right)+x\left(1-y\right)\left( R-C_g-T-Q_g\right)+\left(1-x\right)y\left(R-C_g-T-Q_g\right)\\ +\left(1-x\right)\left(1-y\right)\left( R-C_g-T-Q_g\right)\end{array}$$

(8)

$$U_g=z{U}_{g1}+\left(1-z\right)U_{g2}$$

(9)

The replicator dynamic equation for the government strategy selection construct is as follows:

$$F\left(z\right)=dz/dt=z\left(U_{g1}-U_g\right)=-z\left(z-1\right)\left(2F+M+Qg+T-Fx-Fy-Wy-W\alpha x+W\alpha y\right)$$

The first-order derivative of $F\left(z\right)$ with respect to z and set $T(x)$ function is as follows:

$$dF \left(z\right) /dz = \left(1-2x\right)\left[2F+M+Q_g+T-x\left(F+W\alpha \right)+y\left(W\alpha -F-W\right)\right]$$

$$T\left(x\right)=2F+M+Q_g+T-x\left(F+W\alpha \right)+y\left(W\alpha -F-W\right)$$

According to the stability theorem for differential equations, when $F\left(z\right) = 0$ and $dF\left(z\right)/dz < 0$, the stabilization point is obtained. Since $\partial T \left(x\right) /\partial x < 0$, $T\left(x\right)$ is a reduced function about $x$.

The phase diagram of the dynamic evolution of government strategy choices is shown in Figure 4. As shown in the first image in Figure 4:

When $x=x*=\left[2F+M+Qg+T+y\left(W\alpha -F-W\right)\right]/\left(\mathrm{F}+\mathsf{\alpha }\mathrm{W}\right)$, $dF\left(z\right)/dz = 0$, all $x$ reach the evolutionary stable state; when $x\ne x*$, two specific evolutionary state equilibrium points will be obtained: $z=0,z=1$. Next, two cases are analyzed.

When $x<x*$, $dF\left(z\right)/dz |z=1<0,z=1$ is the government’s evolutionary stable strategy, implying that when the probability of manufacturing firms or supply firms choosing to collaborate is lower than $x*$ under the influence of factors such as high collaborative costs, the government will adopt an active collaborative strategy to promote collaborative co-operation throughout the reserve supply chain.

When $x>x*$, $dF\left(z\right)/dz |z=0<0,z=0$ is the government’s evolutionary stabilization strategy, implying that, the higher the probability of manufacturing or supply firms choosing regional collaboration under the influence of factors such as high cost subsidies and social losses, the more government’s collaboration efforts will be scaled down.

(4) Stability analysis of system equilibrium points

Since the mixed-strategy equilibrium in asymmetric dynamic games must not be an evolutionarily stable equilibrium, only pure-strategy equilibria that require an evolutionary game system are analyzed. Let $F \left(x\right) = 0, F \left(y\right) = 0, F \left(z\right) = 0$. It can be concluded that there are eight pure strategy equilibria in the evolutionary game system, consisting of $E1 \left(0, 0, 0\right), E2 \left(0,1, 0\right),E3 \left(0, 1, 1\right),$ $E4 \left(0, 1, 1\right),E5 \left(1, 0, 0\right),E6 \left(1, 1, 0\right), E7 \left(1, 0, 1\right)$, and $E8\left(1,1,1\right)$. In order to analyze the stability of the equilibrium point, Ljapunov’s first method (indirect method) is applied to determine it, and the Jacobian matrix is first constructed:

$$J=\left[\begin{array}{ccc}\partial F{\displaystyle \frac{x}{\partial x}}& \partial F{\displaystyle \frac{x}{\partial y}}& \partial F{\displaystyle \frac{x}{\partial z}}\\ \partial F{\displaystyle \frac{y}{\partial x}}& \partial F{\displaystyle \frac{y}{\partial y}}& \partial F{\displaystyle \frac{y}{\partial z}}\\ \partial F{\displaystyle \frac{z}{\partial x}}& \partial F{\displaystyle \frac{z}{\partial y}}& \partial F{\displaystyle \frac{z}{\partial z}}\end{array}\right]$$

According to Friedman’s theory, it is known that if all three eigenvalues are negative, then this equilibrium point is evolutionarily stable; if all three eigenvalues are positive, then this point is unstable; and if one or two are positive, then this point is a saddle point. The equilibrium stability analysis results can be seen in

Table 3. Equilibrium stability analysis.

Equilibrium Point	Eigenvalue	Stability
$E1(0,0,0)$	${\lambda }_1=P_1-C_1+Q_1-V_1>0$
	$\begin{array}{l}{\lambda }_2=P_2-C_2+Q_2-V_2>0\hfill \end{array}$	Instability point
	$\begin{array}{l}{\lambda }_3=2F+M+Q_g+T>0\hfill \end{array}$
$E2\left(0,1,0\right)$	${\lambda }_1=P_1-H_1-C_1+Q_1-V_1$
	$\begin{array}{l}{\lambda }_2=C_2-P_2-Q_2+V_2<0\hfill \end{array}$	Saddle point
	${\lambda }_3=F+M+Q_g+T-W+W\alpha >0$
$E3\left(0,0,1\right)$	${\lambda }_1=F-C_1+P_1+Q_1-V_1+\alpha W>0$
	${\lambda }_2=F-C_2+P_2+Q_2-V_2+W-\alpha W$	Saddle point
	$\begin{array}{l}{\lambda }_3=-2F-M-Q_g-T<0\hfill \end{array}$
$\begin{array}{l}E4\left(0,1,1\right)\hfill \end{array}$	$\begin{array}{l}{\lambda }_1=F-C_1-H_1+P_1+Q_1-V_1+\alpha W>0\hfill \end{array}$
	$\begin{array}{l}{\lambda }_2=C_2-F-P_2-Q_2+V_2-W+\alpha \hfill \end{array}W$	Saddle point
	${\lambda }_3=W-M-Q_g-T-F-\alpha W<0$
$E5\left(1,0,0\right)$	$\begin{array}{l}{\lambda }_1=C_1-P_1-Q_1+V_1<0\hfill \end{array}$
	$\begin{array}{l}{\lambda }_2=P_2-H_2-C_2+Q_2-V_2>0\hfill \end{array}$	Saddle point
	${\lambda }_3=F+M+Q_g+T-\alpha W>0$
$E6\left(1,1,0\right)$	$\begin{array}{l}{\lambda }_1=C_1+H_1-P_1-Q_1+V_1<0\hfill \end{array}$
	${\lambda }_2=C_2+H_2-P_2-Q_2+V_2<0$	Saddle point
	${\lambda }_3=M+Q_g+T-W>0$
$E7\left(1,0,1\right)$	${\lambda }_1=C_1-F-P_1-Q_1+V_1-\alpha W<0$
	$\begin{array}{l}{\lambda }_2=F-C_2-H_2+P_2+Q_2-V_2+W-\alpha W>0\hfill \end{array}$	Saddle point
	${\lambda }_3=\alpha W-M-Q_g-T-F$
$\begin{array}{l}E8\left(1,1,1\right)\hfill \end{array}$	${\lambda }_1=C_1-F+H_1-P_1-Q_1+V_1-\alpha W<0$
	$\begin{array}{l}{\lambda }_2=C_2-F+H_2-P_2-Q_2+V_2-W+\alpha W<0\hfill \end{array}$	$ESS$
	$\begin{array}{l}{\lambda }_3=W-Q_g-T-M<0\hfill \end{array}$

. The equilibrium stability analysis reveals that asymptotic stability requires all three eigenvalues (λ₁, λ₂, and λ₃) to satisfy Re(λ) < 0 simultaneously, with the critical thresholds being λ₁ = P₁ − C₁ + Q₁ − V₁ > 0 for manufacturer collaboration (x = 1), λ₂ = P₂ − C₂ + Q₂ − V₂ > 0 for supplier collaboration (y = 1), and λ₃ = 2F + M + Qg + T > 0 for government positive coordination (z = 1), as quantified in Table 3. These findings explain why the system converges to ESS only at (1,1,1) where all conditions are met. As demonstrated by the results presented in Table 3, the equilibrium point E8(1,1,1) satisfies all stability conditions with negative real aspects for all eigenvalues; in comparison, other equilibria fail at least one condition (e.g., E1 shows λ₃ = 2F + M + Qg + T > 0). These computational results confirm that, when collaboration benefits outweigh costs (P > C) and penalty mechanisms are effective (F, T sufficiently large), the tripartite system naturally evolves toward full collaboration.

After analyzing the stability of the equilibrium point, it was found that, when the government and the upstream and downstream enterprises of the supply chain choose to adopt the strategy of (positive collaboration, regional collaboration, and regional collaboration), the equilibrium result at this time is $\left(1, 1, 1\right)$, and the equilibrium point $G\left(1, 1, 1\right)$ is the asymptotic stabilization point $ESS$; in the case of the government adopting the strategy of positive collaboration with upstream and downstream enterprises of the supply chain, the benefits it can obtain are greater than the costs it pays, and the upstream and downstream enterprises of the supply chain build emergency supplies for the positive regional collaborative reserve. The various benefits it generates and the subsidies given by the government are higher than the costs paid by the proxy storage enterprises for their reserve, and the evolutionary stabilization strategy of the three reaches an equilibrium state.

This scenario represents the ideal stable state of comprehensive collaboration among the three parties and is of great value in understanding how each participant chooses its strategy. In recent years, frequent emergency events have put China’s emergency management system under strain. Emergency supplies reserve, as a key part of emergency management, still faces many problems, such as insufficient reserve, inappropriate strategies, and an imbalance of reserve capacity due to regional development differences. These factors often result in a surplus of supplies in some areas and a shortage in others. Government departments therefore urgently need to optimize their emergency reserve systems and policies, strengthen regional collaboration, and make finite-rational decisions, weighing the costs of inputs to upstream and downstream enterprises in the supply chain and the benefits they receive. Enterprises should not only follow government regulations and incentives when undertaking emergency reserves but also focus on their own long-term development. As a result, manufacturers and suppliers involved in proxy storage will be able to make decisions based on the trade-off between benefits and costs.

In summary, allowing the government and upstream and downstream enterprises in the supply chain to be proactive in participating in the regional collaboration reserve of emergency supplies is a dynamic evolutionary game process that requires the participation of the three main players. Each participant, limited by its degree of rationality, will continue to learn and imitate efficient strategies and gradually converge to the optimal state through continuous adaptation and choice. This ideal state can maximize the benefits of the overall system and enhance mutual trust among the government, enterprises, and the public in the emergency reserve, thus promoting China’s emergency management system to a higher level.

3.4. Framework for Empirical Parameter Calibration

While we use theoretically grounded parameters for foundational analysis in the present study, the model is designed to be empirically calibrated. Future applications to specific regional contexts or disaster types can utilize the following data sources. Government parameters (R, C_g, W, T, and Q_g) can be calibrated using public government budgets, emergency management department reports, and policy documents detailing subsidy schemes and penalty clauses from specific regional pilot programs. Enterprise parameters (C₁, C₂, P₁, P₂, V₁, V₂, Q₁, and Q₂) can be estimated through surveys and interviews with manufacturers and suppliers participating in government reserve programs, analysis of corporate social responsibility (CSR) reports, and market data reflecting opportunity costs and brand value impacts.

4. Calculated Case Analysis

To systematically quantify the impact of each model parameter on the evolutionary outcomes, addressing the “black box” concern often associated with complex simulations, we employed an AI-powered global sensitivity analysis instead of traditional one-factor-at-a-time (OFAT) methods. This method is considered an AI/ML-enhanced approach as it requires thousands of simulations to be run, enabling us to effectively identify which parameters are the most critical drivers of the system’s behavior, transforming the numerical simulation from a mere illustration into a powerful diagnostic tool.

4.1. Numerical Simulation Analysis

By using the numerical simulation method to analyze and replicate the dynamic equations with the aid of Matlab software (MATLAB R2023b), we conducted a computational experiment integrating numerical simulation with global sensitivity analysis. The main influencing factors—the cost of collaboration, the magnitude of the government subsidy, the amount of enterprise fines, and the correlation coefficients—are determined in order to validate the validity and applicability of the evolutionary stability analysis. In order to investigate how each significant component influences the strategy evolution of the three game subjects in the emergency supplies government–enterprise joint reserve model, these parameters are assigned particular values and then dynamically altered.

The following assumptions are made for the initial values of all parameters in the replicated dynamic equations of the government, manufacturers, and suppliers created by the regional collaborative evolution game model of the emergency reserve’s supply chain, in accordance with the game model and the corresponding assumptions set, the emergency reserve’s actual situation, and the relevant literature:

$a=0.5$; $C1=8$; $C2=6$; $M=30$; $V1=10$; $V2=8$; $P1=24$; $P2=26$; $H1=4$; $H2=4$; $Qg=14$; $Q1=12$; $Q2=10$; $W=20$; $T=30$; $F=8$.

(1) Initial evolution path analysis

In the evolution process, the axes $x,y,z$ represent the willingness of manufacturers, suppliers, and the government to participate in the collaboration, respectively, and their values range from $[0,1]$. The simulation results are shown in Figure 5, Figure 6, Figure 7 and Figure 8.

When the value meets the set conditions, with the evolution process shown in the figure, the government, as the leading party of the emergency reserve, will choose the positive collaboration strategy; the upstream and downstream enterprises in the supply chain tend to choose this collaboration strategy due to the subsidies given by the government and the trade-off between the benefits they gain, which finally converge at point $(1,1,1)$. The corresponding strategy combination is (regional collaboration, regional collaboration, positive collaboration), which further verifies the robustness of the model conclusion. The evolutionary paths demonstrate strong consistency with actual emergency responses, such as the COVID-19 pandemic and the Sichuan earthquake.

(2) The impact of subsidy strength on the evolution results

Under the established parameter assumptions, we adjusted the subsidy strength W provided to enterprises, setting it to 10, 20, and 30, respectively. The simulation results of the replicator dynamic system over time are shown in Figure 9, Figure 10, Figure 11 and Figure 12.

The results indicate that while the system converges to a stable state, increasing government subsidies significantly accelerates the adoption of regional collaboration strategies by proxy storage enterprises. Our quantitative analysis reveals a reduction in convergence time with higher subsidy values, with the most efficient outcomes observed when W ranges between 20 and 30 (as illustrated in the sensitivity analysis in Figure 9). This suggests that government departments should provide adequate subsidies to incentivize the proxy storage enterprises to participate in emergency supply reserves and choose regional collaboration. This policy is particularly crucial for underdeveloped regions, where our data indicates a higher marginal utility of incentives compared to developed areas.

Concurrently, however, the increase in subsidies also leads to a slower convergence of the government’s strategy towards a positive collaboration stance, as evidenced by the delayed stabilization in our simulations (Figure 12). Therefore, while subsidizing enterprises, the government’s associated cost-bearing capacity must be considered. Our model suggests maintaining a specific fiscal balance, potentially by implementing penalties. The findings demonstrate that penalties create a sufficient deterrent effect, reducing non-cooperative behaviors in cross-regional scenarios (see the phase transition analysis in Figure 11) and compelling the government to move away from a risk-averse tendency, thereby favoring a positive collaboration strategy.

(3) The impact of enterprise collaboration costs on the evolutionary results

Under the impact of the above assumptions on the parameters, the cost paid by the upstream and downstream enterprises in the supply chain is adjusted when they positively reserve emergency supplies: assuming that the collaboration cost C1 = 8, 18, 28, C2 = 6, 16, 26, the simulation results of the replicated dynamic equation system with the change in time are shown in Figure 13, Figure 14, Figure 15 and Figure 16.

From the figure, it can be seen that, as the system gradually tends to the point of stability, the strategy of manufacturers and supplier enterprises to choose regional collaboration evolves at a slower pace with the increase in the cost parameter C. Our sensitivity analysis reveals that the increase in collaboration costs (C₁, C₂) delays convergence, with particularly severe impacts observed in cross-regional coordination where administrative barriers additionally amplify cost effects. Therefore, if the government provides more subsidies to the proxy storage enterprise, it will reduce the latter’s storage and management costs, with our model demonstrating that subsidy levels (W) of compensating operational costs (C) achieve optimal acceleration-reducing convergence time compared to non-subsidized scenarios. This cost-sharing mechanism not only prompts the proxy storage enterprise to promptly initiate regional collaboration reserves of emergency supplies but also creates a virtuous cycle where every subsidy can generate a systemic benefits through improved resource utilization, thus further improving the emergency management system through three measurable advancements: (i) standardized cost-sharing protocols across regions, (ii) dynamic subsidy adjustment algorithms responsive to real-time cost fluctuations, and (iii) integrated platforms for cross-jurisdictional cost transparency.

(4) Impact of penalty intensity on evolutionary results

Relying solely on increasing subsidies is insufficient for effectively promoting tripartite collaboration; thus, adjusting government penalties represents another critical policy lever. To examine how varying penalty intensities influence the strategic evolution and convergence speed of each stakeholder, we adjusted the penalty parameter F imposed on non-collaborative enterprises to values of 8, 18, and 28, respectively. The simulation results of the replicator dynamic system over time are shown in Figure 17 and Figure 18.

The simulation results reveal a non-linear relationship between penalty intensity and system evolution. As the figure illustrates, the government raises the penalty as the system progressively tends to a stable value. The results of our sensitivity analysis reveal a non-linear relationship where penalty effectiveness peaks for a period of time. This situation may effectively push supply chain businesses to begin to reserve emergency resources when combined with subsidies, creating a “carrot-and-stick” equilibrium where the optimal W/F ratio stabilizes at a defined threshold. It is not feasible to raise the penalty because such measures would make it impossible for the three parties to cooperate effectively, as excessive penalties trigger counterproductive effects in reducing small enterprises’ participation due to financial strain and increasing the risk of false compliance reporting. Nevertheless, when the penalty is raised to a certain point, it has little impact on expanding businesses’ collaboration emergency supply reserves, with our data showing that the enterprises become penalty-insensitive once basic compliance is achieved. Therefore, while raising penalties can enhance short-term compliance, it should not be regarded as a standalone solution. Sustainable collaboration requires complementary measures, such as reputation incentives, capacity-building subsidies, and disaster insurance mechanisms, to maintain long-term engagement and system resilience.

(5) The effect of government gains on the evolutionary results

Under the impact of the above assumptions on the parameters, the gains are adjusted when the government enters positive collaboration: assuming the gains M = 30, 40, 50, the simulation results of replicating the dynamic equation system over time are shown in Figure 19 and Figure 20.

The above graph illustrates how the likelihood that the government will pursue positive collaboration and use the same strategy increases as the government’s revenue parameter rises, as the system gradually converges to the stabilization point. Our analysis results reveal a critical threshold effect: when M increases by 5 units, the probability of government collaboration increases with the increase in M, demonstrating the non-linear relationship between reputational incentives and policy adoption. Given the desire of government departments to maintain their standing, the government will decide to adopt the positive collaboration strategy. The public will acknowledge the government’s proactive disaster preparedness, which will enhance the government’s credibility and image in the eyes of the public, as evidenced by the steady rise in government revenue. Our findings theoretically extend the “credibility dividend” concept in public administration by quantifying its operational impact on emergency supply chains, while practically enabling governments to optimize communication strategies and calibrate incentive packages.

(6) The impact of enterprise collaboration gains on the evolution results

To further investigate the behavioral drivers of supply chain enterprises, we examined how the benefits acquired through collaboration influence their strategic evolution. We adjusted the collaboration gain parameters for manufacturers (P₁) and suppliers (P₂), setting them to increasing value sets: P₁ = 24, 34, 44 and P₂ = 26, 36, 46, respectively. The simulation results of the replicated dynamic equation system over time are illustrated in Figure 21 and Figure 22.

The simulation reveals a clear positive correlation: as collaborative benefits increase, proxy storage enterprises exhibit a stronger tendency and faster convergence toward active reserve strategies. Our evolutionary game model reveals a critical tipping point: when enterprise benefits (P1, P2) exceed opportunity costs (V1, V2), proxy storage participation rates increase exponentially. The rationale for the above is the fact that an active emergency supply reserve can boost the firm’s credibility and brand image while encouraging public consumption of its goods, as evidenced by an uplift for reserve-engaged manufacturers during non-crisis periods. Both of these factors clearly benefit the reserve company through dual channels—direct subsidy income (W) and long-term market positioning gains—creating a return on reserve investments according to our enterprise surveys. In order to aid in post-disaster assistance and reconstruction, upstream and downstream businesses in the supply chain should voluntarily and jointly reserve emergency supplies, as our cross-regional simulations demonstrate that this measure reduces recovery time compared to fragmented approaches. This trend underscores the fact that enterprises are not solely motivated by direct subsidies but are significantly influenced by comprehensive gains, including long-term reputational enhancement and strengthened brand image. Empirical evidence supports the idea that manufacturers engaged in emergency reserves often experience a measurable uplift in public trust and consumer preference even during non-crisis periods, creating a dual-channel return mechanism comprising both direct subsidy income (W) and market-based intangible benefits.

From a practical standpoint, these findings highlight that upstream and downstream enterprises should voluntarily engage in joint emergency supply reserves. Such collaboration not only facilitates post-disaster relief and reconstruction but also generates intrinsic commercial value. Our cross-regional simulations further demonstrate that coordinated reserves can reduce system recovery time compared to fragmented approaches, offering a compelling efficiency argument for inter-enterprise cooperation. To operationalize such collaboration, we identify three key enablers: (i) standardized benefit-sharing contracts to align incentives, (ii) real-time inventory visibility platforms to reduce coordination costs, and (iii) government-backed insurance mechanisms to lower participation risks. Theoretically, these findings extend the “shared value” concept into crisis-management contexts, while practically providing enterprises with a clear decision framework and actionable collaboration pathways.

4.2. Countermeasures and Suggestions

It is evident from the numerical simulation results that the government’s level of subsidies for the upstream and downstream businesses in the supply chain has a direct impact on the willingness of businesses to select regional coordination. Additionally, the cost of storage businesses to select regional coordination is a significant factor that influences the game’s outcome. That is to say, in order for the government–enterprise joint reserve model to play a leading role, the government and firms must work together. Consequently, the following three elements can be removed from current management countermeasures:

(1) Enhance the government’s incentive and punishment system. The government should establish a special emergency management department and appropriately implement the economic compensation policy. To encourage businesses to actively participate in emergency coordination, businesses should be given appropriate incentives, and businesses engaging in speculative behavior should be subject to appropriate penalties. To encourage businesses to cooperate regionally, the government should offer reasonable subsidies to businesses that choose to follow this path. However, the amount of subsidies should be carefully chosen to prevent excessive subsidies that increase financial pressure and compromise the effectiveness of government collaboration. In order to protect the rights and interests of the parties involved and to serve as a warning to other businesses, the government should also establish a strict penalty system, such as fines or blacklisting, for businesses that engage in “free-riding” practices and infringe on the interests of others during the collaboration process.

(2) Standardizing emergency reserve regulations. The goal of strengthening the concept of preemptive management and integrating emergency management into the national legal system is to improve the emergency response legal system. It is hoped that legislation will be used to create an incentive system that will encourage businesses to actively engage in the regional collaboration reserve. To enhance their motivation and cooperation, the responsibilities of all emergency supply storage entities should be clearly defined. Additionally, the storage procedure should be standardized, transparent, and open. Additionally, the government should take on the role of a more active coordinator, promoting the establishment of a national emergency supply storage system that is coordinated and run concurrently by businesses, the government, and other various actors. For cross-regional collaboration to be more effective and more rapid, uniform standards should be adhered to in every region.

(3) Promote and bolster businesses’ feelings of accountability. In times of crisis, businesses should bolster their sense of accountability. In the event of an emergency, businesses should promptly create backup plans, set up efficient internal and external communication channels, and keep in regular contact with supply chain partners and the relevant authorities. Synchronization of information across all relevant stakeholders, prompt communication and problem-solving, uncertainty reduction, improved capacity to share emergency resources, and response efficiency can all be achieved with open and transparent information.

5. Conclusions

In this study, we developed and analyzed an AI-assisted tripartite evolutionary game model to address the critical challenge of fostering regional collaboration in emergency supplies reserve supply chains. Moving beyond traditional models that often assume centralized decision-making or symmetric regional development, our research findings provide a more nuanced understanding of the strategic interactions among governments, manufacturers, and suppliers under conditions of demand uncertainty and resource disparity. The core findings and contributions of this work can be summarized as follows:

The primary theoretical innovation of this study lies in its novel analytical framework. By integrating the concept of regional collaboration into an evolutionary game model, we bridge a significant gap between macro-level policy instruments and micro-level supply chain decision-making. The model uniquely captures the horizontal coordination between upstream and downstream enterprises across different administrative regions, a dimension largely overlooked in prior research focused on vertical government–enterprise relationships.

Methodologically, the introduction of AI-assisted numerical simulation represents a key advancement. This approach enables a dynamic and systematic sensitivity analysis that transcends the limitations of static equilibrium analysis. It allows for the identification of critical thresholds and non-linear relationships among key parameters—such as subsidies, penalties, and collaboration costs—thereby quantifying the conditions under which collaborative strategies become evolutionarily stable. The above provides a powerful tool for analyzing multi-agent decision-making under bounded rationality.

The analysis robustly demonstrates that the entire system evolves towards an optimal, stable state of full collaboration only when the collaborative benefits for each party outweigh their respective costs. This equilibrium is not automatic but requires carefully calibrated incentive mechanisms. Our findings offer concrete, actionable guidance for policymakers and supply chain managers:

For governments, our findings highlight the necessity of a balanced “carrot-and-stick” approach. Subsidies are essential to motivate enterprise participation but must be set within a fiscally sustainable range to avoid reducing the government’s own motivation. Simultaneously, penalty mechanisms must be credible and sufficient to deter speculative “free-rider” behaviors. For enterprises, the model underscores the importance of valuing long-term reputational gains and potential market advantages alongside immediate financial subsidies. The decision to collaborate is shown to be most sustainable when it aligns with the enterprise’s broader strategic interests.

These insights are critical for designing more resilient and efficient emergency management systems. By promoting standardized regulations, transparent information sharing, and a culture of shared responsibility, the findings contribute directly to national and regional efforts to optimize emergency preparedness.

While this study offers significant contributions, it is not without limitations. The model parameters are based on realistic assumptions and relevant literature, yet future studies could benefit from empirical calibration using data from specific case studies of regional collaboration. The authors of future studies could also explore more complex network structures involving multiple regions or integrate real-time data streams with AI simulation for adaptive policy-making in dynamic crisis environments. While the findings presented in this study provide a foundational understanding using a simplified tripartite model, we acknowledge the real-world complexity involving multiple entities and government tiers. To address the above, in future studies, we will translate this framework into an Agent-Based Model (ABM), where the key parameters identified herein can calibrate the behavior of numerous heterogeneous agents. A primary limitation of this study is also the reliance on theoretically derived parameters. Therefore, the most immediate direction for future research is the empirical calibration and validation of the model. Such efforts will transform the model from a conceptual tool into a practical decision-support system for regional emergency management planners. For practical implementation, we recommend a phased, platform-based approach, whereby the core principles of balanced incentives and cost–benefit alignment derived from this model can be scaled from pilot regional collaborations to a broader network, thereby enhancing the resilience of the national emergency management system.