**1. Introduction**

As a result of its vast territory, complex environmental conditions, large population, low disaster prevention, and management capabilities, China has suffered serious losses and used a great deal of human and material resources for natural hazard prevention and relief. The government plays an important role in emergency management, but it is difficult to achieve high efficiency by only relying on the strength of the government [1]; therefore, determining how the public can effectively participate in natural hazard emergency relief has become an urgent task. Emergency relief refers to the money, goods, rescue, and services provided for victims after natural hazards [2]. Practice has proven that due to traffic inconveniences, secondary disasters, and an imperfect system, public participation has

caused a certain degree of traffic congestion and disorder. Therefore, determining how to organize the public to participate in emergency relief in an orderly and effective way has become a popular topic.

In 2016, the new "National Natural Hazard Emergency Relief Plan" was released in China, but the subject of public participation in emergency relief has concerned scholars.

Questions to clarify this emergency plan are listed below.


This paper will address these questions. The remainder of this paper is structured as follows. In Section 2, we review relevant literature and identify the research gap that this paper focuses on. In Section 3, we give a formal expression based on a multi-agent system negotiation interval, the participation decision function is presented, and the interaction process of the participants is analyzed. Section 4 gives the formal definition of the participation degree and participation strategy function, and an algorithm for constructing an emergency relief alliance is proposed. In Section 5, we compare our approach and the competing algorithms and summarize our results. Conclusions and an outlook on future research directions are given in Section 6.

Our contributions are as follows: We propose the establishment of a method for creating a natural hazard emergency relief alliance. At present, research regarding the field of public participation in emergency management mainly focuses on macroscopic, qualitative research, such as participation status and problems, influencing factors, and participation strategy; however, quantitative research is mainly concerned with resource allocation, personnel evacuation and path planning, and a lack of quantitative analysis of participation models. This paper presents a model for an emergency relief alliance and proposes a corresponding algorithm.

Moreover, taking into account that different participants have different resources and participation attitudes, we put forward the concept of public participation degree. The adoption of multi-agent technology reflects autonomy, and a natural hazard emergency relief alliance based on the public participation degree is proposed. The experiment shows that our method is reliable and applicable.

#### **2. Literature Review**

#### *2.1. Disaster Risk Reduction*

Many scholars have done large amounts of research in the field of disaster risk reduction. Adnan et al. [3] selected eleven precedent (model) cities to study their various initiatives for reducing coastal flood risks. Their findings showed that protecting cities from flooding and reducing exposure to floods are two different but interrelated approaches to disaster risk reduction (DRR). Few et al. [4] discussed findings from a two-year research project on DRM (Disaster Risk Management) capacity development. They pointed out that adaptability, ownership, sustainability, the inclusion of actors, and scales reflect the capacity of DRR. Sim et al. [5] explored the way that social workers can apply transdisciplinary strategies to work with other professional practitioners and stakeholders in meeting disaster risk reduction needs in a remote Chinese village. Marlowe et al. [6] emphasized the importance of incorporating the concepts of reach, relevance, receptiveness, and relationships into a DRR approach; these concepts were presented as an embedded guiding framework that can aid in disaster communication. Carrao et al. [7] assessed drought hazard (DH) changes for the mid-century (2021–2050) and late-century (2071–2099). The results showed that a major challenge for risk management is for human populations or their activities to not adapt to DH changes; instead, they should continue developing global initiatives that mitigate their impact on the entire carbon cycle by the late-century. Migliorini et al. [8] conducted surveys and interviews with National Sendai Focal Points and stakeholders in science and research, governmental agencies, non-governmental organizations, and industry, enhancing data interoperability for disaster risk reduction. Marchal et al. [9] pointed out

that reinsurance and insurance industries play a role in helping to manage risks and improve disaster risk reduction (DRR) as well as loss prevention. Webb [10] stated that early tropical cyclone warnings are necessary but not sufficient to ensure that households and communities are prepared. There is a strong case for investments in mid- to long-term DRR focused on community and household capacity, prioritizing women's active and equal participation as community leaders, and disability inclusion. Totaro et al. [11] drew maps of monothematic and synthetic indices to describe the hazard status of metropolitan areas; a hazard hotspot map was also elaborated to identify areas with high hazards.

#### *2.2. Emergency Relief for Natural Hazards*

Many scholars have researched in the field of natural hazard emergency relief. Narayanan et al. [12] built a disaster recovery network that provides food, water, shelter, medical services, and other relief resources for the affected people through reliable communication network coordination to avoid social chaos caused by natural hazards. Wachinger [13] pointed out that personal experience, government investment, experts' awareness of disaster risk, and media reports have a significant impact on natural hazard emergency relief. Wex et al. [14] proposed a decision support model of disaster emergency relief based on the heuristic algorithm of Monte Carlo, which can effectively allocate and dispatch rescue units, greatly reducing the casualties and economic losses. Oral et al. [15] assessed the impact of earthquake experience on earthquake preparedness, and the results showed that disaster experience, residential area, and disaster relief work have an obvious relationship, and reasonable suggestions were put forward for disaster relief. Berariu et al. [16] analyzed the cascade effects of natural hazards and investigated their impact on relief operations concerning critical infrastructure; in particular, the transport infrastructure, electricity, and human health.

Alem et al. [17] took into account practical characteristics, such as budget allocation and fleet sizing of multiple types of vehicles, and developed a new two-stage stochastic network flow model to help decide how to rapidly supply humanitarian aid to victims of a disaster. Toyasaki et al. [18] focused on horizontal cooperation in inventory management, which is currently implemented in the United Nations Humanitarian Response Depot (UNHRD) network, and proposed a policy priority for the first-best system of optimal inventory management. Deo [19] used integrated disaster management technology, a quantitative method, and big data analysis to create disaster and early warning models to reduce the impact of these disasters and provide a comprehensive method for disaster management systems. Nassereddine et al. [20] presented a multi-criteria decision-making approach to evaluate emergency response systems by taking into account the interaction synergy. Borowska-Stefanska et al. [21] developed an optimization pattern for the process of a preventive evacuation of people from flood-risk areas aimed at mitigating the negative effects of the flood.

Some scholars have researched natural disaster management based on multi-agent system theory. Izida et al. [22] pointed out that communication in natural disasters can be simulated through the multi-agent paradigm, which provides better assistance in emergencies of natural disasters. Yue et al. [23] built a disaster emergency collaborative decision-making framework based on a multi-agent system, designed a collaborative decision-making model, and achieved a natural disaster dynamic emergency decision-making process. Naqvi et al. [24] proposed an agent-based model of a stylized low-income region to study the impact of natural disasters on population displacement, income, prices, and consumption with a focus on low-income groups. Na et al. [25] developed an agent-based discrete-event simulation (ABDES) modeling framework based on an embedded GIS module for making no-notice natural disaster evacuation planning. Lee et al. [26] presented a novel and efficient rescue system with a multi-agent simultaneous localization and mapping (SLAM) framework, which was proposed to reduce the rescue time while rescuing the people trapped inside a burning building.

#### *2.3. Public Participation in Natural Hazard Emergency Relief*

Since Sherry Arnstein published his famous paper "A Ladder of Citizen Participation" in 1969 and John Clayton Thomas proposed an effective decision model of citizen participation, more and more scholars have studied public participation. The research on public participation focuses on community development, e-government, social governance, environmental science, and democratic construction.

Chen et al. [27] developed an Environmental Community Consultative Group (ECCG), a small local group that can promote public environmental awareness, improve public environmental behavior, and facilitate public engagement in environmental management. Werner [28] pointed out that policymakers' dependencies, motivations, and decision-making processes lead them to evaluate firms by using a sociopolitical reputation as a differentiating heuristic. Say et al. [29] shed light on the basic principles related to the process of public participation and how effective participation can be attained in SEA (Strategic Environmental Assessment) practices at the policy, plan, and program levels. Park et al. [30] aimed to investigate the legal opportunities of public participation in managing state forests in the case of the Republic of Korea (ROK), which provide legal insights on promoting public participation in managing state forests. Xi [31] analyzed the problems of public participation in environmental protection and put forward three ways to optimize the system of public participation in environmental protection. Li [32] indicated that it is not necessary to impose a kind of external force on residents through empowerment, then make them engage in tourism. Self-empowerment means consciousness, awareness, and promotion of participation ability; these are fundamental and critical points for community residents participating in tourism. Xiao et al. [33] established a structural equation model to identify key factors influencing citizen environmental willingness to participate in waste management, and this indicated that the most important influencing factor was citizen knowledge, followed by social motivation, while institutional factors had the smallest positive effect. Ruiz-Villaverde et al. [34] identified several advantages of public and stakeholder participation in water management, such as the better use of knowledge and experiences from different stakeholders, increases in public acceptance, and reduced litigation, delays, and inefficiencies in implementation. Chen et al. [35] pointed out that government supervision has a positive effect on environmental governance and can urge enterprises to actively perform pollution control. The effect of government supervision is constrained by the income and costs of enterprises, and the penalties for passive pollution control should be raised.

In summary, scholars have done a lot of research in the field of natural hazard emergency relief and public participation. However, most of the studies are focused on issues such as the positive role of public participation, the influencing factors, and the ways of public participation. Some scholars have put forward quantitative models and methods of emergency management, but multi-agent system technology is rarely used, and it is mostly used for research of material distribution and personnel evacuation. The agent has the characteristics of autonomy, sociality, and initiative. It also has some human characteristics, such as knowledge, belief, obligation, intention, and so on. So, we can regard the public as agents, and management of such emergency relief is improved by combining human intelligence with efficiencies of multi-agent systems, and we adopt agent automatic negotiation coalition technology [36] to carry out our research.

#### **3. The Establishment of the Natural Hazard Emergency Relief Alliance**

#### *3.1. The Formalization of the Natural Hazard Emergency Relief Alliance*

The public refers to the masses, nongovernmental organizations (NGO), profit organizations, and individuals. The participants of natural hazard relief include governments and the public. We define the government, social masses, NGO, profit organizations, and individuals as agents. The natural hazard alliance, composed of multiple agents, is a connected graph with the agent as the node. *Agents* = {*At1*, *At2*, . . . , *Atn*}, where *Agents* is set of agents. The agent that communicates directly with *At<sup>i</sup>* is defined as the immediate neighbor of *At<sup>i</sup> ,* expressed as *DN*(*i*) = {*i*|(*i*,*j*) ∈ E,*i* , *j*}, where E represents the set of edges that are connected between agents. *R<sup>i</sup>* is the resource that *At<sup>i</sup>* can provide when carrying out natural hazard relief. The multiple agents may form alliances of *RA1*, *RA2*, . . . , *RA<sup>n</sup>* ∈ 2 I , where each member *At<sup>i</sup>* in the alliance *RA<sup>i</sup>* has the resource *R<sup>i</sup>* , so that the

resources that the alliance *RA<sup>i</sup>* owns are described as *RRA* ∈ Q *i*∈*RA Ri* . After the formation of alliance *RA<sup>i</sup>* , the overall utility can be obtained as *P*(*RRAi*) ∈ Z <sup>+</sup>, which represents the aid given to the victims after the alliance was established. Based on the theory of stakeholders, the public, whether organizations or individuals, gains corresponding utility from participating in natural hazard relief activities; for example, an enterprise provides excavators and transportation vehicles, and gets rewards of social recognition and good reputation. The masses and volunteers get psychological satisfaction and a sense of accomplishment through donations and rescue services, which belong to the invisible utility. Thus, each member *At<sup>i</sup>* can get a utility value, which is *pi*∈Z +, P *i*∈*RA p<sup>i</sup>* = *P*(*RRA*).

#### *3.2. Participation Proposal*

*At<sup>i</sup>* can invite direct neighbor *At<sup>j</sup>* to join an alliance by sending a proposal when *At<sup>j</sup>* agrees with the proposal; *At<sup>i</sup>* and *At<sup>j</sup>* will combine their resources to complete a disaster relief mission and receive corresponding utility. *At<sup>i</sup>* sending a proposal to *At<sup>j</sup>* defined as follows.

$$S\_{ij}(\mathbb{R}\_i) = ((R\_{\mathbb{R}A - i\prime} \mathbb{R}\_i)\_{\prime} p\_{i}^{\mathbb{R}\_i}) \tag{1}$$

where *Sij*(*R<sup>i</sup>* ) means that *At<sup>i</sup>* sends a proposal to *At<sup>j</sup>* . *RRA-i* is the total resources owned by alliance *RA* before joining *At<sup>i</sup>* ; *RA-i* stands for the alliance without *At<sup>i</sup>* ; *RA* is the alliance after *At<sup>i</sup>* joins; *p Ri i* is the utility that *At<sup>i</sup>* gained after *At<sup>i</sup>* joins the alliance *RA* and provides resource *R<sup>i</sup>* . When *At<sup>i</sup>* sends a proposal to neighbor *At<sup>j</sup>* , *At<sup>j</sup>* will evaluate the utility before deciding whether to accept the proposal or not. If *At<sup>i</sup>* 's proposal does not meet *At<sup>j</sup>* 's minimum requirement, *At<sup>j</sup>* will send a counter-proposal *S ji oppose* and will require higher utility. *At<sup>i</sup>* might not immediately agree to the counter-proposal, and will continue to send a new anti-proposal *S ij oppose* to *At<sup>j</sup>* , assigning a new utility value to *At<sup>j</sup>* at the same time. In this way, both participants in the alliance will conduct several rounds of negotiation.

#### *3.3. The Negotiation Interval*

The negotiation interval is the fluctuation range of the utility distribution between the two participants, that is, the minimum utility and the maximum utility. Suppose that *At<sup>i</sup>* has formed an alliance *RA* with other agents; when *At<sup>j</sup>* joins the alliance *RA*, the added utility is given in Formula (2).

$$
\Delta P(R\_{RA}, R\_{\dot{j}}) = P(R\_{RA}, R\_{\dot{j}}) - P(R\_{RA}) \tag{2}
$$

The maximum utility that *At<sup>i</sup>* can obtain is given in Formula (3).

$$p\_j^{\text{max}}(R\_{RA\prime}R\_j) = \Delta P(R\_{RA\prime}R\_j) \tag{3}$$

The optimal participation proposal that *At<sup>i</sup>* received is *S* ∗ = max *s*∈*proposali p*(*S*), where *proposal<sup>i</sup>* represents the participation proposal set that *At<sup>i</sup>* received. The *At<sup>i</sup>* 's minimum utility is given in Formula (4).

$$p^{\min}(R\_{RA}R\_j) = \max(0, p\_i(S^\*))\tag{4}$$

Therefore, the negotiation interval is given as follows.

$$\left[p^{\min}(R\_{RA}, R\_j), p^{\max}(R\_{RA}, R\_j)\right] \tag{5}$$

#### *3.4. The Negotiation Decision-Making Function*

The negotiation decision-making function reflects that agents choose their behavior with the principle of maximizing their own utility after receiving the participation proposal. *S<sup>j</sup>* represents the proposal that *At<sup>i</sup>* received from *At<sup>j</sup>* , *Soppose* represents the counter-proposal; *Soppose* is generated according to the participation strategy. *S* ∗ *<sup>m</sup>*\_*join* represents the participation proposal sent to the neighbor *At<sup>m</sup>* that has the highest participation degree. *S* represents the history of the participation proposal; it is a set of participation proposals that *At<sup>i</sup>* had proposed. The participation strategy of the inviter agent is defined as a Formula (6). The participation strategy of the invited agent is defined as Formula (7).

$$T\_i([p\_i^{\min}, p\_i^{\max}], \overline{S}) = p\_i^{\min} + f\_{\overline{\partial}\_{ij}^n}(\mathbf{x})(p\_i^{\max} - p\_i^{\min}) \tag{6}$$

$$T\_i([p\_i^{\min}, p\_i^{\max}], \overline{\mathcal{S}}) = p\_i^{\max} - f\_{\mathcal{Y}\_{ij}^v}(\mathbf{x})(p\_i^{\max} - p\_i^{\min}) \tag{7}$$

where *f*<sup>∂</sup> *n ij* (*x*) represents the participation strategy function; the participation strategy can be divided into three types according to the different values of participation attitude values ∂*ij*:


*S* ∗ *k* (*k* ∈ *proposal<sup>i</sup>* ) represents the participation proposal sent by *At<sup>k</sup>* that belongs to the best alliance, that is, *At<sup>k</sup>* can provide the highest utility for *At<sup>i</sup>* in the neighborhood.

$$\mathcal{S}\_k^\* = \max\_{k \in \mathcal{DN}(i) \cup \{i\}} p\_i(T\_i([p\_i^{\min}, p\_i^{\max}], \overline{\mathcal{S}\_k})) \tag{8}$$

The decision function for *At<sup>i</sup>* that participates in the alliance is given in Formula (9):

$$\mathcal{S}\_{\text{Decisioni}} = \arg\max \left( \mathcal{S}\_{\text{j}}, \mathcal{S}\_{\text{oppose}} \mathcal{S}\_{\text{k}'}^{\*} \mathcal{S}\_{m\_{-} \text{j} \text{on}}^{\*} \right) \tag{9}$$

The negotiation process of agent participation in the alliance is to produce different types of participation proposals: Acceptance proposal, counter-proposal, new proposal, and rejection proposal, so that decisions that can be made by *At<sup>i</sup>* as follows. (1) *At<sup>i</sup>* proposes a counter-proposal according to the participation strategy. (2) *At<sup>i</sup>* sends a proposal to the agent who belongs to the optimal alliance and invites *At<sup>i</sup>* to join the alliance, and *At<sup>i</sup>* joins the optimal alliance. (3) *At<sup>i</sup>* creates a new alliance, namely *At<sup>i</sup>* sends a participation proposal to its neighbor *At<sup>m</sup>* and invites *At<sup>m</sup>* to participate in the alliance.

#### *3.5. The Agent Interaction Process*

The state of agents is defined as *State* = {Sleep, Wait, Done}, where the Sleep indicates the initial state of the agents and when waiting for other agents to send participation proposals; the Wait indicates that the agent has sent a participation proposal to another agent and wait for its response; the Done indicates that the agent has accepted a participation proposal and the participation negotiation has been completed.

The agent interaction process of public participation in the alliance is given as follows.


*At<sup>j</sup>* adds *At<sup>i</sup>* to the alliance set *proposal set<sup>j</sup>* and sets *State<sup>j</sup>* = Sleep, and then *At<sup>j</sup>* can receive the new participation proposal. (4) If *At<sup>i</sup>* sends a counter-proposal to *At<sup>j</sup>* , then *At<sup>j</sup>* sets *State<sup>j</sup>* = Sleep, and *At<sup>i</sup>* sets *State<sup>i</sup>* = Wait. (5) If *At<sup>i</sup>* sends a rejection proposal to *At<sup>j</sup>* , then *At<sup>j</sup>* sets *State<sup>j</sup>* = Sleep. proposal to *Atj*, then *Atj* sets *Statej* = Sleep, and *Ati* sets *Statei* = Wait. (5) If *Ati* sends a rejection proposal to *Atj*, then *Atj* sets *Statej* = Sleep. The agent interaction process is given in Figure 1.

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 7 of 17

The agent interaction process is given in Figure 1.

**Figure 1.** The agent interaction process. **Figure 1.** The agent interaction process.

#### **4. The Participation Strategy Based on Participation Degree 4. The Participation Strategy Based on Participation Degree**

#### *4.1. The Participation Strategy Function 4.1. The Participation Strategy Function*

The participation strategy function given in Formula (10) uses an exponential function to gradually produce new proposals after concessions. The participation strategy function given in Formula (10) uses an exponential function to gradually produce new proposals after concessions.

$$f\_{\hat{\theta}}(\mathbf{x}) = e^{\hat{\theta}\_{ij}^{\mathbf{u}} \mathbf{x}} - \mathbf{1} / e^{\hat{\theta}\_{ij}^{\mathbf{u}}} - \mathbf{1} \tag{10}$$

The domain and range of *f* ( ) *<sup>x</sup>* <sup>∂</sup> are [0.1]; they are monotonically increasing, but with different values of *<sup>n</sup>* ∂*ij* , the concave and convex of the function will change. The agent concession degree can be adjusted by changing the value of *<sup>n</sup>* ∂*ij* ; *<sup>n</sup>* ∂*ij* is *Ati* 's participation attitude value when *Atj* invites *Ati* to participate in the alliance in round *n*. The independent variable *x* is the ratio between the actual number of negotiation rounds and the maximum assignment negotiation rounds. That is, The domain and range of *f*<sup>∂</sup> (*x*) are [0, 1]; they are monotonically increasing, but with different values of ∂ *n ij*, the concave and convex of the function will change. The agent concession degree can be adjusted by changing the value of ∂ *n ij*; ∂ *n ij* is *At<sup>i</sup>* 's participation attitude value when *At<sup>j</sup>* invites *At<sup>i</sup>* to participate in the alliance in round *n*. The independent variable *x* is the ratio between the actual number of negotiation rounds and the maximum assignment negotiation rounds. That is, *x*=*round*/*MAX\_Round*, *round* changes from 0 to *MAX\_Round*, so *x* is guaranteed to change from 0 to 1.

#### *x*=*round*/*MAX\_Round*, *round* changes from 0 to *MAX\_Round*, so *x* is guaranteed to change from 0 to 1. *4.2. The Participation Degree*

*4.2. The Participation Degree*  The concept of the participation degree is proposed to reflect that different participants have different resources and different participation histories, and the participation process, guided by different strategies, is realized through the influence of participation degree on participation attitude value. The different values of the participation degree lead to different values of participation attitude, then lead to different participation strategies, and then lead to construction of different alliances. The The concept of the participation degree is proposed to reflect that different participants have different resources and different participation histories, and the participation process, guided by different strategies, is realized through the influence of participation degree on participation attitude value. The different values of the participation degree lead to different values of participation attitude, then lead to different participation strategies, and then lead to construction of different alliances. The details are given as follows.

details are given as follows. Definition 1. The historical participation degree of *Atj* relative to *Ati*  **Definition 1.** *The historical participation degree of At<sup>j</sup> relative to At<sup>i</sup>*

the probability of *Ati* and *Atj* building the alliance is.

$$pd\_{ij \text{\textquotedblleft}ADV} = \sum\_{\text{negation}}^{\text{partion\textquotedblright}} (i \to j) \Big/ \sum\_{\text{negation}}^{\text{partion\textquotedblleft}out} (i \to j) \tag{11}$$

*Ati* and *Atj*, the molecule is the number of successful historical negotiations between *Ati* and *Atj*. Formula (11) reflects the probability of *Atj* accepting *Ati* 's participation proposal in historical negotiations, which is called the historical participation degree of *Atj* relative to *Ati*. The higher the *where pdij*−*ADV*<sup>∈</sup> *[0, 1]. The denominator is the number of historical mutual proposals proposed by At<sup>i</sup> and At<sup>j</sup> , the molecule is the number of successful historical negotiations between At<sup>i</sup> and At<sup>j</sup> . Formula (11) reflects the probability of At<sup>j</sup> accepting At<sup>i</sup> 's participation proposal in historical negotiations, which is called the historical*

historical participation degree is, the higher the agent's participation consciousness is, and the higher

*participation degree of At<sup>j</sup> relative to At<sup>i</sup> . The higher the historical participation degree is, the higher the agent's participation consciousness is, and the higher the probability of At<sup>i</sup> and At<sup>j</sup> building the alliance is.*

**Definition 2.** *The current participation degree of At<sup>j</sup> relative to At<sup>i</sup>*

$$pd\_{ij\text{-current}} = \left| \text{DN}(i) \right| \left| \sum\_{n=\text{neighbor}\_i} \left| \text{DN}(n) \right| \right. \tag{12}$$

*where pdij*−*current*<sup>∈</sup> *[0, 1]. The denominator is the sum of resources owned by At<sup>i</sup> 's neighbors, and the molecule is the number of At<sup>j</sup> 's resources who are At<sup>i</sup> 's neighbors. When At<sup>j</sup> has more resources relative to At<sup>i</sup> 's other neighbors, we think that At<sup>j</sup> 's current participation degree is high relative to At<sup>i</sup> . The current participation degree reflects the resources owned by an agent. The more resources an agent has, the higher the current participation degree.*

**Definition 3.** *The participation degree of At<sup>j</sup> relative to At<sup>i</sup>*

$$pd\_{ij} = a \times pd\_{ij-ADV} + b \times pd\_{ij-current} \tag{13}$$

*where pdij*∈ *[0, 1], a* + *b* = *1. Definition 3 takes into account the historical participation degree in Definition 1 and the current participation degree in Definition 2. The a represents the weight of historical participation situations, and the b represents the weight of the current resource occupancy.*

Based on the above definition of the participation degree, the participation attitude value of *At<sup>i</sup>* relative to *At<sup>j</sup>* is given as follows.

$$\partial\_{ij}^{\eta} = \begin{cases} \partial\_{ij}^{\eta} - \beta\_{\partial} (pd\_{ij} - \Delta T) \,\partial \ge \partial\_{\text{min}} \\ \partial\_{\text{min}} \, \partial < \partial\_{\text{min}} \end{cases} \tag{14}$$

where the initial participation attitude value ∂ 0 *ij* can be negative, neutral, or positive. β<sup>∂</sup> is the influence factor of the participation degree. The bigger the β<sup>∂</sup> value is, the bigger the impact of participation degree on participation strategy. ∆*T* represents the threshold of the participation strategy. ∂min represents the minimum value of the participation attitude.

By proposing the concept of the participation degree, the participation process presents the following three characteristics.


From these characteristics, we can see that the neighbor's participation attitude value is always referenced in the process of disaster relief alliance formation. The participation attitude values of the neighbors with low participation degree are reduced and the participation attitude values of the neighbors with high participation degree are increased. Agents prefer to negotiate with the neighbor with the best cession strategy and gradually eliminate the neighbor with the lowest participation degree, which makes the participation strategy intelligent and more efficient.

#### *4.3. An Algorithm for Establishing the Emergency Relief Alliance*

We give an Algorithm 1 for establishing the emergency relief alliance as follows:

The algorithm flow chart is shown in Figure 2.

**Figure 2.** The algorithm flow chart.

**Figure 2.** The algorithm flow chart.

#### **5. Experiments and Analysis of Results**

#### *5.1. Experiments and Results*

**5. Experiments and Analysis of Results**  *5.1. Experiments and Results*  The experiments adopt an Eclipse platform to simulate the formation process of a multi-agent alliance; we verify the reliability and validity of our proposed model and Algorithm 1 by comparing with References [37,38].

The experiments adopt an Eclipse platform to simulate the formation process of a multi-agent alliance; we verify the reliability and validity of our proposed model and Algorithm 1 by comparing with References [37,38]. We did three experiments: (1) Comparison and analysis of the influence of different parameters on the participation strategy and negotiation result. (2) By comparing and analyzing Reference [37] We did three experiments: (1) Comparison and analysis of the influence of different parameters on the participation strategy and negotiation result. (2) By comparing and analyzing Reference [37] and our proposed method, we show the concession change caused by the dynamic change of participation degree. (3) As for the performance of alliance establishment, the differences among the References [37,38] and our proposed method are compared.

participation degree. (3) As for the performance of alliance establishment, the differences among the

**Table 1.** Parameter assignment.

and our proposed method, we show the concession change caused by the dynamic change of The initial parameters are given in Table 1.

The initial parameters are given in Table 1.

References [37,38] and our proposed method are compared.

**Parameter Name Parameter Values** 

Single Agent Cost *Ci*~*U*(0,1) Single Agent Utility *pi*~*U*(1.5*Ci*,3*Ci*)

Agent Number 10,50,100,200,500,800


**Table 1.** Parameter assignment.

To validate the adaptability of different alliance construction methods for large-scale agent sets, the agent set scales are 10, 50, 100, 200, 500, and 800; single-agent overhead is a uniform distribution of [0, 1] and the single-agent utility is greater than the overhead. To remove the effect of a series of random numbers on the experimental results, a single alliance construction is repeated 100 times; the confidence level is 95%.

The first experiment is the comparative analysis of the effects of the parameters (a, b), β0, and ∆*T* on the experimental results. The agent set scale is 100. The ∂ 0 *ij* value is a random selection of 15, 0.001, and −5. The experimental results are shown in Tables 2–4.

**Table 2.** The effect of (*a*,*b*) on the alliance construction (β<sup>0</sup> = 5, ∆*T* = 0.5).


**Table 3.** The effect of β<sup>0</sup> on the alliance construction ((*a*,*b*) = (0.5,0.5), ∆*T* = 0.5).


**Table 4.** The effect of ∆*T* on the alliance construction ((*a*,*b*) = (0.5,0.5), β<sup>0</sup> = 5).


We can see from Table 2 that the higher the percentage of *a* in (*a,b*) is, the better the negotiation performance; this is because we do not consider different network topologies so that the resource owned by the agent does not change when an alliance is formed every time, while the participation attitude towards each other is dynamically changed with the process of alliance construction. The higher the proportion of *a*, the higher the performances of negotiation success rate, average agent utility, and negotiation efficiency are. As can be seen in Table 3, the bigger the β<sup>0</sup> value, the better the performance of the alliance construction; this is because when the β<sup>0</sup> value increases, the participation degree will play a greater role in the participation strategy. As shown in Table 4, the performance of the alliance construction becomes worse when the ∆*T* value increases; this is because ∆*T* is a threshold of the participation strategy—only when the participation degree is greater than ∆*T* will the participation attitude tend to be moderate. When the ∆*T* value increases, the agent requires that negotiation partners have higher participation degrees, which makes the moderate participation attitude more difficult to achieve. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 12 of 17 because Δ*T* is a threshold of the participation strategy—only when the participation degree is greater than Δ*T* will the participation attitude tend to be moderate. When the Δ*T* value increases,

As can be seen from the above experiment, the different sets of parameters have a significant influence on the performance of the participation alliance. In order to make good comparisons among our method and the References [37,38], in the following experiments, the parameters are set compromise values, namely a = 0.5, b = 0.5, β<sup>0</sup> = 5, ∆*T* = 0.5. the agent requires that negotiation partners have higher participation degrees, which makes the moderate participation attitude more difficult to achieve. As can be seen from the above experiment, the different sets of parameters have a significant influence on the performance of the participation alliance. In order to make good comparisons among

In the second experiment, in order to obtain an empirical set of participation degrees, the 10 alliances are constructed and recorded as Alliance1, Alliance2, . . . , Alliance10. The parameters are given in Table 1; the alliances are constructed by Reference [38] and our method, respectively. The agent set scale is 100, and positive, neutral, and passive participation strategies are adopted to construct alliances. The comparison between Reference [38] and our method (Alliance5, Alliance8, Alliance9, Alliance10) is given in Figure 3. our method and the References [37,38], in the following experiments, the parameters are set compromise values, namely a = 0.5, b = 0.5, β<sup>0</sup> = 5, Δ*T* = 0.5. In the second experiment, in order to obtain an empirical set of participation degrees, the 10 alliances are constructed and recorded as Alliance1, Alliance2, ..., Alliance10. The parameters are given in Table 1; the alliances are constructed by Reference [38] and our method, respectively. The agent set scale is 100, and positive, neutral, and passive participation strategies are adopted to construct alliances. The comparison between Reference [38] and our method (Alliance5,Alliance8,

**Figure 3.** Experiment results. **Figure 3.** Experiment results.

As we can see from Figure 3, our proposed method can achieve better negotiation results by

∂*ij* is assigned in advance and cannot change in the negotiation

∂*ij* is not assigned in advance in our proposed

process, and the concession extent is relatively fixed. The ability of negotiation participants cannot be

convergence after fewer rounds of negotiation. In Reference [38], the parameter *<sup>n</sup>*

fully reflected in Reference [38], while the parameter *<sup>n</sup>*

As we can see from Figure 3, our proposed method can achieve better negotiation results by dynamically changing participation attitude values and selecting neighbors with better participation attitude values. The participation attitude becomes more and more moderate and achieves convergence after fewer rounds of negotiation.

In Reference [38], the parameter ∂ *n ij* is assigned in advance and cannot change in the negotiation process, and the concession extent is relatively fixed. The ability of negotiation participants cannot be fully reflected in Reference [38], while the parameter ∂ *n ij* is not assigned in advance in our proposed method. Once each alliance is formed, the negotiation data will be saved. The alliance formed in the next round will use this series of data to calculate the agent's participation degree, followed by the dynamic assignment of ∂ *n*+1 *ij* according to the ∂ *n ij* value. This participation strategy based on participation degree fully reflects the difference in negotiation ability and negotiation situation.

The third experiment is a comparative analysis of References [37,38] and our proposed method. The parameters are given in Table 1. The agent set scales are 10, 50, 100, 200, 500, and 800; a single alliance construction is repeated 100 times, and the confidence level is 95%. The tenth-time experimental results of alliance construction (Alliance10) are adopted for comparison. The comparison results are given in Table 5.


**Table 5.** The performance comparison for alliance construction.

As can be seen from Table 5, our proposed method has obvious advantages in the negotiation rounds; because of the lack of an intelligent negotiation strategy in Reference [37], the negotiation rounds are greater than in the latter two methods, and our method reduces the negotiation rounds by more than 50% compared with the Reference [38], which is without the participation degree. In terms of the negotiation success rate, the negotiation success rate of Reference [37] is much larger than the latter two methods because Reference [37] is based on the candidate union set, and the negotiation success rate of our method is 10% higher than that of Reference [38]. As for average agent utility, Reference [38] is significantly higher than Reference [37] and our method, which shows that Reference [38] achieves higher individual utility. As for the negotiation efficiency, our method is much better than the other two methods, which shows that our method can achieve higher utility in a few rounds of negotiation, so the negotiation efficiency is high.

As for the different scales of the agent set, we can see that the larger the agent scale, the higher the negotiation success rate in our method; this is because the resources among agents have increased. However, in Reference [37], the negotiation success rate decreases with the increase of agent scale. From the perspective of negotiation efficiency, our method has the absolute advantage compared with the other two methods; this advantage is more obvious with the increase of agent scale, and shows that our method is suitable for large-scale agent systems. This is because the heterogeneity among agents is more pronounced when the agent scale increases. Our method, based on the participation degree, can well reflect this heterogeneity to improve the negotiation efficiency and make it easier to reach agreement.

#### *5.2. Results Discussion*

#### 5.2.1. Experiment 1

As can be seen from Table 2, the bigger the proportion of *a* in (*a,b*), the fewer the negotiation rounds, and the higher the negotiation success rate and negotiation efficiency. However, the bigger the proportion of *b* in (*a,b*), the more the negotiation rounds, and the lower the negotiation success rate and negotiation efficiency. The reason is that for the public, such as enterprises, NGOs, and volunteers, if they have higher historical participation degrees, that means that they have a sense of participation and positive participation attitude, and it is easy to build alliances in the face of natural hazards, so the negotiation efficiency and negotiation success rate are high. If the public provides more resources and gets less utility from participating in natural hazard relief, they will refuse to build an alliance. Therefore, more rounds of negotiation are needed, and the negotiation efficiency and success rate are low.

As can be seen from Table 3, the bigger the β0, the fewer the negotiation rounds, and the higher the negotiation success rate and negotiation efficiency. The reason is that the bigger the β<sup>0</sup> is, the higher the participation degree is, and, the more positive the public's participation attitude is, the easier it is to build alliances in natural hazard relief, so the negotiation efficiency and success rate are high.

As can be seen from Table 4, the bigger the ∆*T*, the more negotiation rounds there will be, and the lower the negotiation success rate and negotiation efficiency. The reason is that if a certain type of public continuously requires others to provide more resources and services, but does not provide more resources and services itself, it will be difficult to build an alliance, so the negotiation efficiency and success rate are low.

#### 5.2.2. Experiment 2

Experiment 2 verified that our proposed method can achieve better negotiation efficiency by dynamically changing participation attitude values. The reason is that the traditional negotiation-based alliance formation method does not take into account the heterogeneity of agents, but makes consistent assumptions about the negotiation attitudes and resources of all agents. The negotiation is conducted in the order of agent ID number.

Our method proposes that the participation degree will be dynamically calculated in real time according to the agent's historical participation performance and the resources owned by the agent. The agent dynamically adjusts the participation degree, and chooses agents with higher participation degrees to negotiate and build alliances. This is also in line with the actual situation of emergency relief for natural hazards. The public can dynamically decide whether to join or leave an alliance according to their participation attitude and resources.

#### 5.2.3. Experiment 3

Compared with Reference [37], our method is only slightly lower in the negotiation success rate, and has better performance than that of the Reference [38]. This shows that our method has better negotiation efficiency and higher agent utility than the other two methods, and is more suitable for

large-scale multi-agent systems. This coincides with the fact that many public subjects participate in natural hazard relief. With the increase of the public scale, the number of interactive objects also increases, which greatly improves the possibility of forming alliances. Therefore, our method is practical.

### **6. Conclusions**

Although the government is the main body of emergency management, our era is an unprecedented stage in which natural hazards frequently happen, so relying solely on the government's efforts to deal with natural hazards is almost impossible. We must actively encourage the public to participate in natural hazard relief. Through this research, we draw following conclusions.


Based on this, to encourage the public to participate in natural hazard relief, we put forward the following countermeasures and suggestions.

On the one hand, foster a sense of public participation. The government makes use of traditional and modern media, such as radio, television, portal site, government websites, mobile phone short messages, WeChat, official accounts, and short videos, to widely publicize the knowledge of crisis management. Meanwhile, disaster relief courses, lectures, training, and professional schools should be set up to cultivate the awareness of public participation, improve the attitude of participation, and enhance public participation.

On the other hand, incentives should be increased. Through preferential tax policies, financial discount policies, commercial publicity, material rewards, and other incentive measures, the government guides NGOs, enterprises, institutions, and the masses to actively participate in emergency relief. At the same time, spiritual incentives are given to the public, such as the honorary titles of "philanthropist" and "good citizen". When the utility for the public increases, the public is willing to build an emergency relief alliance.

In our future research, we will consider a different network topology and a dynamic agent resource change process in each instance of the construction of an alliance. More factors will be taken account, such as rescue time, participants' characteristics, and so on.

**Author Contributions:** Methodology, Y.C.; software, Y.C.; validation, Y.C.; formal analysis, Y.C.; data curation, Z.W.; writing—original draft preparation, Y.C.; writing—review and editing, Y.C., J.Z., Z.W., and P.R.T.; funding acquisition, Y.C. and J.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by "National Social Science Fund of China, grant number 17BGL181", "National Natural Science Foundation of China (NSFC), grant number 51679058", and "Shandong Natural Science Foundation in China, grant number ZR2019LZH005".

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
