شبیه سازی عامل مبتنی بر رفتارهای انسانی بر اساس کارایی در تخلیه اضطراری

Complex cognitive processes corresponding to human control behaviors cannot be easily inferred using (1) a logical rule-based model, (2) a statistical model, or (3) an analytical predictive model. Predicting human behaviors in complex and uncertain environments like emergency evacuation is considered almost impossible (at least NP hard) in systems theory. In this paper, we explore simulating human behaviors using affordance-based finite state automata (FSA) modeling, based on the ecological concept of affordance theory. To this end, we introduce the conceptual and generic framework of affordance-based human behavior simulation developed through our previous work. Following the generic framework, formal simulation models of affordance-based human behaviors are developed, especially for emergency evacuation, to mimic perception-based dynamic human actions interacting with emergent environmental changes, such as fire. A “warehouse fire evacuation” case is used to demonstrate the applicability of the proposed framework. The human action planning algorithms in the simulation model are developed and implemented using the Adjusted Floor Field Indicators, which represent not only the evacuee’s prior knowledge of the floor layout but the perceivable information about dynamic environmental changes. The results of our simulation study verify that the proposed framework accurately simulates human fire evacuation behavior. The proposed framework is expected to capture the natural manner in which humans behave in emergency evacuation and enhance the simulation fidelity of analyses and predictions of perceptual human behaviors/responses in the systems by incorporating cognitive intent into human behavior simulations.

Recently, the need to observe, analyze, and predict human behaviors using computer simulation technologies has emerged in public and social system design, where humans and their inherent capabilities have eluded formal analysis. For example, in a terrorism-driven evacuation situation, a fire, or a natural disaster such as The station nightclub fire in 2003 [1] (the fourth deadliest nightclub fire in American history, killing around 100 people), understanding and predicting how a human will respond (or, more properly, how a crowd of humans will respond) within certain circumstances will allow law enforcement agencies to be better prepared and will greatly reduce the damage associated with these disasters. However, the cost of holding practical experiments is prohibitive, and the experimental data are difficult to capture [2]. Unlike traditional physical systems, from which humans are excluded, research on modeling and simulating human behaviors in human-environment complex systems has been slow due to the challenges associated with the nondeterministic and dynamic nature of the human action/decision making processes. The cognitive processes in human behaviors cannot be described simply using logical rule-based models, since predicting human behaviors under uncertainty is a highly complex systems problem [3]. While systems theory has grown rapidly, the modeling and simulation of human-environment systems that accommodate both human cognitive models and discrete system representations have not kept pace. This is especially true for discrete event-based systems, which comprise the best computational modeling methods of predicting physical system behaviors, as they can model event occurrences and changes of system states in either a deterministic or stochastic manner [4]. However, this is not so for human-environs, in which both discrete and continuous characteristics exist together, creating a major modeling void, given that most complex systems of interest to modern society are composed of human activities. To overcome the challenge of addressing the human in systems, we start our discussion on the premise that perception guides a human’s actions towards his or her goal. To this end, two related hypotheses regarding cognitive human actions are employed: (1) humans use perception-based actions in an ecological environment [5] and [6] and (2) humans utilize goal-directed actions through prospective control [7]. The former supposes that a human makes a decision to take an action based on the perceived information he or she takes from the environment. An ecological understanding of perception-based human actions in animal-environment systems was initiated by Gibson in 1979 [5]. He defines an affordance as “a property of the environment that provides an action opportunity offered to an animal (human), either for its good or ill.” According to him, a human action is regarded as a consequence of direct perception of affordance and effectivity (an individual’s ability to take a specific action). Thus, a human makes decisions to take action based on perceived information regarding sets of affordance–effectivity. On the other hand, the latter assumes that every human action has its own objectives or intentions for prospective control (i.e., the projection of control into the future); thus, every future human action can be interpreted as an intermediate goal to be realized to reach a final goal in the future, and a human makes a plan (a series of actions) to achieve the goal and anticipates the perceptually available outcomes and opportunities to take the series of actions that advance the plan. From the viewpoint of systems theory, there are two kinds of approaches to modeling human behaviors: the experimental modeling approach and the formal modeling approach. The former method builds a model using an experimental monitoring of human behaviors through human-in-the-loop simulation [8]. This approach is seen as lacking in generality and completeness of experimental results due to its simplified and controlled laboratory conditions [14]. On the other hand, the latter is an attempt to represent the qualitative and dynamic nature of nondeterministic perception-based human behaviors using quantitative formal models. Kim et al. have advocated affordance-based descriptive formalism for modeling complex human-environment systems [9] and [10]. Formalism describes a system as a set of discrete states and regards the transitions between states as triggered by certain human actions leading to the next states in the computable models. In our previous conceptual investigation of the overall system architecture and generic functional components of an affordance-based simulator, we employed the modeling formalism of Kim et al. as the basis of our formal scheme [11]. The objective of this paper is to develop and verify an agent-based formal simulation framework of affordance-based human behaviors in emergency evacuation situations using formal modeling methods. We formalize perception-based human evacuation behaviors into a simulation model via the affordance-based FSA model, agent models, and a human action plan. Specifically, an algorithm of human behavior logic is presented. Using this algorithm, each human agent establishes and schedules a series of actions based on the dynamic perceptual properties of affordance and effectivity to reach the goal state. The proposed simulation framework is illustrated using a Warehouse Fire Evacuation (WFE) problem. The simulation results of a few different scenarios are studied to demonstrate the model’s capacity to solve the considered problem. When modeling and simulating human behaviors, other human attributes (such as emotions, cultures, knowledge levels, and social factors) need to be considered along with the perceptual properties. In particular, social factors such as interactions and communications within and between groups of people should be considered in human behavioral simulation models, especially in the case of an emergency evacuation [12] and [13]. In this research, however, our perspective on human behavior at this stage is limited to individual decision making with human perceptions rather than more complex problem domains involving human interactions and communication. Thus, only limited communication among human agents within their perceivable (or communicable) ranges is considered in the presented simulation framework. Other human attributes, such as crowd psychology and swarm intelligence, will be considered more specifically in the future. The framework and simulation models proposed in this paper will provide a predictive analysis capability for the design of human-involved systems, especially for emergency evacuation design and control. The remainder of this paper is organized as follows: Section 2 provides the related work and background of this research. Section 3 reviews the overall system architecture of affordance-based human behavioral simulation. Section 4 presents a formal simulation framework of affordance-based human behaviors using the WFE problem. We demonstrate the applicability of the proposed framework and the capability of the simulation models by running a fire evacuation scenario in Section 5. Finally, we conclude this paper with a discussion of possible directions for future research in Section 6.

Emergency evacuations from disastrous situations represent a kind of complex human-environment system in which humans dynamically interact with environmental elements. In this research, we have explored a novel simulation methodology for emergency evacuation situations that incorporates the ecological concept of affordances. The modeling and simulation architecture proposed in this paper provides powerful capabilities for planning, scheduling, and executing human actions in prospective control over the simulation because it incorporates perception-based human decision-making processes into the discrete event-based system structures. We have demonstrated the applicability of the proposed framework by a simulation study using the illustrative example of the warehouse fire evacuation problem. The simulation capabilities of the proposed evacuation application have been demonstrated by being tested through simulation performances using different space layouts and numbers of system agents. The simulation model presented in this paper is agent-based as well as perception-based. It allows a human agent to make decisions based on perceived affordance given by surrounding environment. Since the proposed framework interprets human actions as a set of perceivable action opportunities, it can dramatically reduce the complexity of human behavior (for example, in the evacuation model, agents do not have many alternatives, they can only decide which cell they should occupy next) and give a new perspective of simulating human behaviors in systems. The human agent (evacuee) in the demonstrated simulation model has its own algorithm, the “minimum cost analysis algorithm,” which generates evacuation plans based on both prior knowledge of the floor layout and what it perceives in the perceivable boundary. The static and dynamic floor field indicators are applied to represent the human planning algorithm, and they are summed up to create the AFFIs indicating human behavioral patterns in the space. The proposed agent model can regenerate evacuation plans whenever its perceivable surrounding environments change. This simulates realistic human behaviors, which are dependent on perception-based decision-making, as asserted by ecological psychologists. We have implemented our simulation framework using the AnyLogic© simulation package. Our WFE problem was developed using FSA (cellular automata), a popular approach to system representation, with computability and scalability. The proposed framework is applicable to other problem domains [9] and [10]. Kim et al. provide manufacturing control and highway driving examples using affordance-based FSA formalism in [9] and [10], respectively. The examples developed using FSA formalism have computability and scalability [38]. The environmental model offers information on dimensional states, while the agent models indicate individual states under specific conditions. The combined FSA model should be able to simulate highly complex situations by including environmental and multiple agent dynamics simultaneously because the state transitions are all under the pre-defined conditions described in the FSA model. This will give us a systematic procedure by which to build affordance-based simulation models in complex human-involved systems for the simulation of human actions and the design of human-involved systems. While we have focused on the framework and human perception (a major function) in this work, the underlying assumption in this paper is that the agent-based simulation framework considers only perception-based action, which implies the ecological properties of affordance and effectivity, rather than the social factors that might affect human decision making. However, many problems remain to be overcome. The proposed model lacks the reality of an actual fire situation and should thus be used to model more realistic scenarios. Our warehouse layout, for example, is much simpler than actual warehouse layouts. In addition, although we suppose that the evacuees within each others’ PBs can affect each other, this does not necessarily mean that the social behaviors of the human agents have all been modeled. Social factors such as gender, race, age, and physical attributes were not considered in this research. We will consider these human social attributes (e.g., social psychology, emotions, cultures, and knowledge levels) in future research. We shall also have to validate the simulation results through human experiments in a suitable task environment (e.g., in a virtual reality). Using the proposed modeling and simulation framework for a variety of applications (such as pedestrian moving and building evacuation) will further the breadth and depth of the human-integrated systems and simulation areas containing complex and multiple human activities. The deliverables in this research are expected to merge system engineering technologies with human factors and generate a novel informational technology that can be used for future disaster simulations and homeland security management.