Searching efficient plans for emergency rescue through simulation: the case of a metro fire

Transcription

1 Cogn Tech Work (2004) 6: DOI /s ORIGINAL ARTICLE Nikos Zarboutis Æ Nicolas Marmaras Searching efficient plans for emergency rescue through simulation: the case of a metro fire Received: 5 June 2003 / Accepted: 20 January 2004 / Published online: 23 March 2004 Ó Springer-Verlag London Limited 2004 Abstract In this paper we present the way we modelled and simulated a metro system in the case of a tunnel fire, and discuss the ways this simulation may support the search for efficient rescue plans. The metro system was modelled as a complex adaptive system, comprising four interacting and co-evolving subsystems: (i) the fire and the released smoke, (ii) the group of passengers, (iii) the technological system, and (iv) the metro personnel. Based on this model, an agent-based simulation was developed. This simulation provides an appropriate dynamic representation of the designer s problem space, enabling him (i) to apprehend the critical dependencies and invariants of the system under consideration, (ii) to identify the features that should characterise the designed emergency rescue plans, and (iii) to assess their efficiency. To demonstrate the usefulness of the adopted approach for the design of an efficient emergency rescue plan, the results of two experiments exploring alternative sequences of the metro personnel s actions under different circumstances are presented. Keywords Metro systems Æ Emergency rescue plans Æ Complex adaptive systems Æ Agent-based simulation Æ Co-evolution 1 Introduction In the last 50 years, in railway systems there is about one severe accident every 2 years. In rail or metro tunnels in particular, several incidents have questioned the efficiency of traditional approaches to evacuation design: N. Zarboutis Æ N. Marmaras (&) School of Mechanical Engineering, National Technical University of Athens, Heroon Polytechniou 9, Zografou, Greece marmaras@central.ntua.gr In 1995 a fire in a metro tunnel at Baku, Azerbaijan resulted in 292 human losses and other 168 injuries, in the worst tragedy in the history of metro systems. In 1996 a fire in a heavy goods vehicle shuttle train a special train used to transfer automobiles with their drivers on a separate car in the Channel Tunnel, resulted in no human losses but led to seven injuries, caused significant damage to the infrastructure and the rolling stock, and initiated a long debate on the safety of railway tunnels. In 1999 a fire in a railway tunnel near Salerno, Italy resulted in only four human losses and nine injuries, because the driver managed to get the train out of the tunnel. In 2000 a fire in a funicular train near Kaprun at the Austrian Alps resulted in 155 human losses as only 12 of them managed to escape from the tunnel. In 2003 at Daegu, South Korea a fire in a train (as well as in a neighbouring one, minutes later) resulted in 189 human losses and in more than 140 injuries, due to smoke inhalation. The analysis of the above incidents illustrates a direct coupling between the personnel s activity and the passengers behaviour, which being mediated by the evolution of fire, influences the overall efficiency of the evacuation process. In the Kaprun incident for example, this coupling was evident in the initial stages of the evacuation process. Rolling the train backwards and outside the tunnel could have resulted in a better performance, since the train would have been evacuated at a more convenient place (as it was the case in the Salerno incident). The presence of early evacuators in the tunnel though, led the station master to instruct the driver not to roll the train backwards. On the contrary, in the Channel Tunnel case, the personnel s choice to stop the train next to an emergency exit (the train had the capability to move, even though it was on fire), and the fact that the passengers remained in the car for as long as it was necessary, led to an efficient evacuation and ultimately to a success story. As the Channel

2 118 Tunnel Inquiry Report concludes ( gov.uk/stellent/groups/dft_railways/documents/page/ dft_railways_ hcsp p. 51): Overall therefore the performance of the Eurotunnel system, set against the key safety criteria, illustrates both the unavoidable complexity of its interrelated elements and the absolute importance of the actions of operators. While we consider that the existing design of the HGV shuttles can be utilised in a way that secures the safety of people, all designs may be capable of improvement as technology develops. In the future Eurotunnel should not neglect to explore other options. We can, therefore, conclude that in railway operations involving fire in tunnels, the coupling between the triplet passengers fire personnel, given the technological, physical and contextual environment, is of great importance, and should be considered as a whole in order to design effective evacuation plans. When referring to a socio-technical system, a plan designates an organisational artefact determining the required actions to achieve a goal, the humans and the machines that will carry out these actions, and the sequence by which these actions will be performed. The more the system under consideration is governed by a stable transformation function, i.e. the inputs to the system, the laws governing the transformation of these inputs and the desired outputs are known in advance, the easier the conception of an efficient plan is (Nathanael et al. 2002). Consider for example the case in which the goal is to produce a mechanical product, consisting of four parts, manufactured by three machine-tools and assembled in a predetermined way, and where the manufactured times of the parts are known, and a sufficient number of workers is available. In this case, the conception of an effective production plan is a problem that can be solved rather easily. In contrast, the more the transformation function governing a system is not stable, i.e. the inputs to the system, the laws governing the transformation of these inputs and the desired outputs cannot be accurately defined in advance, the more difficult the conception of an efficient plan is. Consider for example the case of an emergency rescue plan for a metro train on fire. The goal of such a plan is to facilitate the passengers escape towards a safety place. The propagation of the fire and smoke, the number of passengers, their physical abilities, etc., are not known in advance, and consequently the actions to be carried out and the sub-goals to be achieved can hardly be predetermined. Designing an efficient organizational artefact for such systems is a complex problem solving task. The complexity is mainly due to the nature of these systems as the designer has to deal with a system characterised by: A large number of elements, e.g., the passengers, the metro personnel, the metro s technological system, the fire and the smoke. These elements interact either physically or by transferring information, and the interaction is dynamic. The interaction is fairly rich, i.e. any element influences and is influenced by a multitude of others. The interactions are characterised by non-linearity. The interactions usually have a fairly short range. There are loops in the interactions (feedback and recurrent loops). The system interacts with its environment, i.e. it is an open system. Each element in the system is ignorant of the behaviour of the system as a whole. In other words, the designer deals with a complex system (Simon 1996; Cilliers 1998), whose behaviour is rather impossible to be modelled and predicted with conventional analytic approaches. In cases where the system can be observed, the designer can form a fairly accurate idea about the mechanisms underlying the system s functioning, and speculate on the effects of alternative design solutions, or test them. In contrast in cases as the design of a plan for emergency rescue, the system under consideration cannot be observed. The designer can form an image about the functioning of the system only under normal situations, as well as from descriptions and analyses of past events. Modelling and computer simulation, using the formalisms originating from Complexity Science, may provide a valuable tool for the designer to acquire a better understanding of the complex system under consideration and to investigate the effects of alternative design solutions (Dugdale et al. 2000). In this paper we present the way we modelled and simulated a metro system for the case of a tunnel fire, and discuss the ways this simulation may support the search for efficient rescue plans. The present study is different from previous attempts of developing simulation tools for the case of fire mainly in two respects: (i) of the simulation s scope and (ii) of the way the whole system was modelled. In fact, the scope of most of the existing simulation tools is to facilitate engineering decisions regarding mainly the space layout (how to eliminate bottlenecks and avoid congestion phenomena, which are the optimal evacuation paths etc.). Furthermore, most of the existing simulation tools do not model the whole system as a complex adaptive system. Consequently, the human agents are modelled as stochastic agents who move according to rules that depend only on the space constraints and the evolution of the fire; the environmental changes caused by decisions taken by human agents far from the place of the fire (e.g. firemen or other specially trained personnel) are not considered. For a comprehensive up-to-date review of fire related computer models see at firemodelsurvey.com.

3 119 2 Descriptive analysis of a metro system on fire In this section we present a descriptive analysis of the studied metro system, when a train is stalled in a tunnel between two stations, due to a fire. This analysis was carried out considering a specific metro system; for confidential reasons, the studied metro system is not identified. Adopting a systemic view, we can identify four dynamically interacting wholes (i.e. systems): (i) the fire, (ii) the passengers, (iii) the technological system, and (iv) the metro personnel. 2.1 The fire The fire in a metro railway system can be the result of multiple causes, internal or external to the trains. Thus, the fire may be initiated due to some electrical or mechanical failure, or it can be the product of a wider failure, such as a derailment, a collision or a crash on an obstacle. Furthermore, it may be caused by external events such as a natural phenomenon (e.g. an earthquake), arson, sabotage etc. (Diamantidis et al. 2000). Finally, the fire may be burning on or underneath the train. Fire is a complex phenomenon (Xue et al. 2001); it propagates on a flame-spread rate and raises the overall temperature, through the thermal energy that is being released in the environment (the temperature may raise up to 500 C within even 20 min (Cheng et al. 2001)). Fire propagation depends on multiple factors, such as the ignition point, the proximity of flammable materials (e.g. engines, batteries, baggage etc.), airflows (Gann et al. 1994; Drysdale 1997) etc. Finally, fire releases toxic smoke on a specific release rate, propagating on a specific diffusion rate, depending on the geometry of the tunnel, the properties of the materials involved and the airflows. 2.2 The passengers The number of passengers on a train may vary from few to many. They may be of different ages and physical abilities. The main risks that a passenger runs in the tunnel stem from (Purser 1992, 2000): The toxic potency of the environment (smoke inhalation leading to asphyxia). The thermal energy released by the fire, causing serious burns and/or fatalities. The power tracks, causing electrocution upon contact with them. The other passengers and the limited space to move, causing falls, injuries, tripping etc. In a train full of passengers, fire may create out-of-control situations. Fire and smoke affect not only the well-being of the people exposed to it, but their behaviour over time as well (Purser 2000). In such critical situations passengers on the train face a highly constrained world, both physically and temporally, where they should come up with the appropriate actions in order to save themselves. At each point in time and space, each passenger is expected to maintain only a basic level of control, if any at all. Planning and situation assessment within a crowd are rather limited at short-term goals, and involve instinct-based reflexive acts (Le Bon 1971). More specifically, passengers exhibit preference in pursuing goals with short-term effects regarding their next move, rather than goals that would seem as better options in the long-run (Helbing et al. 2000). Human behaviour is rather event-driven, as if it lacked meta-cognition. Such behaviour resembles to what Hollnagel (2002) refers to as scrambled mode of control. As the author states in such mode, the next action is in practice unpredictable... this corresponds to trial-and-error type of performance where cognition plays little or no role.... Consequently, it could be assumed that the passengers who are evacuating the train will constantly tend to avoid the fire and to join groups, instead of planning their way out of the tunnel. Each action will be based on what can be done in order to avoid imposed constraints, by exploiting the immediate affordances of the environment. 2.3 The technological system The technological system comprises all the elements that enable or facilitate metro operation. It is spatially distributed in multiple areas of responsibility, comprising the trains, the tunnels hosting the transition of the trains, the stations, the operations control centre (OCC), consisting of multiple vigilance and remote control equipment, and the depot, where parking and maintenance tasks are carried out. Operation and control is performed either locally from the stations or centrally from the OCC, with a preference to the centralised control. The tunnels are of one-tunnel-double-track type, with two 80 cm sidewalks on their waysides, having an average length of 1 km. Each train is 106 m long, it has a capacity of 1,030 passengers and runs on direct current provided by the power tracks installed in the middle of the tunnel, along the normal lanes. It consists of six cars each having four double-doors, about 2 m wide, equipped with manual emergency handles. The cockpit contains the telecommunication equipment as well as indicators about the maximum speed allowed (i.e. the automatic train protection indicator), the status of the doors (i.e. whether they are open or closed), and the eventual delays in the itinerary. Tunnel ventilation is achieved through fans, installed in the lips of the tunnel (i.e. at the stations), as well as in the inter-shafts, that are shafts linking the tunnel with

4 120 the outer environment and are placed close to the middle of each tunnel. The fans can be operated at various speeds either as exhaust or as inlet fans (i.e. extracting the smoke or bringing in clean air respectively), except for the fans at the inter-shafts which act as exhaust fans only. Monitoring and control is carried out primarily through the OCC. The OCC is equipped with fixed-line diagrams, displaying information about the relative position of every train on every line (but not its precise position), its identification number and the status of the signalling system. Power in the network is monitored through the power mimic diagram and controlled by means of a computer system, permitting the shutdown of the power on any part of the line and any other area within the metro. The ventilation system is controlled remotely, either in manual mode (i.e. control of each fan separately) or through pre-defined settings (i.e. simultaneous activation of multiple fans on predetermined settings regarding speed, exhaust or inlet modes). Monitoring of the public areas is achieved through a closed circuit TV system, continuously sending real time images to a set of monitors in the OCC. Finally, communications are enabled through the telecommunication system, comprising all the systems that connect the OCC with the entire network (i.e. the train cockpit, the stations and individual agents carrying hand-portables, either individually or through a general call), as well as with external agencies such as the fire brigade or the police. Local level monitoring and control in the vicinity of the stations can be carried out from the stations. More specifically, the movement of the trains just before and after the station, and all the public areas can be monitored through displays and the closed circuit TV system installed in the station. The ventilation system can be activated for the nearby tunnels, but only through predefined settings. The power can be shut down on the parts of the line that run through the areas of the station platforms. Finally, the fire protection equipment is activated from each station. The telecommunication system of the stations comprises both hand-held and fixed devices that enable direct communication between the stations and the OCC. 2.4 The metro personnel The metro personnel, who retain the responsibility of the operation of the metro system consists of multiple agents located in the tunnel, the nearby stations (i.e. the two stations around the area of the incident) and the OCC. After the onset of a tunnel fire, the following human agents would normally be involved: The train driver, who is the person in the closest proximity to the fire and the passengers. The traffic regulator in the OCC, who is responsible for directing the movement of trains within the tunnel. The power controller in the OCC, who is responsible for controlling the power. The information controller in the OCC, who is responsible for monitoring the closed circuit television monitors of the network. The safety controller in the OCC, who is responsible for the coordination of the safety personnel of the system. The network controller in the OCC, who is responsible for the coordination of all the agents and all the decisions that are taken in the OCC. The station masters in the stations, who are responsible for the local monitoring and control. Finally, the ticket sellers may also be involved in the evacuation process. 2.5 The problem space of designing an emergency rescue plan The actions that can be carried out to facilitate the passengers rescue, and the persons who may perform them, are as follows: Shutting down the power of the line, to prevent passengers from being electrocuted. This action may be performed at the OCC by the power controller, at the stations by the station masters, or at the tunnel by the train drivers, or even by any other person having access to the respective switches. Adjusting and activating the ventilation system in order to bring clean air and extract the toxic smoke from the area to be evacuated. These actions may be performed at the OCC by the power controller or at the stations by the station masters. Opening the train doors. This action may be performed by the train driver (if not injured), as well as by the passengers. Informing the passengers. This action may be performed by the train driver (if not injured). Directing the other trains away from the incident area. This action may be performed by the traffic regulator at the OCC, through instructions broadcasted to the train drivers. The train driver, if conscious, may be the only agent who has an immediate recognition of the situation. The personnel in the OCC and the station masters can form an image of the evolving situation partly through verbal communications with the train driver in the tunnel and partly through the mimic diagrams. It is important to note that the actions performed remotely from the OCC or the stations, act on the passengers indirectly (i.e. changing the environment in which passengers are forced to adapt). Therefore, the existence of adequate feedback is indispensable and the efficiency of the whole rescue operation is very sensitive to the communication between the train in the tunnel, the OCC and the personnel in the stations.

5 121 All the above actions, which facilitate the passengers rescue, are interdependent and their implementation involves the recognition of multiple effects on the same or on multiple agents. For example, opening the train doors immediately after the fire is detected would jeopardise passengers lives risking electrocution. Cutting off the power immediately, would remove the danger of electrocution but would increase the probability of stalling other trains in the tunnel, thus jeopardising the lives of the passengers on other trains. Guiding the other trains, then cutting off the power before opening the train doors, would lead to a situation where the tunnel would already be full of smoke, and evacuation might be more difficult. It should be noted that the ventilation system cannot be operated at velocities greater than 11 m/s, since people will experience difficulty in moving (Cheng et al. 2001), as the airflow will be very strong. Furthermore, if the metro personnel delays opening the doors, it is highly probable that passengers will make use of the manual door release handles. The above description of the metro on fire shows that the designer of a rescue plan should dispose an appropriate representation of a complex dynamic problem space. Such a representation will enable the designer to obtain a better understanding of the dynamics of the system and experiment with alternative rescue plans, i.e. to examine the efficiency of alternative task allocations as well as sequences and times of carrying out the actions facilitating passengers rescue. In order to provide the designer with such a representation of the problem space, the metro system was modelled as a complex adaptive system, and an agentbased simulation was developed. The developed model and simulation are presented in the following sections. 3 Modelling the metro on fire as a complex adaptive system (CAS) 3.1 Elements of CAS theory The study of CAS has been recently granted with much attention. Based on the notion of complexity and having roots in many disciplines (Biology, Non-linear Systems, Artificial Intelligence, etc.), CAS theory offers a formal way to describe the behaviour of a complex system within a dynamic environment (Holland 1992; Bernon et al. 2001; Railsback 2001). Its focus is on the interplay between the system and its environment as well as in their co-evolution (Choi et al. 2001) Evolution in a CAS Evolution in a CAS can be described by a set of three elements, namely the system s internal mechanisms, the environment and the co-evolution (Kauffman 1996; Choi et al. 2001). The internal mechanisms describe the structure of the system in terms of agents that colonise the system, its dimensionality (i.e., the degree of autonomous behaviour), their interactions and the way that collective behaviour is unfolded and shaped (i.e. emergence and control). The environment which is considered external to the CAS, is dynamic and the changes that occur within it are au fond unpredictable. It provides the changes upon which the system has to adapt, in order for its constituent parts to survive. Finally, the way that the system evolves over time is described by the notion of co-evolution. According to this notion, the system reacts to environmental changes by altering its borders (i.e. adding or removing interactions among agents, excluding or including agents). Thus the system is both reacting to and creating its environment. The latter provides the stimuli to the system at its borders and in turn, the CAS reacts to these changes by altering its internal structure (i.e. by adaptation to the environmental changes). Thus, system and environment co-evolve, in a non-random way Emergence and self-organisation in a CAS The system comprises many independent but interacting agents, who perform partial actions, with respect to the common goal of the system (Bernon et al. 2001; Casti 1997, 1999). These interactions may be either direct or indirect (Bonabeau et al. 1997). In the former case, agents interact through the exchange of information and/or materials and/or energy. In this way, they alter their individual behaviours and systemic behaviour is altered accordingly. In the latter case, the indirect interactions of the agents take place through the modification of their immediate environment, affecting thus implicitly their relative movements in the problem space, which they seek to adapt Hierarchical control in a CAS Complex adaptive systems are hierarchical, comprising multiple levels of organisation, on which the behaviour of the system emerges. Individual agents interact with each other at the basic level. The product of their interactions is evident at an upper level the level of emergence. The emergent properties have to be meaningful at that level and this is assured by the imposition of constraints, top down, at the interactions at the lower level. Thus, system behaviour is unfolded bottom up and simultaneously is constrained top down (Checkland 1984). 3.2 The simulation model General properties The metro has been modellecomplex adaptive systemlaned as a CAS, comprising four interacting and co-evolving wholes; (i) the technological system, (ii) the fire, (iii) the group of passengers and (iv) the metro personnel.

6 122 The technological system comprising the equipment and the infrastructure of the system, was modelled at the level of the constraints and affordances it provides to the human agents of the system. As far as the infrastructure is concerned, the tunnel was divided in a grid of cells, each having the size of an average person. Depending on the point of the tunnel (e.g. train, sidewalk, power track, line etc.) each cell was modelled as holding environmental information that affects the behaviour or the state of the human agent that stands on it. For example, a cell may or may not contain a line, may or may not contain a power track, etc. Thus, if a passenger steps on a cell which contains a line, her/his speed is decreased; if s/he steps on a cell containing a power track the passenger loses her/his life; if a passenger faces cells that represent the walls s/he can not step on them. The fire has been modelled as a cellular automaton, where each cell could be in one of two discrete states: burning or not burning. In this cellular automaton, fire propagates at a specific propagation rate, spreading to its neighbouring cells randomly, after a given number of iterations. Each burning cell was modelled as releasing a specific amount of smoke, which is evenly diffused at its neighbouring cells. Thus, each cell holds also information about the propagation of the fire and the smoke in the tunnel. This information is dynamically updated as the fire propagates. The group of passengers comprises all the human agents that are present in the area of the incident and are not members of the metro s personnel. We assume that they form a constantly changing ecology, sharing a number of common attributes, and lacking training or experience in similar situations. Their behaviour was modelled as being rule-based type, assuming that it is only reactive and not proactive, given the absence of meta-cognition. The main rules guiding passengers behaviour are: Move only if the space around you is free. Move towards the nearest door; if it is open, exit the train; if it is closed, wait until patience time lapses, and then open the door and exit the train. When out of the train, head to the sidewalks and then to a station. If you are near the fire, move to the opposite direction. If the atmosphere is full of smoke, move towards the direction which is less charged with smoke. Don t stay isolated from the others. If alone, head towards the nearest passenger(s). Each passenger was modelled having certain age and related physical abilities (i.e. walking/running speed, visibility range, resistance to smoke inhalation). For each passenger, individual risk is calculated by the equation (Gann et al. 1994; Purser 1992): Risk ¼ C t C is the concentration of the toxic smoke present in the atmosphere and t is the time of exposure. For values above a lethal level, passengers were modelled to lose their lives. Additionally, passengers were modelled to lose their lives upon touching the power track while power is on. The metro personnel comprises the agents of the system that are responsible for the operation of the equipment and for the execution of the tasks that aim to facilitate the evacuation of the passengers. The behaviour of these agents was not simulated; it was described by a number of open variables in order to test their effects on the passengers rescue. Consequently, the actions of the personnel were modelled in an indirect fashion, that is, by the effects they have on the environment of the passengers in the tunnel. For example, when the power is cut off from the line, the passengers that step on the power track are not electrocuted; when the ventilation is activated, the distribution of smoke changes accordingly. Constraints associated with personnel s actions were also included in the simulated model (e.g. the time necessary to carry out a specific task; the inability of the same person to perform certain tasks in parallel, etc.). These constraints were identified during the analysis of the studied metro system Co-evolution in the metro system At every point in time and space, each passenger is standing on a cell. Evolution is generated by the application of the rules guiding passengers behaviour, which lead to a dynamic set of possible actions, depending on the environmental conditions and on the individual characteristics of the agent standing on the cell. Each passenger identifies the state of the cell s/he colonises as well as of the neighbouring cells (i.e. free or occupied? mounted smoke? on fire? on the train? etc.), and moves or changes her/his state accordingly. At each iteration of the simulation, the environmental parameters change (e.g. the fire propagates, the smoke concentration changes), the status of each agent changes (e.g. a passenger may lose hers/his life, acting then as a physical constraint to the others), and consequently the inputs of the rules change accordingly. Thus, the collective behaviour of the passengers is continuously updated, through modifications in the structure of both the system and its immediate environment. In this manner, behaviour is modelled as unfolding bottom-up, through the co-evolution of the elements of the whole system Simulation as a test-bed for experimentation As already stated, the scope of the simulation is to provide an appropriate representation of the problem space of the designer of a rescue plan, and to become a tool for examining the effects of alternative design solutions. Thus, the simulation was built leaving enough variables open, to be set by the designer. These variables concern either the time it will take the metro personnel to carry out actions, or the different circumstances regarding both the fire and the passengers. The following variables were left open:

7 123 The number of passengers on the train. The time it will take the train driver to open the doors, and thus the moment that the evacuation will formally begin. The time it will take the metro personnel to start up the ventilation system as well as the direction of the airflow. The time it will take the metro personnel to shut down the power line. The patience time of the passengers, before they manually open the train doors. After the completion of this time, the passengers who are closer to the fire open the doors in front of them. After a time period twice the length of patience time, the passengers open the remaining closed doors manually, regardless of the actions of the train driver. Fig. 1 A snapshot of the simulation. The screen is divided into three areas. The left area (the animation window) displays the movements of the simulated agents, i.e. passengers, fire (the smoke is not represented so that the observation of passengers movement is not obstructed). The central area (results area) presents the values of critical parameters (e.g. percentage of passengers alive; average value of inhaled smoke etc.), and three graphs that are updated dynamically, displaying the dynamic changes of some other variables (e.g. the average speed of the passengers). The right area (control area) contains graphical sliders for the adjustment of the values of the open variables Whether the driver is alive or not, and thus whether s/ he is capable of transmitting information to the metro personnel, of giving instructions to the passengers and of opening the doors. The specific car of the train that is on fire. Whether the fire is being ignited in or underneath the train. The magnitude of the fire and the amount of emitted smoke, through variables such as smoke release rate, diffusion rate, flame propagation rate, density of fire, etc. The relative position of the train in the tunnel (i.e. close to a station, between stations, under an intershaft, etc.). The simulation was developed in an agent-based simulation tool, the NetLogo 1.0. Its graphical interface (Fig. 1) permits the control of the open variables, the observation of the evolution of the evacuation, and the performance of the examined rescue scenarios. 4 Using simulation to support the design of an emergency rescue plan The developed simulation of the metro system offers to the designer of an emergency rescue plan a tool for exploring his complex problem space and for

8 124 apprehending the main features that an efficient plan should dispose. More specifically, s/he can easily test the effects of alternative sequences of the metro personnel s actions, under different circumstances regarding both the fire and the passengers. In this way, the designer can form a clear image of the invariants of the complex situation on which s/he can base the designed emergency rescue plan. The following examples of experiments aim to demonstrate the support provided by the developed simulation. 4.1 Experiment 1 Suppose that the designer of an emergency rescue plan would like to explore the effects of alternative sequences of the following actions: (i) opening the train doors, (ii) shutting down the power, and (iii) starting up the ventilation. The reason for such an investigation lies on the following contradictory perspectives: Opening the doors before the power has been cut off jeopardizes the safety of the passengers who might step on the power lines. On the other hand, opening the doors as soon as possible minimises the risks created by the smoke and the fire propagation. Opening the doors, after having the power shut down, minimises the risk of electrocution, but the evacuation will take place in a more hazardous atmosphere, because of the smoke and fire propagation. Waiting for the ventilation to be activated, could possibly offer better environmental conditions (i.e. less smoke). However, the prolonged waiting of the passengers in the train, while the fire propagates, may create panic, as well as a more hazardous environment because of the increased magnitude of the fire and quantity of smoke. We ran the simulation for the following alternative scenarios: 1. The train driver opens all the doors 5 min after the ignition of the fire. 2. The train driver opens all the doors after the power has been shut down, 8 min after the ignition of the fire. 3. The train driver opens all the doors after the power is shut down and after the ventilation system has been activated, 13 min after the ignition of the fire. The time values used for these scenarios were estimated based on the time needed to diagnose the event as well as on the actual technical and organisational constraints of the studied metro system. For these three scenarios we modified the total number of passengers (from 10% up to 100% of the train capacity), and ran the simulation a number of times until we obtained a clear trend, under the following settings: The train was stalled between the departure station and the shaft of the tunnel. The fire started at the front car of the train and was assumed to burn underneath the train. The fire was modelled as propagating randomly and releasing toxic smoke at a constant rate. The smoke was evenly diffused in the tunnel, before the activation of the ventilation. The ventilation removed smoke at a constant rate. The passengers were patient enough and did not open the doors by themselves, following the train driver s instructions, who was assumed to be alive and conscious. The number of human fatalities due to smoke inhalation, burns or electrocution was used as a measure of the efficiency of the evacuation. In this way, the designer could assess the effects of the alternative sequences of the metro personnel s actions on the safety of the passengers. The following conclusions can be drawn, based on the results of the simulated alternative scenarios (Fig. 2): Opening the doors before any other action appears to be the most efficient scenario, regardless the number of passengers on the train. Waiting for the power to be cut-off from the line and then opening the doors, does not appear to be a better scenario. In fact, the human losses that were avoided due to electrocution were fewer than those caused by the prolonged waiting in the train. The worst scenario appears to be when the doors open after the ventilation has been activated. This is mainly due to the increased magnitude of the fire and amount of smoke at the moment the evacuation starts. 4.2 Experiment 2 In this experiment we explored the effects of the limited patience time of the passengers. Three cases were considered: Fig. 2 Results of the simulations of Experiment 1

9 The passengers who are closer to the fire, open the doors in front of them at t =1 min after the fire starts, using the emergency manual handles, while the other passengers open the remaining doors at t =2 min. 2. The passengers who are closer to the fire, open the doors in front of them at t =2 min after the fire starts, while the other passengers open the remaining doors at t =4 min. 3. The passengers who are closer to the fire, open the doors in front of them at t =3 min after the fire starts, while the other passengers open the remaining doors at t =6 min. The remaining variables were set at the same values as for Experiment 1. Figure 3presents the results of the simulation for these three cases, for different numbers of passengers (from 10% up to 100% of the train capacity). The following conclusions can be drawn, based on the results of the simulated alternative scenarios (Fig. 3): Increasing the patience time of the passengers, the number of evacuated passengers decreases, regardless the number of passengers on the train. At low capacity rates, the passengers find it easier to exit the train, regardless of the length of patience time. 4.3 Overall conclusions from the two experiments Considering the results of the two previous experiments, the following conclusions can be drawn, forming a basis for the design of an effective emergency rescue plan: There is a clear trend that the sooner the evacuation begins the more efficient the evacuation will be; therefore, an efficient emergency rescue plan should ensure that the evacuation start as soon as possible. The danger of electrocution is not so critical; consequently, shutting down the power from the lines should not delay the start of the evacuation. Smoke inhalation causes the majority of the fatalities; thus an effective emergency rescue plan should find out ways to minimize this risk. At low capacity rates, the passengers find it easier to exit the train, regardless of the length of patience time. As the number of passengers increases, the efficiency of the evacuation decreases. The passengers, who delay their departure from the train or those who chose to stay in the train for a prolonged period, run more risks; consequently, metro personnel would have to ensure that the train is evacuated as soon as possible. Most deaths occur in the area around the train. As a result, passengers should be guided to move out of the area around the train as soon as possible. 5 Epilogue While designing a complex system may be relatively easy, designing for a complex socio-technical system is a much more difficult task, as it requires the appreciation of multiple nonlinearities and a compromise among conflicting objectives, dynamically changing over time. In this study, we used agent-based modelling and computer simulation as a method to model a complex system and to explore the effects that individual actions and their interdependence can have on the performance of the whole system. The effects of alternative sequences and durations of the metro personnel s actions, while varying some environmental factors, have been explored to demonstrate the usefulness of this method for the design of an effective emergency rescue plan. It has been shown that even within a complex system, there are invariant features (e.g. a particular sequence of actions) that can serve as a basis for the design (Schwaninger 2000). It could be concluded that agent-based modelling and computer simulation may provide a useful method for the design of artefacts addressed to complex sociotechnical systems, when human agents can be considered acting reactively and not proactively. The main strength of the method is that it provides an appropriate dynamic representation of the designer s problem space, enabling him (i) to concurrently apprehend the critical dependencies and invariants of the system under consideration (Axelrod 1997a, b), (ii) to conceive the features that should characterise the designed artefact, and (iii) to assess their efficiency. References Fig. 3 Results of the simulations of Experiment 2 Axelrod R (1997a) Advancing the art of simulation in the social sciences. Complexity 3(2):16 22 Axelrod R (1997b) The complexity of cooperation. In: Agent-based models of competition and collaboration. Princeton University Press, Princeton Bernon C, Camps V, Gleizes MP, Glize P (2001) La Conception de Systèmes Multi-Agents adaptifs: contraintes et Spécificite s.

10 126 Bonabeau E, Theraulaz G, Deneubourg JL, Aron S, Camazine S (1997) Self-organisation in social insects. Trends in Ecology and Evolution 12: Casti JL (1997) Would-be worlds. How simulation is changing the frontiers of science. Wiley, New York Casti JL (1999) The computer as a laboratory: toward a theory of complex, adaptive systems. Complexity 4(5):12 14 Checkland P (1984) Systems thinking, systems practice. Wiley, Chichester Cheng LH, Ueng TH, Liu CW (2001) Simulation of ventilation and fire in the under-ground facilities. Fire Safety J 36: Choi TY, Dooley KJ, Rungtusanatham M (2001) Supply networks and complex adaptive systems: control versus emergence. J Oper Manag 19: Cilliers P (1998) Complexity & postmodernism. Taylor and Francis, London Diamantidis D, Zuccarelli F, Westhäuser A (2000) Safety of long railway tunnels. Reliability Eng Syst Safety 67: Drysdale D (1997) An introduction to fire dynamics. Wiley, Chichester Dugdale J, Pavard B, Subie J-L (2000) A pragmatic development of a computer simulation of an emergency call centre. In: De Michelis G. et al (eds) Designing cooperative systems, Proceedings of COOP IOS Press, Amsterdam Gann RG, Babrauskas V, Peackock RD, Hall JR Jr (1994) Fire conditions for fire toxicity measurement. Fire Mater 18: Helbing D, Farkas I, Vicsek T (2000) Simulating dynamical features of escape panic. Nature 407: Holland JH (1992) Adaptation in natural and artificial systems. In: An introductory analysis with applications to biology, control and artificial intelligence, 2nd edn. MIT Press, Cambridge Hollnagel E (2002) Time and time again. Theor Issues Ergon Sci 3(2): Kauffman S (1996) At home in the universe. The search for laws of complexity. Penguin Science, London Le Bon G (1971) Psychologie des Foules, nouvelle e dition, 2 e tirage. PUF, Paris Nathanael D, Marmaras N, Papantoniou B, Zarboutis N (2002) Socio-technical systems analysis: which approach should be followed? In: Bagnara S et al (eds) Cognition, culture and design, Proceedings of the ECCE-11, Catania, September 2002, pp Purser DA (1992) The evolution of toxic effluents in fires and the assessment of toxic hazard. Toxicol Lett 64/65: Purser DA (2000) Toxic product yields and hazard assessment for fully enclosed design fires. Polym Int 49(10):1,232 1,255 Railsback SF (2001) Concepts from complex adaptive systems as a framework for individual-based modelling. Ecol Model 139:47 62 Schwaninger M (2000) Managing complexity-the path toward intelligent organizations. Syst Pract Action Res 13(2): Simon HA (1996) The sciences of the artificial, 3rd edn. MIT Press, Cambridge Xue H, Ho JC, Cheng YM (2001) Comparison of different combustion models in enclosure fire simulation. Fire Safety J 36:37 54

11 本文献由 学霸图书馆 - 文献云下载 收集自网络, 仅供学习交流使用 学霸图书馆 ( 是一个 整合众多图书馆数据库资源, 提供一站式文献检索和下载服务 的 24 小时在线不限 IP 图书馆 图书馆致力于便利 促进学习与科研, 提供最强文献下载服务 图书馆导航 : 图书馆首页文献云下载图书馆入口外文数据库大全疑难文献辅助工具