مدل توسعه آتش خانه و فرار از سکونت با استفاده از شبکه های بیزی

The concept of probabilistic modelling under uncertainty within the context of fire and rescue through the application of the Bayesian network (BN) technique is presented in this paper. BNs are capable of dealing with uncertainty in data, a common issue within fire incidents, and can be adapted to represent various fire scenarios. A BN model has been built to study fire development within generic dwellings up to an advanced fire situation. The model is presented in two parts: part I deals with “initial fire development” and part II “occupant response and further fire development”. Likelihoods are assessed for states of human reaction, fire growth, and occupant survival. Case studies demonstrate how the model functions and provide evidence that it could be used for safety assessment, planning and accident investigation. Discussion is undertaken on how the model could be further developed to investigate specific areas of interest affecting dwelling fire outcomes.

In this paper, a Bayesian network (BN) model is presented to investigate the development of fires within dwellings and assess the probability of the associated consequences. A ‘dwelling’ can be defined as a house which is used solely for living in; it excludes any make-shift shelters, caravans, or buildings. Managing fire safety within society is immensely complex due to the sheer number of different environments or situations that exist. Not only are there diverse types of locations, for example factories, warehouses, high rise buildings, dwellings, etc., but there are also many different circumstances within each type of location. In dwellings for example, variations will arise in terms of fire safety arrangements, geographical location relative to fire stations, characteristics of occupants, activities of occupants, characteristics of dwelling, culture among others. Should a fire occur, each incident will be unique in terms of the type of fire, time of the day, state of occupants, fire cues, etc. What all these variations signify is that the development of the next fire event and the magnitude of its consequences are generally unpredictable. The upshot of these complications is that a high degree of uncertainty is often attached to the management of fire safety, in particular within housing communities. Fire statistics indicate that dwelling fires result by far in the greatest number of fatalities when compared to other types of locations across the UK [15]. Fatalities primarily result from being overcome by fire gases, or heat exposure and burns. The causes which actually lead to such tragedies are numerous. During nighttime fires for instance, many people do not awaken at all or in time to escape from the dwelling. This may be because of an absence of a smoke alarm, a non-operative smoke alarm, the occupant being intoxicated and not reacting to sounds, the physical condition of the occupant, or reasons relating to the characteristics of the dwelling. For cases when fire cues succeed in producing early warnings, occupant survival will likely be influenced by the decisions a person might take. Fire risk assessment is undeniably a complex task due to the multiple variables and potential outcomes that must be examined in a holistic manner. Reasoning and decision making regarding what fire prevention and mitigating measures to reinforce, are often undertaken with a degree of uncertainty. In order to diminish some of this uncertainty and improve confidence in fire safety decision making, this paper presents a two-part BN. The first part of the model examines the early stages of a fire focusing on human reaction to fire cues; the second part of the model addresses, among other things, the follow-up actions an occupant may take in responding to the situation, as well as the intervention of the fire and rescue service (FRS). The model is designed for application in the UK but may be tailored for other locations should relevant data be obtained. The purpose of the study is to model the sequence of events which may occur during a fire at a given location, and to examine the critical variables that intervene in human reaction, fire growth, and survival. Several case studies are presented in the paper to demonstrate how the model works and to investigate particular fire scenarios.

This paper presents a two-part BN to model the development of fire in a generic dwelling up to an advanced fire situation. The model is set-up to assess the likelihood of human reaction, fire growth, and occupant survival, among other events of interest. Part I of network has been used to investigate, through the first case study, the significance of fire types, smoke alarm, and the time of day. The case study concludes that humans are less likely to react to fire cues during a smouldering fire than a flaming one, but that this difference is far greater when there is no smoke alarm and it is night; for nighttime situations when there is no smoke alarm present, the probability of human reaction during a smouldering fire is nearly half compared to a flaming fire. In a second case study, the model is used as a fire investigation tool to determine the likely causes of no human reaction. A third and forth case study are conducted with part II of the model; in the third case the probability of becoming trapped within a dwelling fire is determined based upon the time taken to make a 999 call and the actions undertaken by the occupant. The case reveals the importance of making an immediate 999 call and provides justification for evacuating the dwelling above any other action. The forth case study examines the impact of location and LSOA risk level upon the probability of remaining trapped; it was noted that neither have a major impact implying that FRS response is set up in a way that provides relatively equal cover for everyone. Diminishing probabilities of remaining trapped were noted for increasing levels of risk but it was recognized that this was offset by increasing frequency of incidents associated with higher risk. A final case study examines the effect of certain nodes from part I of the model, upon a focus node from part II. This served a dual purpose, firstly to investigate how smoke alarm, sprinklers, and time of day affect an occupant’s chances of remaining trapped in a fire, and secondly to demonstrate how part I and part II function together. The BN model has shown to be useful for undertaking dwelling fire analysis from various perspectives. Results such as the ones produced may be useful for planning and management of fire safety. Findings could be extrapolated for assessment of residential communities with similar characteristics such as type of housing, level of risk, new homes with sprinkler systems, areas targeted for smoke alarm campaigns, and so forth. It has also been shown that the model is adaptable and can be expanded for example, to investigate cases of variable occupancy, to assess the impact of social deprivation upon human reaction, and to serve as a decision making tool.