Real Time Building Evacuation Modeling with an Improved Cellular Automata Method and Corresponding IoT System Implementation

Abstract

Facility emergence evacuation is often a complicated process under extreme conditions. Most of the buildings today use pre-installed signages to guide the emergence evacuation. However, these guidances are sometimes insufficient or misleading, particularly for evacuating from high-rise buildings or complex buildings, such as schools, hospitals, and stadiums. Following a planned route may lead the crowd to move towards dangers, such as smoke and fire. The future emergency guidance system should be more intelligent and be able to guide people to evacuate with a higher survival possibility. This study proposes a real-time building evacuation model with an improved cellular automata (CA) method. This algorithm combines cellular automata with the potential energy field (PEF) model in fluid dynamic theory (FDT) to choose safe paths for the crowd and reduce the possibility of stampedes. Custom-designed wireless sensors, artificial intelligence (A.I.) enhanced surveillance cameras, intelligent emergency signage systems, and edge computing servers are used to sample fire and crowd data, operate the intelligent evacuation algorithm, and guide the crowd with the signage system in real-time conditions. In addition, we performed the algorithm simulation on a two-dimensional plane generated based on the building structure of the Beijing Capital Airport Hospital. The evacuation drill simulations show that the average escape time is significantly shortened with optimal real-time guidance. In one case, a 72% reduction in evacuation time is achieved compared with evacuation using pre-installed signages. The results also demonstrated that the proposed model and system’s evacuation time reduction performance is particularly good in crowded buildings, such as schools or stadiums.

1. Introduction

Building fires cause tens of thousands of casualties and billions of dollars in economic damage worldwide [1,2]. In a United States Fire Administration (USFA) report, it was shown that casualties, as well as property damage due to fire hazards, have continued to increase in the past decade, with 13,195,000 fires resulting in 3400 deaths and 14,670 injuries and 23 billion U.S. dollars in direct property damage in 2017 alone [3]. For fire safety issues in buildings, the study of evacuation path planning has always been an important field of research. Correct and effective building emergence evacuation plans can minimize casualties and have been the focus of research. Complex buildings, such as high-rises, hospitals, schools, theaters, and stadiums, mandate effective evacuation path design and simulation drills.

The fire evacuation path planning started as a simple shortest path graph theory problem. However, based on many real-world cases, more factors are taken into consideration in the evacuation problem, including crowd size, obstacles on the evacuation route, smoke and flame coverage scope of the building, and the psychological state of nervousness and panic. For more complex evacuation scenarios, such as hospitals, more additional factors must be considered, for example, the patient’s mobility, ongoing surgeries, and critical medical resources, including plasma and donated organs [4,5,6,7]. More recently, the COVID-19 pandemic has further complicated the evacuations of medical facilities, with some of the constraints being conflicting, or even impossible, to achieve simultaneously.

Many methods of building emergence evacuation modeling exist in the literature. Ni Li and Yunjun Xu categorized these models into two types: macroscopic models and microscopic models [8]. The macroscopic models mainly use numerical analysis models to solve evacuation problems holistically. The macroscopic models include pseudo-polynomial methods, classical Hughes models, and finite element analysis models [9,10]. These models are mainly applied to large-scale evacuation under the continuity of time when a flood, fire, or seismic event occurs. It mainly considers factors such as the macro-mechanics of the crowd and the directionality of the crowd, which can effectively avoid large-scale congestion and human injuries. However, this macroscopic model cannot respond well to the individuals’ complicated crowd dynamics and various micro factors.

The microscopic simulation models approach this problem based on individual behavior modeling and analysis from a bottom-up perspective. The concurrent mathematical models such as cellular automata (CA) methods [11], lattice gas methods, and social force methods are frequently utilized in these approaches, in addition to classic models such as the Dijkstra algorithm [12] and the Floyd algorithm [13]. These models consider key factors, such as building plans, crowd mobility, reaction time, the psychological state during the evacuation, age, and the individual’s health condition, to calculate each individual’s automaton movement from the initial location to the exit. Then, through repeated simulations with different initial conditions, the optimal evacuation path models and corresponding signages are formulated. Microscopic models are now commonly used since they can better characterize interactions among evacuees [14]. The Cellular Automata models are widely used since this model is well studied and suitable for a large-scale evacuation because today’s computer processing capability can better facilitate its homogeneous and concurrent mathematical calculations [15,16,17].

The ultimate goal is to achieve the best evacuation route and the shortest evacuation time [18,19]. During the evacuation, it is necessary to comprehensively consider different factors of the personnel and buildings in the event of a fire and other emergencies, such as the specific location of the smoke and fire [20,21], the building structure [22], and the psychology of the personnel. Many uncertainties also exist in the evacuation process, such as the open/closed exit and exit width. Depending on these factors, a more reasonable egress route is planned [23,24,25,26,27].

A disadvantage in most of the existing microscopic model methods is that although the microscopic method can dynamically analyze the crowd evacuation process, there is a lack of technical methods to change the evacuation signage dynamically. Even if the fire scene of the building is very different from the preset model, it is impossible to adjust the signages after the fire emerges. Some research finds the CA model is sufficient and relatively accurate for most small to medium-sized buildings, even when the evacuation scenario is different from the predicted conditions and the evacuation signage cannot change its preset route [9,16,17]. However, it may not be sufficient for more complex buildings to achieve optimal evacuation and may lead to wrong evacuation instructions and cause more severe casualties. If the CA model is able to interact with the sensors and signages of the building, its dynamic analysis characteristics would make it possible to formulate evacuation paths better and adjust them in real-time according to fire and personnel conditions.

With the Internet of Things (IoT) technology development, more sensor and actuator systems are installed and used in smart buildings, making information perception and device remote control more feasible. This interactive IoT technology situation brings new perspectives to the study of CA models and other microscopic evacuation models. The researchers are no longer faced with the study of the evacuation route when the building is built but formulate the evacuation plan in real-time according to the risk assessment when the building encounters a specific hazard. This type of interactive research method started to emerge in recent years [28]. Wire-line [29] and wireless sensor networks [30] have been installed in buildings to detect building-related emergencies such as smoke and fire. When a fire is detected, an early warning signal is issued and sent to the control terminal to evacuate people in the building. In the evacuation process, Bluetooth and other wireless communication technologies in IoT devices are used in IoT terminals to transmit the sensor data [31,32,33,34,35]. However, data integrity can be compromised due to the possible loss of IoT devices, deterioration of the wireless signal, or power failure caused by fire. More recently, artificial intelligence (A.I.) algorithms have been introduced in building evacuation in conjunction with IoT and BIM [36] to improve the robustness of the IoT enhanced evacuation model.

In addition, surveillance camera equipment is covering a more significant proportion of the building area in large public buildings, such as railway stations, airports, stadiums, schools, and hospitals. Some latest surveillance camera equipment can perform intelligent scene analysis, which can output data such as the people count in zones, graphical flame/smoke analysis results, and obstacles blocking evacuation routes. There are two specific data analysis methods: (1) uploading the image data to the cloud/local server in streaming files and performing data analysis on the cloud/local server; and (2) performing “edge computing” data analysis on the camera device, feedback on the abnormal circumstances results, and performing video/image backup storage on the cloud or local server [37,38]. Furthermore, some scholars are studying how to evacuate effectively in case of sudden power failure of IoT devices when a fire occurs [39,40,41].

The building information model (BIM) acts as a “digital map” for CA evacuation model simulation and IoT system deployment layout. Many key parameters and simulations need to be implemented on the BIM model [42,43]. In addition, an improved BIM model [44] is proposed with personnel visualization information to make the model more consistent with the actual scene. Subsequently, several researchers have developed evacuation simulation software, such as the evacuation simulator [45], the fire system [46], and the emergency management system [47].

This paper proposes a new cellular automata model and its corresponding IoT system. This novel dynamic cellular automata (DCA) model adjusts the evacuation plan based on BIM, crowd, and fire alarm data wirelessly collected by IoT sensors and smart surveillance cameras. An IoT system with LoRa wireless communication module transmits the evacuation path decision (signage direction) data obtained by the CA model to customized emergency signages/lights, which are also equipped with the LoRa communication module. The improvement of the cellular automata model in this paper is mainly the enhancement of the moving direction of the individual cell so that the individuals in this CA model can adapt to the complex environment in the building. Compared with the traditional CA cell model, each cell can now move at any angle instead of only moving in four directions, up, down, left, and right, which is more suitable for actual evacuation scenarios. This paper uses an efficient combination of the improved cellular automata and potential energy field model, which can significantly reduce simulation time. This new fast simulation method can utilize devices with relatively low computing power to achieve fast simulation within tens of seconds, which is of great significance for real-time systems. Considering the possible loss of computing equipment and partial communication interruption in a fire emergence, this paper proposes a low-cost edge computing gateway that can be deployed in many different areas of the building. This multi-disaster recovery edge computing system is expected to improve the system’s robustness under fire conditions.

The remainder of this paper is organized as follows. Overall structure and hardware design of evacuation system is presented in Section 2. Section 3 shows the theory and modeling of the developed evacuation model. Section 4 presents the verification of the designed intelligent evacuation model. Finally, the conclusions are drawn in Section 5.

2. System Architecture and IoT Design of Evacuation System

The system architecture and operation process of this work is shown in Figure 1. The system architecture and operation process of this work is shown in Figure 1. The detailed operation of the system is divided into the following five steps: (1) fire broke out inside the building. (2) Fire was detected by the building’s smoke and fire sensors, and an evacuation began. (3) IoT system and DCA model analysis start initialization. The IoT system starts monitoring smoke and fire alarm signals, analyzes the people count from the surveillance video system, and uploads the data to the edge computing gateway through the LoRa IoT wireless communication module. (4) Using a BIM-based visualization system, the DCA model is run on the edge computing gateway to calculate evacuation paths based on the current fire situation and personnel conditions. On this basis, the system gives the direction value on each signage. (5) The system transmits signage direction data to smart signage devices through IoT LoRa communication to guide the crowd to evacuate according to the optimal path.

This case study’s BIM data and evacuation scenarios are based on the West Wing of Beijing Capital International Airport Emergency Center (Formally known as Capital Airport Hospital) in Chaoyang District, Beijing, China, as shown in Figure 2.

The exterior view of the emergency center is shown in Figure 2a, while Figure 2b shows the first-floor plan of the emergency center building. In this case study, 29 rooms and 3 main corridors are included, which are mainly the emergency room, outpatient doctor’s office, treatment room, and reception room. The airport emergency center is a small-scale hospital that mainly treats emergency patients with sudden onset inside the airport or during the flight and does not have an inpatient department or ward.

The IoT hardware part of this system mainly includes wireless sensors, A.I. enhanced cameras, edge computing gateway and wireless emergency signage.

2.1. Sensor and Edge Computing

The sensors in the evacuation system are shown in Figure 3. It consists of temperature sensor, smoke sensor, humidity sensor and image sensor. The image sensing is based on a CCTV camera and is used for flame image identification, while the smoke and temperature sensors are used to collect the smoke and temperature parameters in order to assist the fire monitoring. When a fire occurs, the sensors detect the abnormal data, processed by STM32 MCU onboard the IoT circuits, and use wireless communication module to transmit the data. The edge computing platform receives the data and the system evacuation algorithm is finally run on the edge computing platform.

2.2. Emergency Signage and Lighting

After using the improved cellular automata algorithm to plan the evacuation route on the edge computing platform, the optimal path for evacuation is sent to the back-end emergency signage and lighting system. The emergency signage light guides people based on the route planned by the algorithm. The hardware composition of the system emergency light is proposed based on the original circuit of the traditional emergency light [48]. Most traditional emergency lights cannot change the evacuation direction on the signage. This traditional signage often features good backup battery design, which can work in power blackouts for several hours. A battery analysis showed that the over 30–50 Ah capacity battery can supply the additional, ultra-low-power 10 mA current consumption IoT circuit with no significant battery lifetime reduction. In this paper, the emergency signage part mainly uses the SI10XX series chip as the core communication and control circuit. The SI10XX is an ultra-low-power single-chip wireless micro-controller (MCU) that is optimized for power supply and meets the power consumption and RF signal transmission requirements of indoor monitoring and security systems. It can collect different voltage parameters in the existing fire emergency light switching power supply. The internal modular composition of the emergency light fixture is shown in Figure 4.

The ultimate goal of building the evacuation system is the efficiency in the shortest time. The required evacuation time (t_all) is the time from the start of the fire to the end of the evacuation of the crowd to a safe area. It can be calculated based on the fire detection, alarm time (t_alarm), pre-action time (t_pre), and evacuation movement time (t_move):

t_all = t_alarm + t_pre + t_move

(1)

Dynamic evacuation systems are mostly used in corridors when crowds exit from the rooms. A dynamic signage system to evacuate the crowd greatly reduces the necessary time for evacuation. This can ensure that the required evacuation time (t_all) of people in the fire site is reduced, and the evacuation efficiency is greatly improved. Therefore, the dynamic evacuation indication system is very suitable for the safe evacuation of people in large-scale building fires [49,50].

In this paper, a wireless LED smart signage system based on LoRa wireless communication is designed. When a fire occurs, the image recognition system uses the LoRa wireless transmission technology to recognize the fire [51]. The location information is sent to the Jetson Nano edge computing device. The Jetson Nano edge computing device is a real-time data processor for the data side. It is a product released by NVIDIA in 2019, which allows you to port your own algorithms to this device and complete algorithm applications and research. The shortest escape route is determined by the improved cellular automata algorithm. The route information is sent to the LED lights placed on all the corridors of the emergency center through the LoRa transmission technology. The LED indicator completes the indication of the valid path according to the transmitted path information. The circuit diagram of the LoRa-based smart indicator developed by our team is shown in Figure 5, where the micro control processor (MCU) is an ARM Cortex-M3-based chip, the specific model of which is STM32F103RCT6, with an operating frequency of 72MHz, while the SX1278 chip is a radio frequency chip (RFIC) based on the LoRa communication technology. The power management chip is RT7272 synchronous step-down DC/DC converter, the voltage regulator circuit uses AMS1117 chip, the USB to serial circuit uses CP2102 chip, and the LEDs are selected with conventional indicators. The temperature sensor is used to detect the heat generation of the hardware. In Figure 5, the top physical diagram and the bottom schematic diagram with capital letters for one-to-one correspondence, from the schematic diagram produced a PCB layout and according to the PCB layout design, show the hardware circuit production.

3. Building Evacuation Model Theory and Modeling

This paper considers the west side building at Beijing Capital Airport Emergency Center as an example. The building is first abstracted and mapped on a two-dimensional plane. The cells of cellular automata are then used to simulate the evacuees. We first propose an improved meta-cellular automaton model based on the existing meta-cellular automata. The improved meta-automaton model is used to construct the evacuation algorithm, in which we consider factor 1: first calculate the distance of a particular evacuee from different exits, and use probabilistic priority to calculate the optimal possible evacuation exit based on the different distances plus the density of people on the scene. Factor 2: we added the potential field theory to determine the optimal exit route. Factor 3: we consider factors such as the speed of evacuees, the density of people, and the width of the evacuation exit. The three factors are used to effectively design the optimal escape algorithm rules and design the algorithm. Through simulation, we analyzed the results and compared the analysis with the shortest path algorithm Dijkstra’s algorithm to conclude the superiority of our improved meta-cellular automata algorithm. The overall method diagram of building evacuation model is shown in Figure 6.

3.1. Cellular Automata Model

3.1.1. Cellular Automata Theory

The cellular automata is a dynamic system which evolves in a discrete time dimension [52,53,54,55]. In cellular automata, a regular grid of a certain form is divided into many cells to form a cellular space. The cell is the unit in these regular grids, and its value range is a finite set of discrete states. The states of all the cells are updated according to the same evolution rules and certain local rules. The cellular automata is like a “field” acting in a shorter distance in traditional physics, which is a model of a discrete field [56]. The structure diagram of the cellular automata is shown in Figure 7. It should be noted that this is a plan view of the structure of the cellular automata.

Cellular automata is composed of five parts: cell, cell space, state set, field, and a state transition function [57]. The model has a simple framework, which can simulate very complex system behaviors and has strong vitality. Its general formula is given by:

$$\mathrm{CA}=\left({\mathrm{L}}_{\mathrm{d}},\mathrm{S},\mathrm{N},f\right)$$

(2)

where CA represents a cellular automaton, L_d denotes a d-dimensional cell space finite sequence subset, S represents the cell state set, N denotes the cell’s neighbor set, and f represents the evolution rule of the cell state. The relationship between the different parameters of the cellular automata is shown in Figure 7. Together, these factors form the most basic cellular automata model.

3.1.2. Improved Cellular Automata Model Selection

In this paper, the two-dimensional cellular automata model is used. The two-dimensional cellular automata models include Von Neumann type, Moore type, extended Moore type, and Margolus type. Most of the cellular automata evacuation models are shown in Figure 7. In this paper, the extended Moore type is considered. Based on the extended Moore type, the exit priority probability of sports is used to calculate the next possible exit. The potential energy method is then used to determine the optimal route from the exit, as shown in Figure 8a.

The operating rules of improved cellular automata are parallel. They are summarized as follows:

• All the cell states happen at the same time;
• The state of the i-th cell at time t + 1 is determined by the state of the i-th cell at time t and the adjacent 2r cells whose distance does not exceed r.

3.2. Modeling of the Evacuation System Based on Improved Cellular Automata

Based on the improved cellular automata evacuation model, the closest exit distance model, priority selection mobile model, potential energy field model, and building factor model are all included.

3.2.1. The Closest Exit Distance Model

The layout of the west side building of the Beijing Capital Airport Emergency Center is set in a two-dimensional space. The cell space uses the top left corner as the 0 point coordinate to form a plane coordinate system. The position of each cell is represented by (i, j), while the coordinate value is 1 or 0. Let any cell A (i, j) have center coordinates (x, y). According to the building standard, the design cell step length is 1.2, and the coordinate formulas are given by:

$$\mathrm{x}\left(\mathrm{i},\mathrm{j}\right)=1.2\mathrm{i}-0.6$$

(3)

$$\mathrm{x}\left(\mathrm{i},\mathrm{j}\right)=1.2\mathrm{i}-0.6$$

(4)

The process of evacuation of people is parallel. Therefore, the movement of cells uses the parallel rule, that is, the state of all the cells changes at the same time, and each cell can only move one step or not move at a time. The cell at (i, j) can then be the target network at the next moment, including eight neighborhoods, up, down, left, right, top left, top right, bottom left, and bottom right. Assuming there are two exits, the exit center coordinates are 01 (x₁, y₁) and 02 (x₂, y₂). The distance formula between A (i, j) and the nearest exit is then expressed as:

$$\mathrm{D}\left(\mathrm{i},\mathrm{j}\right)=\min \left(\begin{array}{c}\sqrt{{\left(1.2\mathrm{i}-{\mathrm{x}}_1-0.6\right)}^2+{\left(1.2\mathrm{j}-{\mathrm{y}}_1-0.6\right)}^2}\\ \sqrt{{\left(1.2\mathrm{i}-{\mathrm{x}}_2-0.6\right)}^2+{\left(1.2\mathrm{j}-{\mathrm{y}}_2-0.6\right)}^2}\end{array}\right)$$

(5)

Note that D(i, j) is also used to find the shortest distance from the two exits. The distance formula of the outer cell is similar to that in Equation (5). Through the two preset exits, the general formula for the nearest distance of each cell to the exit is given by:

$$\mathrm{D}\left(\mathrm{i}\pm n,\mathrm{j}\pm n\right)=\min \left(\begin{array}{c}\sqrt{{\left(1.2\mathrm{i}-{\mathrm{x}}_1-0.6\pm n\times 1.2\right)}^2+{\left(1.2\mathrm{j}-{\mathrm{y}}_1-0.6\pm n\times 1.2\right)}^2}\\ \sqrt{{\left(1.2\mathrm{i}-{\mathrm{x}}_2-0.6\pm n\times 1.2\right)}^2+{\left(1.2\mathrm{j}-{\mathrm{y}}_2-0.6\pm n\times 1.2\right)}^2}\end{array}\right)$$

(6)

3.2.2. Development of the Improved Cellular Automata Cell Priority Selection Mobile Model

After calculating the distance between the cell and the preset exit, the next step is to consider the principle of preferential movement of the cell. By calculating the probability of each cell moving to each preset exit, this probability is set as the exit attraction probability. In addition, the midpoint probability of the cellular automaton is set to p(i, j). The extended Moore-type movement probability is shown in Figure 8b. The export attractiveness probability is expressed as:

$$\mathrm{p}\left(\mathrm{i},\mathrm{j}\right)=\frac{\max \mathrm{D}\left(\mathrm{i},\mathrm{j}\right)-\mathrm{D}\left(\mathrm{i},\mathrm{j}\right)}{\max \mathrm{D}\left(\mathrm{i},\mathrm{j}\right)-\min \mathrm{D}\left(\mathrm{i},\mathrm{j}\right)}$$

(7)

where D(i, j) is the distance from the grid (i, j) to the evacuation exit, max D(i, j) is the maximum grid distance from the exit, and min D(i, j) is the minimum distance from the exit grid distance value (the closer to the evacuation exit, the greater the probability of location attraction, and vice versa).

3.2.3. Development of the Improved Cellular Automata Potential Energy Field Model

In the case where different people have the same attractive probability, conflict detection is required when the same cell will be selected. At the same time, when there are fire burning points or obstacles during the escape process, after the introduction of these obstacles to escape, it should be considered that some of the shortest escape routes may cause fires. Smoke sensors or image sensors can be used for monitoring, in order to re-plan the escape route. Therefore, the theory of the potential energy field is introduced. When the sensor detects the fire, the improved cellular automata and the potential energy field algorithm are used to achieve effective escape. All the obstacles in the potential energy field that hinder the escape are represented by a repulsive force, introducing attraction to redirect people to choose the best route.

The gradient potential field method is commonly used to establish a potential energy field. It generates a virtual force for pedestrians caused by the negative gradient of the potential field. The target point generates the attractive force to the pedestrian, and the obstacle generates the repulsion force to the pedestrian. The combined force of the attractive force and the repulsion force is used as the pedestrian moving forward. The pedestrian moves toward the target without hitting the wall under the force “push”. The force of a single cell in the potential energy field and the direction of the escape route, are shown in Figure 9. The position of the cell in the figure is set to 4, while the other positions (1, 2, and 3) are three exits. When a fire occurs, Equation (6) is used to calculate the distance between the cell and the three exits, and the nearest exit is chosen. When a fire occurs, the nearest exit is directly selected. At this time, the cell moves to the position No. 5. At this time, the route should be re-planned. The potential energy method is then used to analyze the force, but finally waits in escape exit 3, and it is then escaped.

The theory of artificial potential energy field determines the pedestrians in any environmental space, as long as there is a target point, and regardless of whether there are static obstacles or other moving obstacles in the environment, and therefore an artificial potential energy field diagram can be generated and calculated. In the actual realization, the target point acts as an attraction pole to generate attractive force, and the surface of the obstacle generates repulsive force. In any state, the location of the pedestrian, received potential field, attractive potential field generated by the target point and repulsive force potential field generated by the obstacle are denoted by q, U(q), U_A(q), and U_R(q), respectively. The potential energy field of the pedestrian q in the environmental space can then be expressed as:

$$\mathrm{U}\left(\mathrm{q}\right)={\mathrm{U}}_{\mathrm{A}}\left(\mathrm{q}\right)+{\mathrm{U}}_{\mathrm{R}}\left(\mathrm{q}\right)$$

(8)

It is assumed that for every pedestrian in the space, the total potential energy U(q) is differentiable. The virtual force received by the pedestrian is the resultant force of the attraction of the target point and the repulsive force of the obstacle. The potential field force is defined as the gradient function of the potential field function, as shown in Equations (9)–(11).

$$\overline{{\mathrm{F}}_{\mathrm{A}}\left(\mathrm{q}\right)}=-\mathrm{grad}\left[{\mathrm{U}}_{\mathrm{A}}\left(\mathrm{q}\right)\right]$$

(9)

$$\overline{{\mathrm{F}}_{\mathrm{R}}\left(\mathrm{q}\right)}=-\mathrm{grad}\left[{\mathrm{U}}_{\mathrm{R}}\left(\mathrm{q}\right)\right]$$

(10)

$$\overline{\mathrm{F}\left(\mathrm{q}\right)}=-\mathrm{grad}\left[{\mathrm{U}}_{\mathrm{A}}\left(\mathrm{q}\right)\left]-\mathrm{grad}\right[{\mathrm{U}}_{\mathrm{R}}\left(\mathrm{q}\right)\right]$$

(11)

where $\mathrm{grad}\left[\mathrm{U}\left(\mathrm{q}\right)\right]$ represents the gradient of U at q, which is a vector pointing in the direction where the potential field change rate of the pedestrian q is the largest. There are many ways to define the potential field U(q). For the attractive potential energy field U_A(q) and the repulsive potential energy field U_R(q), the most commonly used definition is the electrostatic field potential field model, expressed in Equations (12) and (13).

$${\mathrm{U}}_{\mathrm{A}}\left(\mathrm{q}\right)=\frac{1}{2}{\mathsf{\delta }\mathrm{p}}_{\mathrm{g}}^2\left(\mathrm{q}\right)$$

(12)

$${\mathrm{U}}_{\mathrm{R}}\left(\mathrm{q}\right)=\{\begin{array}{c}\frac{1}{2}\mathsf{\eta }\left(\frac{1}{\mathsf{\rho }\left(\mathrm{q}\right)}-\frac{1}{{\mathsf{\rho }}_0}\right),\rho \left(\mathrm{q}\right)\le {\mathsf{\rho }}_0\\ 0 , \rho \left(\mathrm{q}\right)>{\mathsf{\rho }}_0\end{array}$$

(13)

where ${\mathsf{\rho }}_{\mathrm{g}}^2\left(\mathrm{q}\right)=\Vert \mathrm{q}-\mathrm{q}\left(\mathrm{q}\right)\Vert$ represents the Euclidean distance from q to q(q), ρ₀ is a normal number which indicates the maximum distance that obstacles have an impact on pedestrian movement, ρ(q) represents the minimum distance from the obstacle to the pedestrian q, and δ and η are constant coefficients.

Consequently, the potential field area is limited to the obstacle surface, and when the pedestrian q is infinitely close to the obstacle area, U_R(q) approaches infinity. By combining the previous formulas, the attractive and repulsive forces of pedestrians can be obtained as shown in Equations (14) and (15).

$$\overline{{\mathrm{F}}_{\mathrm{A}}\left(\mathrm{q}\right)}=-\mathrm{grad}\left[{\mathrm{U}}_{\mathrm{A}}\left(\mathrm{q}\right)\right]=\mathsf{\delta }\left(\mathrm{q}-\mathrm{q}\left(\mathrm{q}\right)\right)$$

(14)

$$\overline{{\mathrm{F}}_{\mathrm{R}}\left(\mathrm{q}\right)}=-\mathrm{grad}\left[{\mathrm{U}}_{\mathrm{R}}\left(\mathrm{q}\right)\right]=\{\begin{array}{c}\frac{\mathrm{n}}{{\mathsf{\rho }}^2}\left(\frac{1}{\mathsf{\rho }\left(\mathrm{q}\right)}-\frac{1}{{\mathsf{\rho }}_0}\right)\nabla \rho \left(\mathrm{q}\right),\rho \left(\mathrm{q}\right)\le {\mathsf{\rho }}_0\\ 0 ,\rho \left(\mathrm{q}\right)>{\mathsf{\rho }}_0\end{array}$$

(15)

The obtained resultant force is given by:

$$\overline{\mathrm{F}\left(\mathrm{q}\right)}=\overline{{\mathrm{F}}_{\mathrm{A}}\left(\mathrm{q}\right)}+\overline{{\mathrm{F}}_{\mathrm{R}}\left(\mathrm{q}\right)}$$

(16)

This combined force determines the state of movement of pedestrians. In potential energy, pedestrians move into the moving space and are always attracted by the target exit. The target exit determines the direction of the pedestrian’s entire movement process. When a pedestrian moves to an obstacle or a fire point, it is affected by the repulsive force, and the pedestrian looks for a quick escape route under the reasonable action of the attractive and repulsive force.

3.2.4. Construction of Building Factor Model for Improved Cellular Automata

People will consider several factors in the process of escaping in the building, including their own factors and the factors under the building’s own building standards. In this paper, a comprehensive optimization algorithm is built from the factors that are currently considered in the evacuation process. The factor correlation is expressed as:

$$\mathrm{T}=f\left(\mathrm{V},\mathrm{TOT},\mathrm{EW},\mathrm{PF},\mathrm{PA},\mathrm{NE}\right)$$

(17)

where V represents the speed of pedestrian evacuation, TOT denotes the density of people, EW represents the width of the escape exit, PF is the point of fire, PA represents the age of the person, and NE represents the number of people evacuated.

In an evacuation process, the speed depends on the density. In this paper, a quantitative expression is provided. When the pedestrian density is less than 2 people/m², people can walk in the building at an average speed of 1.2 m/s. However, when the density increases, the speed will decrease. At a density of 6 or 7 people/m², people can hardly move. The three factors of exit width, obstacles and fire burning point can be quantitatively set, and the relationship between people speed, people density, and personnel age is variable. In this paper, the relationship between the three is given by:

$$\mathrm{V}=\frac{1}{\mathrm{A}}{\mathrm{e}}^{-\frac{2\times TOT}{5\times 5}}$$

(18)

where A represents the age coefficient which can carry out the speed of people of different age levels and choose the scene environment to escape, and 5 × 5 represents the selected improved cellular automata model.

3.2.5. Improved Operating Rules and Flowcharts of Cellular Automata Evacuation Model

According to the proposed improved cellular automata building evacuation algorithm, and to the improved cellular automaton model, as shown in Figure 8a, the cell size of our assumed evacuees is $\frac{1}{5}$ m × $\frac{1}{5}$ m. The operating rules are summarized as follows:

• Each cell can only represent being occupied by one person, or occupied by obstacles, or the opening and closing state of the escape door;
• The pedestrians use statistical actions to move towards neighboring cells. The initial speed of pedestrians is uniformly 1.2 m/s, and the initial time T starts at 0 s;
• The number of escaped people is set. The escape map (including the length and width of the room, and the location of the room) is set. The location of the fire, and the width and switch state of the escape exit, are also set;
• In the process of escape, the distance between the people and each exit is first calculated. The probability of each exit is then calculated based on the distance. The escape exit is initially determined, and the potential energy method is used to check the exit during the movement. Whether there are fires or obstacles in the process of the preliminary selected escape exits, the force analysis is performed when the fires or obstacles are encountered. Aiming at the attraction of the exit and the repulsive force of obstacles, the combined force is formed to provide the effective evacuation process of people;
• Congestion will occur when a large number of people escape. The escape density is then used to determine the escape situation. The speed of escape has a certain quantitative relationship with the density of the escape. A threshold is set for the density of people. When the threshold is exceeded, the escape density is too high, and serious congestion occurs, making the speed of people equal to 0;
• The cellular automata potential energy method is used to determine the shortest escape route;
• When all the people in the building are evacuated, the evolution process is stopped. The flowchart of this process is shown in Figure 10.

The core program of the intelligent evacuation system with improved cellular automata is shown in

Table 1. Core Procedures of the Intelligent Evacuation System.

Abstract Classes and Methods	Description
Person (id, pos_x, pos_y)	Represents the evacuated crowd. The attribute has a unique identification id and a two-dimensional attribute
People (self,cnt, myMap)	self: main function, cnt: number of people, myMap: current map. This class exists as an abstract class of people
Map (self, L, W, E, B, F, A1, A2)	The method of generating a map, where the attributes include length and width, escape exit, obstacle, fire area and arrow indicator set
rmap	People density attributes of the map
thmap	Heat density properties of the map
setMapValue (self, mp, x, y, val = 0)	Methods of creating a map
addMapValue (self, mp, x, y, add = 1)	The method of adding attributes to the two-dimensional characters of the map. The parameters passed in are the heat distribution map of the map and the people
getSpeed (self, p)	Get people’s map speed
tot	Personnel density
move (self, p, dire, show = False)	A method for two-dimensional personnel to move on the map by adding algorithmic path and potential energy attributes
run	When using the Breadth First Search Algorithm (BFS) to calculate a high probability escape path, the attractive and repulsive forces of the potential energy field are added during the escape process. These two forces are based on the ignition point.
Dire	we can move in eight directions (forward, backward, turn left, turn right, front left, front right, bottom left, bottom right)
checkSavefy (self, pos)	Check if the cell 2D character is out of bounds
Init_Exit (P1, P2)	Create exit method
Init_Arrow1 or 2 (A, B, C)	Methods of creating arrows
getDeltaP (self, P1, P2)	Obtain the potential energy difference between the two cells
space	Potential energy of a two-dimensional cell. After adding the potential energy field, calculate the distance between the cell and several exits, and calculate the probability of going out from each exit based on the distance. The magnitude of this probability determines the direction of escape.
BFS (self, x, y)	Breadth First Search Algorithm

.

3.3. Evacuation Simulation and Data Analysis

The cellular automata grid is used to construct a building plan. The width of the building in the plan should be quantitatively designed according to the standard hundred people width index of the construction industry. In this paper, the evacuation path in the building evacuation process is a plane evacuation, not a staircase evacuation. The width of a single stream of people on the occasion is almost 0.55–0.60 m. The building plan designed by the cell is suitable for the construction industry standard. Each grid is a cell, which is occupied by pedestrians, obstacles, walls, and rooms. In this model, the size of the grid is set to $\frac{1}{5}$ m × $\frac{1}{5}$ m. At the same time, each grid can only be occupied by one pedestrian or empty. The evacuation time of pedestrians is sampled into equal time steps. At each moment, pedestrians move to the adjacent cell or stop waiting. In the unit of the grid, the initial speed of pedestrian evacuation is 1.2 m/s (in the horizontal or vertical direction), and 1.2$\sqrt{2}$ m/s along the diagonal direction. The speed and density of people will then change with the evacuation time. Pedestrians cannot cross the fence, and cannot leave the set building plan system. In the algorithm, after the escape route is planned, the route cannot be used as the escape route or the indicator light on the route to a fire point turning red, which indicates that the road is blocked. By the simulation algorithm on the edge computing, it is possible to effectively know the hardware indicator light to achieve the evacuation effect.

In the process of evacuation, many human factors are taken into account. This algorithm takes into account the speed of personnel escape, the density of personnel escape, the size of the fire point, the width of the escape exit, and the detection factors of the exit switch. The potential energy field and the element, as well as the combination of cellular automata, form a path planning algorithm, which is more in line with the actual escape. In the simulation process, a cell is set, that is, a person’s escape trajectory, in order to visually understand the escape process of evacuation, as shown in Figure 11. In the latter, the yellow rectangle is the hypothetical ignition point. The pedestrian escape route marked in blue in is 1→2→3→4→5→6, and finally an effective escape is realized. This system can directly display the entire process of personnel escape. The process of personnel escape is a dynamic process. Therefore, it is a dynamic graph. Note that the algorithm is shared on Github.

In Figure 11, a dynamic analysis of a single cell is performed at an interval of 5 s. In the evacuation process, each cell needs to use the cell movement probability selection shown in Figure 9 and the force analysis of the potential energy field method to obtain the most suitable path. The exit path of escape can efficiently avoid the point of fire.

3.4. Comparison between the Improved Cellular Automata Personnel Evacuation Algorithm with Other Path Planning Simulations

The designed people evacuation algorithm compares the evacuation of the same building with the currently most used personnel evacuation algorithm which is the Dijkstra algorithm. The Dijkstra algorithm is a widely used shortest path search approach. Its function is to find a node in the graph. The shortest path between different nodes is determined, and the least cost path from any given destination to the source node is extracted in the process, as shown in Figure 11. Fourteen people are set up in the building to evacuate at the same time interval contrast.

It can be seen from Figure 12 that the proposed intelligent building evacuation algorithm based on improving cellular automata has an obvious evacuation efficiency for evacuation in the same scene. Compared with Dijkstra’s algorithm, it reduces evacuation time by 55 s and it is more effective for evacuation of people.

4. Case Study

The intelligent evacuation system developed in this paper is simulated in the context of the emergency center of Beijing Capital International Airport. Beijing Capital International Airport Emergency Center is located in Chaoyang District, Beijing, and is used to respond to sudden illnesses of airport personnel. Due to the special nature of the emergency center, we hope more than ever that the emergency center will be more effective in evacuating people in case of emergencies such as fire, which is the significance of our research on intelligent evacuation. With adequate access to emergency center building information, the improved cellular automata algorithm of this system is used to effectively perform personnel evacuation simulation. We need to clarify that we cannot evacuate patients from their wards during a simulated evacuation; our work is only a simulation of the most hospital scenario. With the evacuation system developed in this paper, we are able to consider more evacuation factors and study more comprehensively the evacuation of some special buildings such as medical or school. The internal structure of the building is relatively simple, as shown in Figure 13. All the fire-fighting measures are ignored. Fire points and other facilities are set up for the simulation.

The simulation of the floor plan of the west side building of the Capital Airport Emergency Center (visual interface, dynamic analysis, and guidance) can effectively guide people of different ages to efficiently evacuate, and perform the hypothesis and simulation analysis that are developed. This theory opens up a new analysis method under the artificial intelligence environment for the evacuation of buildings. At the same time, according to the building code, the evacuation time of the first and second fire resistance buildings is controlled within 2 min, and the evacuation time of the third fire resistance buildings is controlled. Within 1.5 min, it meets the construction industry standards, and is better than the industry standards. During the simulation process, a comparative analysis of the different parameters in the process of personnel escape is performed, as shown in

Table 2. Comparison of the simulation parameters.

Number of Evacuated (People)	Map Selection	Different Ignition Points	Exit Width (m)	Final Evacuation Time (s)	Exit Width (m)	Final Evacuation Time (s)
50	Map 1	Fire point 1	5	24	10	22
		Fire point 2	5	28	10	26
		Fire point 3	5	23	10	22
		No fire	5	21	10	20
	Map 2	Fire point 1	5	96	10	94
		Fire point 2	5	51	10	38
		Fire point 3	5	40	10	39
		No fire	5	46	10	38
200	Map 1	Fire point 1	5	34	10	38
		Fire point 2	5	30	10	28
		Fire point 3	5	32	10	31
		No fire	5	30	10	30
	Map 2	Fire point 1	5	109	10	33
		Fire point 2	5	47	10	45
		Fire point 3	5	47	10	45
		No fire	5	46	10	44

.

In Table 2, certain presets are made for the speed and density of people in the building according to the standards of the construction industry. The opening and closing of the escape door in the building are completed by the algorithm. Through the visual selection interface, the dynamics of the exit width, the number of escaped and the fire point are set to observe the effect on the final escape time. It can be deduced from Table 2 that, as the width of the exit becomes larger, the escape time is improved. However, the difference of the firing point may have a certain impact on the escape time, but the potential energy field theory added in the proposed algorithm is the solution to this problem. The potential energy field theory can efficiently determine the ignition point of the best exit, which is also an innovation in this paper. In general, the designed improved cellular automata algorithm effectively improves the evacuation time. Through path planning, the results are sent to a wireless emergency light part to efficiently guide the evacuation of the personnel.

5. Conclusions

This paper studies a novel dynamic cellular automata algorithm and its related intelligent IoT hardware system for better guidance in building fire evacuation. This system uses IoT sensors and AI-enhanced surveillance video cameras to collect information on the location and number of people in the building and the location of fires, and transmit the collected information to the edge computing gateway for rapid analysis. An improved cellular automata algorithm was developed. In addition to using the omnidirectional cellular movement model, a potential field theory based on calculating the exit distance and the optimal evacuation probability was also designed. The comparison with the existing shortest path algorithms shows that the dynamic evacuation algorithm we study is superior in evacuation timeliness and coping with changing situations. Many factors need to be considered in densely populated and complex buildings, such as schools and hospitals, and the evacuation process is complex. These improved cellular automata and corresponding IoT intelligent evacuation algorithms present a new idea to evacuate in these scenarios.