Agent-based Modeling and Simulation for Evacuation of Landslides Natural Disaster

Landslides are natural disasters that pose a threat which is quite high in the area of Batu, East Java, Indonesia. The occurrence of landslides has a negative impact on environmental damage and even fatalities. These impacts can arise due to a lack of planning in disaster management preparedness. Therefore better planning is needed to minimize the negative impacts that arise. Improvement of planning can be done by conducting evacuation simulations. However, the existing evacuation simulation is still static with one scenario that is done repeatedly. Therefore, a more dynamic evacuation simulation is needed to represent the various parties involved in it and to apply various scenarios. Such dynamic simulations can be facilitated using agent technology. Agents can describe autonomous behaviour and can communicate in their environment to achieve a goal. Apply the capabilities of these agents by modelling and simulating the evacuation process can provide an illustration for a more dynamic process of landslide evacuation. This research presents an agent-based landslide evacuation model and the simulation results from this model. The results are concerned that, by using agent technology can apply simulations with various conditions. So with these results can be used as a reference in the handling of natural disasters that occur landslides. Keyword : Goal, Agent, BPBD


Introduction
Landslides are one of the disasters that often occur in Indonesia and are still vulnerable in the future. One area in Indonesia that is vulnerable to landslides is in Batu City, East Java. The vulnerability of landslides that occur due to various factors. There are two types of factors causing potential landslides [1]. The first factor is natural factors which include rainfall with high-intensity several days, the slope on the land, the geological conditions of the land, the existence of steep cliffs and the depth of the weathering. The second factor is management factors caused by human behaviour such as land use, infrastructure conditions, and settlement densities. All these factors make Indonesia very vulnerable to landslides and have high potential to cause casualties.
According to the records of Badan Nasional Penanggulangan Bencana (BNPB) in the last ten years, there have been 4,898 landslide incidents throughout Indonesia which claimed up to 1,865 lives. Meanwhile, in Batu City, according to Badan Penanggulangan Bencana Daerah (BPBD), 27 recorded landslides have occurred that caused 28 victims and 89 points of damage during 2018. The number of victims caused by landslides still has the potential to recur in the future. So need a solution to reduce the risks posed.
The emergence of negative impacts caused by landslides according to BNPB is the lack of awareness, alertness, and preparedness in dealing with disasters [2]. Therefore, good planning is needed in terms of increasing alertness and preparedness for the estimation of facing disasters, especially landslides. The planning process for disaster management can be improved efficiently by conducting evacuation simulations [3].
Evacuation simulation in the planning phase of a landslide disaster management is expected to reduce the negative impacts caused. However, evacuation simulations in Batu City currently still apply static simulations with one scenario that is repeated. With static simulations, it is not enough to describe the actual evacuation process. That is because a disaster event is an unexpected event so that various dynamic scenarios in the simulation are needed to provide more overviews for handling landslides if it happens.
The technology that can be used to address these needs is agent technology. Agents are part of a computer system that can act flexibly and autonomously in their environment to achieve design goals [4]. The ability of an autonomous agent makes it independently possible to make decisions about what must be done to meet the goals [5]. Besides, agents can also dynamically cope with unexpected changes [6]. In practice, agents can interact with other agents to achieve individual goals and global goals [7]. The ability of the agent is the basis of several studies that have been conducted in the application of agents for natural disaster management [8]. More specifically, research on evacuation simulations has also been carried out [9][10] [11].
Using the capabilities possessed by agent technology, all parties involved in the process of evacuation of landslides can be represented as agents. Also, the process of coordination between parties in evacuation can be simulated. The results of the simulation carried out, can be used as one of the references in handling the evacuation of ground disaster that occurred. So in this study, the authors analyzed the evacuation process of landslides in Batu City. The analysis was conducted based on the results of interviews with the BPBD of Batu City in handling disasters so far. The analysis results obtained related to the parties involved along with the interactions conducted are modelled using the Prometheus methodology. Furthermore, the model is simulated using several conditions that may occur during the evacuation. The simulation process is implemented using the Java Agent Development Framework (JADE).

Study Area
Batu City is one of the regions in Indonesia located in East Java Province. This area has a majority of land slope and rainfall is quite high. Geographically this area is located at an altitude of between 700 -2,000 meters above sea level with an average altitude of 800 meters. So that some regions in this area have a slope of 25% to 40%. This condition causes most of the Batu City area to potentially experience landslides (Fig. 1). Following the catalogue of villages and kelurahan prone to landslides issued by BNPB in 2019. There are 20 villages or kelurahan in 3 sub-districts in Batu City that have the potential to experience landslides in both high and moderate hazards [12].

Data and Method
The data used as a basis in this study were interview data with BPBD of Batu City. Interviews were conducted to explore various information related to landslides in the Batu City area. There are several speakers as objects of information retrieval, including BPBD secretary, head of the Prevention and Preparedness Section and the head of the Emergency and Logistics Section. Interviews conducted in this study used an interview approach with general instructions [13]. Apart from the results of interviews, modeling and simulation conducted are also based on the disaster contingency planning documents owned by BPBD [14].
The results of the analysis of all instruments obtained from BPBD, there is information related to landslides that occurred in Batu City. The information included locations in the Batu City area that was prone to landslides. Whereas for disaster management, BPBD refers to Batu City Regional Government Regulation No. 2, which was set in 2015 [15]. The regulation explains that a situation is called a disaster if an event or series of events becomes a threat or disturbance to the life and livelihood of society. So that the incident had an impact on the occurrence of casualties, property losses, environmental damage, and psychological impact. Referring to the regional regulation, the handling of landslides by BPBD of Batu City is divided into two types: 1) Handling of landslides as a threat, which does not affect society or settlement. In handling this condition after BPBD gets from Tim Reaksi Cepat (TRC). Furthermore, BPBD will contact the Tentara Nasional Indonesia (TNI) / Polisi Republik Indonesia (POLRI) to secure the location of the incident as well as Pekerjaan Umum (PU) to clean the scene if there is material that needs to be cleaned. 2) Handling of landslides as a disaster, which has a direct impact on society or settlement so that the evacuation process is needed. Handling landslides with this condition involves more parties. After the BPBD gets confirmation from the TRC, the BPBD will directly contact the TNI/POLRI, PU, Search and Rescue (SAR) team and the HEALTH team to carry out the evacuation process for victims.
All parties involved in handling the evacuation have their respective duties and responsibilities. These parties and their duties and responsibilities are as follows: 1. BPBD of Batu City. Acting as a coordinator for handling landslides. The involvement of all parties in handling evacuation is to fulfil the main purpose of evacuation. The main objective that must be met is to move the victim from the disastrous area to the post that has been provided. The victim transfer post is divided into two, namely the evacuation post and the health post. Evacuation posts provided by BPBD in the case of landslides are the closest and safe areas from the affected locations. Locations that are usually provided are like village halls or other locations where refugee tents might be set up. As for the health post, BPBD of Batu City has determined four hospitals to be used as a reference location. These hospitals include Baptist Hospital, Dr. Etty Asharto General Hospital, Bhayangkara Hasta Brata Hospital, and Karsa Husada Hospital, all of which are located in Batu City. The location of the incident and the available posts are illustrated in Fig. 2.

Prometheus Modeling
Prometheus Methodology is an agent-based methodology from several methodologies that can be used in agent modelling [16]. This methodology is one of the Agent-Oriented Software Engineering (AOSE) methodologies which has documentation and is widely researched and published. There are several phases in the Prometheus methodology that can help to model the agent system that will be created. The phases include system specifications, design architecture and design details in which each phase adds system entities and design artefacts [17]. The main phases in the Prometheus methodology are explained as follows [16]:  System Specification, the initial phase of the Prometheus methodology that focuses on the basic function of system identification which includes inputs in the form of perceptions and outputs or actions.  Architectural Design, the phase in which the output of the previous phase is used to determine the agents that interact.  Detailed Design, focusing on the internal nature of each agent and how the agent completes all his tasks in the system.
All of these phases can be applied to do agent modelling using the Prometheus Design Tool (PDT). PDT is a tool that supports the implementation of the Prometheus methodology. In addition to supporting agent-oriented system design, PDT is also able to automatically generate designs into code frameworks from agent programming languages such as JACK. Then the code can be edited, compiled, tested, debug and packaged to be implemented [18].
Evacuation simulation modeling in this research uses several notations contained in PDT. These notations can be seen in Table 1. Message sent from one agent to another and contains Belief-Desire-Intention (BDI)

Simulation
The simulation is done by using the results of modelling that have been done previously using the Prometheus methodology which is explained in the design model section. Modelling results that include scenarios from the system to the agents involved and their interactions are simulated using the Java Agent Development Framework (JADE). Each individual agent that has been modelled will be applied into JADE by taking into account the capabilities of each agent according to the design details that have been outlined. The process in the simulation is done by considering scenarios based on the landslide evacuation process obtained from the results of the previous interview.

Agent Characteristics Evaluation
Evaluation is carried out to determine whether the agents involved can operate according to the defined characteristics of the agents. Several criteria are used based on the agent characteristics that have been defined by Dam [19]. Criteria are selected by adjusting the criteria that agents may need in the simulation being run. Some of these criteria are explained in the following table:  c) The agent can simultaneously complete more than one goal All criteria of the agent characteristics described will be assessed by each agent with the following rating ranges: 1. High (H), the agent has all the criteria that have been determined.
2. Medium (M), the agent has only two of all criteria.
3. Low (L), the agent has only one of all criteria.

Result and Discussion
The results of the analysis obtained are used as a basis for modelling, simulating, and evaluating the characteristics of all agents involved in the simulation process.

Design Model
There are several actors involved in the evacuation simulation made. These actors are then modelled as agents. All agents will act following the scenarios that have been made in Table 3. Scenarios were started by BPBD agent as coordinators of landslide disaster management. BPBD agent contacts TRC agent to monitor landslide locations as a basis for determining disaster status. After the information is obtained and the status of the disaster requires an evacuation action, the BPBD agent immediately follows upon the information. BPBD agent followed up by contacting TNI/POLRI agent to carry out site security and evacuation obstacle handling if there was an evacuation assessment by the victim. After that, the BPBD agent ordered PU agent to prepare evacuation routes and clean up the landslide material. Then proceed with the order to rescue and evacuate victims to the SAR agent by BPBD agent. Subsequent instructions were delivered by BPBD agent to the HEALTH agent to identify victims and determine referrals for VICTIM agent affected by landslides. Evacuation scenarios carried out in simulations can run well if the main goal of the scenario is met. The main goal of the scenario is evacuating victims of landslides. In achieving the main goal, several sub-goals must also be fulfilled. The main goal and sub-goals in this study can be seen in Fig. 4. The figure explains the goal overview of scenarios that have the main goal of landslide evacuation. For the main objective to be achieved, several sub-objectives must also be fulfilled, such as evacuation coordination team, disaster status determination, and other sub-goals.

Fig. 4. Goal Overview of landslide evacuation
The main scenario is carried out by following the evacuation of the VICTIM agent. Evacuation is based on the results of identification carried out by the HEALTH agent. Using the identification results, the HEALTH agent determines the next action for the VICTIM agent. If the VICTIM agent is willing to be evacuated, the HEALTH agent orders the SAR agent to carry the victim to the designated post. Whereas if the VICTIM agent refuses to be evacuated, the HEALTH agent coordinates with the BPBD agent for handling these obstacles. Coordination of the constraints obtained by the BPBD agent is then forwarded to the TNI/POLRI agent for further action. All of the evacuation scenario models can be seen in Fig. 5.

Evacuation Simulation
The simulation is carried out by using an evacuation scenario under normal conditions of action. This condition was when the VICTIM agent was willing to be evacuated. In this simulation, BPBD agents contact all agents involved in the evacuation process. The simulation process is completed when the objective is fulfilled, namely the evacuation of victims by the SAR agent under the instructions of the HEALTH agent. The results of the application of the evacuation simulation using JADE can be used to determine the interaction of all agents in completing the purpose of the scenario. All agent interactions involved in the simulation can be seen in Fig. 6.

Agent Characteristics Evaluation Result
Based on simulations that have been run using JADE, it can be seen the internal characteristics of each agent involved. These characteristics are known based on the results of evaluations that have been carried out using predetermined criteria. The results of the evaluation of all the agents involved are presented in Table 4. Based on the evaluation results it is known that all agents have a high value in several characteristics. High-value characteristics are autonomous with a percentage of 100% High because all agents involved have high values on these characteristics. Then the proactive characteristics with the percentage of 71% High and 29% Medium, because five of the seven existing agents have high values while the other two agents get medium values. Furthermore, the characteristics of social abilities with a percentage of 100% Medium. As for the percentage value of the reactive and concurrency characteristics, the scores were quite low in all agents. Reactive gets a percentage value of 29% medium and concurrency of 57% medium. The low yield of these characteristics is due to several criteria that are not needed by the agent in the evacuation simulation.

Conclusion
This research presents a simulation of a landslide evacuation disaster using agentbased technology. Agent-based simulations can provide some illustrations related to evacuation activities. Several conditions related to the evacuation process can be simulated compared to previous simulations. The agent involved in the simulation can also fulfil the characteristics possessed by the agent. This can be seen from the agent characteristics evaluation result. Almost all agents have a high enough value on several characteristics such as the autonomous, proactive and social ability. These results indicate that overall the agent can interact and coordinate in a system to solve a problem. So that the agent-based simulation that has been done can illustrate a more dynamic evacuation process for BPBD in handling landslides. That is because the simulation can illustrate the two-way interaction process between agents, and run scenarios with various conditions according to the needs of the field during the evacuation process so that affected victims can be minimized.
In the future, the application of evacuation simulations can add some behaviour to the agents involved. For example, by adding behaviour for the VICTIM agent and the HEALTH agent in the identification process. So that the negotiation process in evacuation can be more dynamic. Also, the time needed in the evacuation process is worth considering. So by applying more complex interactions in the evacuation simulation, it is expected to provide a broader overview of the evacuation process in a more tangible way.