In this study, the fire evolution process of subway stations is divided into three stages, based on the case studies of many typical fire accidents in subway stations, and a further evaluation is made on the judgment of experts. Based on the three stages, we propose 28 nodes that have dependency relationships for representing the subway station fires, from causes to consequences. The proposed Bayesian nodes are divided into three types, and their state classifications are described below.
3.1.1. Root Nodes
(1) Fire source material (state: Circuit; Luggage; Commodity)
Different to public places on the ground, the types of combustible materials in subway station are relatively specific, which could lead to a variety of consequences. Fires in subway stations are mainly caused by combustible substances carried by passengers or man-made attacks. From the previous statistics, the fire combustibles in subway stations can be divided into three major categories, which include circuit, luggage, and commodity.
(2) Fire alarm system (state: Normal; Malfunction)
This node represents the presence of fire alarm systems. According to the national criteria in China, the fire alarm system must be equipped in subway stations. The state “Normal” refers to that the fire alarm system has been equipped and is in a normal working condition, while the state “Malfunction” means that the fire alarm system might be a substandard product or aging, and might not function in case of fire.
(3) Sprinkler system (state: Normal; Malfunction)
This root node represents the presence of an automatic response sprinkler system in subway stations. The state “Normal” refers to that a standard sprinkler system has been installed, and it can work well in case of fire, while the state “Malfunction” means the sprinkler system is a substandard product, aging, or faces accidental blockages that could be useless in case of fire.
(4) Extinguisher (state: Normal; Malfunction)
This root node presents the performance of extinguishers in subway stations. The state “Normal” indicates that extinguishers are in a normal working state, while the state “Malfunction” means they might be substandard, out-of-date, used before, and may not work in case of fire.
(5) Fire time (state: Peak period; Other time)
Different subway stations generally have different passenger flow densities at distinct time periods. In this study, we just classify the time into two periods: “Peak period”, i.e., at the rush hours to or off work, and “Other time” when there are not many passengers.
(6) Fire source position (state: Inside train; On platform)
According to previous statistical data, the main fire locations are in the train and on the station platform, and quite a few of the fire source positions are in the subway tunnel [
42]. Herein, we do not pay more attention to the subway tunnel and other locations, and simply classify fire source positions according to two states: Inside train and On platform, and the probabilities are distributed to this node according to the statistical results.
(7) Subway station type (state: Single floor; Double floor; Complex)
The vertical structure of subway stations is divided into three levels, according to the actual practice in Chinese subway stations. The subway station of single layer is only one floor underground. The underground part of the double floor subway station consists of two floors. For example, the station hall is located on the first underground floor, while the platform, the trains, and the tunnel are located on the second underground floor. Another kind of vertical subway structure is called complex, whose vertical structure is more complex, which is more common in interchange stations [
43].
(8) Platform type (state: Island type platform; Side type platform; Combined type platform)
There are generally three types of subway station platforms: island type, where the platform is designed between two tracks; side type, where the platforms are located on two sides of the track; combined type, that integrated the island type and the side type. Each kind of platform type has its advantages. The subway station area and changes of the passenger flow are different for these three platform types, which have a great influence on economic loss and casualties in case of subway station fire [
44].
(9) Fire brigade location (state: Within 3 min distance; Over 3 min distance)
This root node is set up to examine the significance of firefighters for firefighting and rescue. The distance from the subway station to the fire brigade has a significant impact on the rescue effectiveness organized by firefighters. Previous studies show that the best escape time is 300 s during fire development in an underground subway station. According to experts’ experience, the proper duration for firefighters’ arrival is three minutes. Therefore, the states of this node are set as “Within three minutes” and “Over three minutes”. The former represents that the firefighters could arrive at the subway station within three minutes, and the latter one represents that the firefighters take more than three minutes to reach the subway station.
(10) Smoke extraction system (state: Normal; Malfunction)
Smoke is a critical factor affecting casualties in case of fire. A subway station can be regarded as a closed or semi-closed underground space, so the ventilation system is crucial for the diffusion of fire smoke. The states of this node are set as “Normal” and “Malfunction”, to represent if the smoke extraction system could work well.
3.1.2. Intermediate Nodes
(11) Fire detection by human (state: Yes; No)
Whether the fire can be detected by field staff or passengers in time has great impact on the probability of successful evacuation and dialing the fire alarm phone, which will significantly influence the casualties. The states of this node are set to “Yes” and “No”. “Yes” means that field person could detect the fire in time, and vice versa.
(12) Safe evacuation (state: Yes; No)
As we know, evacuation will directly affect the number of casualties. Since underground subway stations are relatively closed spaces, safe evacuation is more difficult than other places. In the early stage of fire development and expansion, hot and hazardous smoke flow is the primary influencing factor to escape routes and trappers’ safe evacuation.
(13) Fire call (state: Yes; No)
When a fire occurs, if field staff or passengers do not discover it at the initial stage of the fire, the fire call will be delayed, which may cause more serious fire loss.
(14) Firefighters’ arrival time (state: Before flashover; After flashover)
Firefighters play a major role in personnel rescue, fire control, and fire extinguishment. Therefore, it is important to discuss firefighters’ arrival time, and the states of this node are divided into “Before flashover” and “After flashover”. Flashover is a critical stage in fire evolution.
(15) Fire alarm system response (state: In time; Not in time)
A fire alarm system is a crucial factor in the detection of fire, and the response time is associated with the time when people detect the fire. The states of this node are classified into “In time” and “Not in time”. To specifically reflect the response of a fire alarm system, “In time” represents that the fire alarm system responds within 30 s after it detects a fire (China, GB50166-2007).
(16) Initial stage of fire (Stage 1) (state: Smolder; Fire)
In the initial stage of fire, also called Stage 1 of fire development, there are two different combustion modes: smolder and fire. If smolder occurs in the initial stage, it is very hard for the alarm system to detect the fire, and the sprinkler system to conduct a prompt response.
(17) Sprinkler system response (Stage 1) (state: In time; Not in time)
Sprinkler systems can effectively control the spread of fire. If the sprinkler system responds in time, especial in Stage 1 of fire development, there may be a higher probability that the fire can be put out or well controlled. The states of this node can be set to “In time” and “Not in time”, whereby “In time” means that the sprinkler system responds within 30 s in the initial stage of fire development.
(18) Put out early (Stage 1) (state: Yes; No)
This node is set to reflect extinguishment effectiveness during the fire development. At the initial stage of fire, i.e., Stage 1 of fire development, it is possible to put out the fire by the field people.
(19) Growth stage of fire (Stage 2) (state: None; Slow development; Flashover)
If the fire is not put out in Stage 1, it will develop further to flashover or slow development, which is called Stage 2 of fire development in this study. If the fire is put out in stage 1, it will not develop into Stage 2.
(20) Sprinkler system response (Stage 2) (state: In time; Not in time)
According to the response effectiveness of automatic fire extinguishing system in Stage 2, this node was divided into two states: “In time” and “Not in time”. Different from the sprinkler system response in Stage 1, it is more difficult to extinguish the fire in Stage 2, but it is likely to mitigate the spread of fire.
(21) Put out late (Stage 2) (state: Yes; No)
Depending on fire development situation in Stage 2, and the working effectiveness of automatic fire extinguishing systems and firefighters, the fire in Stage 2 could be extinguished, or not be extinguished.
(22) Fully developed to quenched (Stage 3) (state: Yes; No)
This node is to represent the final stage of a fire, i.e., Stage 3: the fully-developed fire to quenched (going out) at last. If the fire develops to this stage, it indicates that the fire is not effectively controlled.
(23) Severity of fire development (state: Slight; Moderate; Serious)
The burning proportions of main fire body parts are generally divided into local burning (<30%), major burning (30–70%) and total burning (≥70%). According to this, the node is classified into three states to represent the severity of fire: Slight, Moderate, and Serious.
(24) Smoke extraction system response (state: In time; Not in time)
The response of the smoke extraction system determines whether the smoke generated by fires can be discharged in time, which is of great significance for the safe evacuation of trapped people in the subway station.
(25) Concentration of hazardous gas (state: Less than critical value; More than critical value)
The concentration of hazardous gases, such as CO, HCl, H2S, and others, will increase faster in fire, and the toxic and harmful gases in the fire are often the main cause of casualties. Hence, the states of this node were divided into “Less than critical value” and “More than critical value”. The accumulated critical values for 5 min causing death of CO, HCl, and H2S, are 5000 ppm, 3000 ppm, and 800 ppm, respectively.
(26) Temperature (state: Less than 100 °C; More than 100 °C)
According to the statistics of related medical and physiological experiments, high temperature smoke in fires will cause a certain degree of harm to the human body. To judge whether the smoke temperature is dangerous or not, the smoke temperature nearby human eyes is usually used as the critical temperature. Herein, this node is divided into “Less than 100 °C” and “More than 100 °C”.