Fire emergency command system and method based on dynamic fire scene modeling and double-algorithm cooperation

Abstract

The invention discloses a fire emergency command system and method based on dynamic fire scene modeling and double-algorithm cooperation. The system comprises a fire scene intelligent map system, a fireman terminal, an unmanned fire fighting equipment terminal and external information. The method comprises the steps that the information management layer extracts the latitude and longitude of a fire point, calculates a spreading rate and a high-risk area mark and pushes the same to the display module; a commander issues a fire-fighting instruction to an AI auxiliary decision-making layer; a fire fighting task plan and a target fire point priority ranking table are generated by a target distribution module and a Hungary algorithm of the AI auxiliary decision-making layer; the AI auxiliary decision-making layer drives a fire extinguishing planning module to cooperate with an action path through an MADDPG algorithm, plans an unmanned aerial vehicle group airdrop route optimization and robot group breakthrough route, carries out data and instruction interaction to generate an end-to-end execution instruction set, and transmits the instruction set to the instruction layer; and the instruction layer sends the generated fire extinguishing instruction to each terminal to guide the firefighter and the unmanned equipment to extinguish the fire. According to the invention, full-chain optimization of fire rescue is realized through multi-dimensional technology fusion.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent fire fighting technology and fire fighting command, and particularly relates to a fire emergency command system and method based on dynamic fire scene modeling and double algorithm cooperation. BACKGROUND

[0002] With the acceleration of urbanization and the increasing complexity of fire scenes, once a fire occurs, if the fire fighting forces cannot be timely commanded to control the spread of the fire, it will cause immeasurable economic losses and even casualties. The traditional fire emergency command system is limited by the difficulty of multi-platform information integration, the lack of intelligent decision support, and the difficulty of cross-platform cooperation, making it difficult to quickly form effective fire scene information integration and disaster relief command. In recent years, many fire command systems have emerged, showing many advantages in rapid command and information integration, but still have many limitations.

[0003] Chinese patent application CN 119838173A discloses a forest fire monitoring and extinguishing system based on unmanned aerial vehicle cluster and reinforcement learning, which identifies fire through multi-source data fusion, but has too strong environmental dependence, and communication depends on high-speed transmission network, making it difficult to integrate fire scene conditions.

[0004] Chinese patent application CN 119733200A discloses a fire fighting command system integrating sensor network, AI decision and real-time scheduling, which uses hierarchical-gray correlation analysis method to optimize resource scheduling. The system algorithm lacks real-time performance and is difficult to quickly command rescue and fire fighting.

[0005] Therefore, it is urgent to build an intelligent command framework of "perception-decision-execution" closed loop, integrate multi-platform fire scene data, quickly build a dynamic fire scene map, and use AI technology to effectively command unmanned devices and firefighters. SUMMARY

[0006] The present application discloses a fire emergency command system and method based on dynamic fire scene modeling and double algorithm cooperation, which integrates the vital sign data of firefighters, geographic information system data, and unmanned fire fighting device data through a fire intelligent map system, and intuitively displays them to commanders. The control system allocates fire extinguishing resources through AI assisted decision making based on existing trends, predicted conditions, and instructions from commanders, according to fire scene environment information and predicted fire development. Finally, the generated fire extinguishing instructions are sent to each terminal by the instruction layer to guide firefighters and unmanned devices to extinguish the fire.

[0007] Technical scheme: The fire emergency command system based on dynamic fire scene modeling and double algorithm cooperation of the present application comprises a fire intelligent map system, a firefighter terminal, an unmanned fire fighting device terminal, and external information.

[0008] The fire site intelligent map system comprises a display module, an information management layer, an AI assisted decision making layer and an instruction layer; the information management layer obtains the longitude and latitude of the fire point and calculates the spreading rate according to the fire site environment and personnel equipment information returned by the external information and the unmanned fire equipment terminal and firefighter terminal, analyzes the fire site thermal radiation distribution map, analyzes the key fire site data of the terrain and topography and constructs the real-time information of the fire site; the AI assisted decision making layer issues the fire fighting task and fire fighting instruction of the firefighters and the unmanned fire equipment, and receives the real-time information of the fire site output by the information management layer.

[0009] The AI assisted decision making layer generates an instruction set for the firefighter terminal and the unmanned equipment terminal according to the fire fighting strategy, the fire fighting instruction and the real-time information of the fire site.

[0010] The display module comprises a tile map service, an intelligent map system front end and an intelligent map system back end; the tile map service and the intelligent map system back end interact through the HTTP protocol; the intelligent map system back end and the intelligent map system front end interact through the HTTP protocol and the WebSocket protocol.

[0011] The AI assisted decision making layer is composed of a target allocation module, a Hungarian algorithm, a fire extinguishing planning module and a MADDPG algorithm; the target allocation module is driven by the Hungarian algorithm and generates a target fire point priority ranking table; the fire extinguishing planning module is driven by the MADDPG algorithm.

[0012] The fire emergency command method based on dynamic fire site modeling and double algorithm cooperation of the present application comprises the following steps:

[0013] 1) The information management layer tracks the geographic position coordinates of the firefighters through the position information returned by the firefighter terminal, and monitors the physical condition of the firefighters through the risk index R:

[0014] 2) The information management layer constructs a dynamic fire site map, determines the number of personnel in the fire area according to the GNSS positioning information in the external information, updates the image of the fire area according to the visible light image, and updates the intelligent map according to the physical condition of the disaster relief personnel fed back by the firefighter terminal;

[0015] 3) The information management layer uploads the fire site environment information, external information and personnel equipment information to the display module;

[0016] 4) The information management layer updates the fire site map and equipment, extracts the longitude and latitude of the fire point, calculates the spreading rate and high-risk area marking and pushes to the display module; the process of calculating the spreading rate is:

[0017] The level set function φ(x, y, t) is used to implicitly represent the fire line: Γ(t)={(x, y)∈Ω|φ(x, y, t)=0}

[0018] Where, φ(x, y, t) < 0 is the burned area, and φ(x, y, t) > 0 is the unburned area.

[0019] Fire line normal propagation velocity: V n = V0+ V w + V s

[0020] Where, V0 is the basic rate; wind speed driven term V w and slope driven term V s are:

[0021] V w = β w ·||W||·|cosθ| V s = β s ·tanα·I {▽h·n＞0}

[0022] θ is the angle between the wind direction and the fire line normal; α is the slope angle; β w and β s are the driven term coefficients;

[0023] The level set evolution equation is discretized using an upwind difference format:

[0024]

[0025] Where, is a first-order upwind difference operator; the instantaneous spread rate at the fire line point is:

[0026]

[0027] The cumulative exposure of thermal radiation is determined and normalized to obtain the thermal radiation risk factor:

[0028]

[0029] The structure risk factor R S is composed of static risk S static and dynamic risk S dynamic :

[0030] R S = 0.3S static + 0.7S dynamic

[0031] The dynamic risk is expressed as:

[0032] Where, δ is the structure deformation, C gas is the toxic gas concentration, and σ p is the standard deviation of water pressure fluctuation;

[0033] The escape difficulty risk factor RE is the shortest escape cost

[0034] where C esc is the shortest escape cost ρ(s) is the path cost function;

[0035] 5) The commander issues fire-fighting instructions to the AI auxiliary decision layer; the target assignment module and the Hungarian algorithm of the AI auxiliary decision layer generate fire-fighting task planning and target fire point priority ranking table for unmanned devices and firefighters, which are passed to the instruction layer; the process is:

[0036] The fire point target value function is regarded as f i and the value of the fire point is determined: f i = αV i + βT i + γP i

[0037] wherein V i is the target comprehensive value evaluation score, T i is the target threat level, P i is the target protection capability, and α, β, γ are weight coefficients;

[0038] The data set structure is constructed as: Data = {(g1, V1, T1, P1),..., (g N , V N , T N , P N )}

[0039] g1, g2,..., g N are fire locations, and e1, e2,..., e M are combat elements;

[0040] The cost calculation formula between the target and the combat element is:

[0041] wherein V gi is the comprehensive value of the target g i , d(g i , e j ) is the distance between the fire location g i and the combat element e j , and d max is the maximum distance between all targets and combat elements;

[0042] Suppose that the fire-fighting resource limit of the participating unit is R j , the Hungarian algorithm is used to maximize the attack on the fire location, the constraint condition is that the resource of each fire-fighting unit is less than the maximum fire-fighting capability, and the optimization model is:

[0043]

[0044] wherein, r ij is the fire distribution amount of the fire fighting unit j to the target i, and the optimal distribution of the fire resource is performed by solving the fire distribution weight w ij and the fire resource distribution amount r ij .

[0045] 6) The AI assisted decision-making layer drives the fire extinguishing planning module to generate a coordinated action path through the MADDPG algorithm, plans the UAV group air drop route optimization and robot cluster breakthrough route, and generates an end-to-end execution instruction set through data and instruction interaction through the Http protocol and the WebSocket protocol, and transmits it to the instruction layer;

[0046] 7) The instruction layer issues the marching route and coordinated timing fire extinguishing instructions to the unmanned fire fighting equipment terminal, and the unmanned fire fighting equipment terminal controls the unmanned equipment to extinguish the fire; the instruction layer sends the fire extinguishing task to the firefighter terminal and pushes the issued instructions to the display module;

[0047] 8) The firefighter extinguishes the fire according to the fire field information displayed by the firefighter terminal and the fire extinguishing task issued by the instruction layer, and the firefighter terminal detects and uploads the vital sign data of the firefighter to the information management layer;

[0048] 9) The unmanned fire fighting equipment terminal receives the instructions of the instruction layer, the fire unmanned plane collects and returns the fire field data to the information management layer, and the fire unmanned vehicle extinguishes the fire and transmits the fire field data to the information management layer;

[0049] 10) The information management layer processes the information returned by the unmanned fire fighting equipment terminal, calculates the spreading rate and high-risk area, and uploads the updated real-time fire field information to the display module;

[0050] 11) The commander judges the fire field situation and fire fighting strategy according to the updated real-time fire field information of the display module, changes or maintains the command to the AI assisted decision-making layer;

[0051] 12) Steps 9)-11) are circularly executed until the fire is controlled.

[0052] In step 1), the information management layer receives the external information source data composed of map information, visible light image information and GNSS positioning information, synchronously acquires the personnel state data uploaded by the firefighter terminal and the fire field environment parameters and equipment operating state feedback by the unmanned fire fighting equipment terminal.

[0053] In step 1), the information management layer tracks the geographic position coordinates of the firefighters through the position information returned by the firefighter terminal, and monitors the physical condition of the firefighters through the risk index R:

[0054] R = a · HR + b · SpO2 + g · T norm 2norm + y · T norm + l1 · HR D + l2 · SpO2 2D + l3 · T D

[0055] wherein HR norm denotes heart rate, T norm denotes body temperature, SpO2 denotes blood oxygen saturation, R is risk index, HR D denotes heart rate, T D denotes body temperature, SpO2 2D denotes rate of change of blood oxygen saturation:

[0056]

[0057] t1 is n times of system refresh frequency; a, b, g, l1, l2, l3 are influence weights of heart rate, body temperature, blood oxygen saturation and their rates of change, which are obtained based on clinical data.

[0058] In step 6), the MADDPG algorithm is as follows: the state space of each fire-fighting unit includes two-dimensional coordinate position (x t ,y t ), velocity v t , and acceleration a t of the fire-fighting unit at time step t, denoted as vector state t :

[0059] state t = [x t ,y t ,v t ,a t ]

[0060] The action space of each fire-fighting unit is composed of displacement

[0061] (Δx t ,Δy t ) and velocity change Δv t in horizontal and vertical directions at time step t, denoted as vector action t : action t = [Δx t ,Δy t ,Δv t ]

[0062] The state update of the fire-fighting unit is as follows:

[0063] Δt is time step, (x t ​y t is the current position of the fire-fighting unit, v t is the speed, a t is the acceleration.

[0064] In step 9), the fire unmanned vehicle starts the fire extinguishing device, and uploads the water pressure, device data, fire scene monitoring image, and fire scene temperature to the information management layer.

[0065] In step 9), the fire unmanned vehicle performs high-altitude fire extinguishing agent delivery, high-altitude collection of fire scene environment, real-time return of fire scene image, fire scene temperature, fire scene mutation and device data to the information management layer (3).

[0066] Working principle: The display module of the present application pushes information to the mobile terminal for reference decision-making by the commander, and the firemen master the fire situation. The information management layer is the core data hub of the fire intelligent command system, which real-time gathers the multi-dimensional fire environment information including remote sensing map information, fire temperature and combustible state through external information and terminal return fire data, and processes the fire environment information, and then pushes the fire situation to the display module of the mobile terminal.

[0067] The AI assisted decision-making layer is composed of two algorithms based on deep learning network: fire extinguishing task target allocation algorithm and multi-agent collaborative action planning algorithm. The fire extinguishing task precise target allocation algorithm is based on fire data set, which evaluates the task value through fire intensity and spread risk, optimizes the resource coverage range through fire equipment range and path accessibility, and combines the Hungarian algorithm for target allocation. The multi-agent collaborative fire extinguishing planning algorithm adopts MADDPG (Multi-Agent Deep Deterministic Policy Gradient) algorithm, which optimizes the device collaborative path on the basis of task allocation to solve the problems of multi-device obstacle avoidance, resource complementation and timeliness collaboration.

[0068] The instruction layer is responsible for distributing the generated fire extinguishing instructions to the device terminal, and the terminal links to manage the fire unit cluster including fire unmanned vehicles and fire unmanned vehicles. The fire units are based on radio communication protocol to build collaborative networking, and the command terminal realizes instruction transmission and state monitoring through standardized interface, and synchronously returns the fire environment data to the fire intelligent map system.

[0069] Various devices in the device terminal of the system are based on the fire extinguishing strategy issued by the instruction layer, and realize accurate task execution and dynamic feedback through multi-modal perception. After receiving the fire point coordinates and fire extinguishing agent delivery parameters, the fire-fighting unmanned aerial vehicle autonomously plans the flight path combining GNSS (Global Navigation Satellite System) positioning and visual obstacle avoidance algorithm, executes high-altitude fire line monitoring and accurate delivery tasks, and real-time returns visible light / thermal infrared images to the information management layer; the fire-fighting unmanned vehicle enters the core area of the fire scene, and real-time collects temperature, gas concentration and other data through multi-sensor fusion, and synchronously updates to the intelligent map system; the devices rely on the dual-mode communication protocol to build a self-organizing network, and realize the transmission and state monitoring through the standardized interface.

[0070] The health assessment model built in the firefighter device terminal real-time warns the device failure risk, and sends an alarm when detecting that the firefighter is in danger. During the execution process, the state data of the terminal and the multi-source information of the fire scene are real-time fused, the AI assisted decision layer is driven to dynamically optimize resource allocation and path planning, and finally the "perception-decision-execution-feedback" whole chain closed loop control is realized through the intelligent map, so that the fire extinguishing efficiency and system reliability in complex fire scene environment are improved.

[0071] Advantages: Compared with the prior art, the core advantage of the fire emergency command system based on dynamic fire scene modeling and double algorithm cooperation in the present application is that the whole chain optimization of fire rescue is realized through multi-dimensional technology fusion; the emergency command system breaks through the limitation of traditional information island by integrating satellite remote sensing, device sensing and firefighter vital signs and other multi-source data to build a dynamic fire scene intelligent map, realizes global situation visualization and real-time interaction; the AI assisted decision layer adopts a deep learning model and a multi-agent cooperative algorithm to dynamically generate task priority ranking; combined with the resource scheduling mechanism, the cooperative path and obstacle avoidance strategy of the unmanned aerial vehicle group, the fire extinguishing robot and the manned device are optimized to form a three-dimensional fire extinguishing network in the air and on the ground, which significantly improves the device reliability and personnel safety. Compared with the prior art, the system innovatively combines dynamic fire scene evolution and deep reinforcement learning to realize breakthroughs in decision-making timeliness, resource utilization and complex scene adaptability, and provides an efficient closed-loop solution for high-rise buildings, forest fires and other scenes. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 It is a structural schematic diagram of the fire emergency command system based on dynamic fire scene modeling and double algorithm cooperation in the present application;

[0073] Figure 2 It is a schematic diagram of the display module in the embodiment of the present application;

[0074] Figure 3 It is a schematic diagram of the AI assisted decision layer in the embodiment of the present application;

[0075] Figure 4 is a schematic diagram of a set of system operation in an embodiment of the present application; wherein

[0076] Figure 4(a) is a schematic diagram of system initialization and terminal registration;

[0077] Figure 4(b) is a schematic diagram of fire scene dynamic map construction and information uploading;

[0078] Figure 4(c) is a schematic diagram of multi-dimensional fire scene information visualization interaction;

[0079] Figure 4(d) is a schematic diagram of fire spread modeling and high-risk area updating;

[0080] Figure 4(e) is a schematic diagram of commander instruction decision and issuance;

[0081] Figure 4(f) is a schematic diagram of multi-agent collaborative task allocation;

[0082] Figure 4(g) is a schematic diagram of layered execution of fire extinguishing instructions;

[0083] Figure 4(h) is a schematic diagram of firefighter task execution and vital sign monitoring;

[0084] Figure 4(i) is a schematic diagram of unmanned equipment collaborative fire extinguishing and data backhaul;

[0085] Figure 5 Figure 4(j) is a schematic diagram of display module information. DETAILED DESCRIPTION

[0086] As shown in the figure, the fire emergency command system based on dynamic fire scene modeling and dual algorithm collaboration of the present application includes a fire intelligent map system 1, a firefighter terminal 6, an unmanned fire equipment terminal 7, and external information 8. Figure 1 The fire intelligent map system 1 is composed of a display module 2, an information management layer 3, an AI assisted decision making layer 4, and an instruction layer 5, and is used to provide fire scene information and instructions to personnel and equipment.

[0087] The display module 2 is used for data visualization, receives real-time fire scene information processed by the information management layer 3 and fire extinguishing instruction information output by the instruction layer 5, and presents intuitive fire scene information and equipment terminal information to the commander and the firefighter terminal 6.

[0088]

[0089] ​The information management layer 3 is used for receiving and processing fire environment, personnel and equipment information, and outputting visual information and fire data; the information management layer 3 receives three kinds of external information 8 including remote sensing map information, GNSS positioning information and visible light image information, and fire environment information and personnel and equipment information returned by the unmanned fire-fighting equipment terminal 7 and the firefighter terminal 6; the information management layer 3 obtains the longitude and latitude of the fire point, calculates the spread rate, analyzes the fire heat radiation distribution map, analyzes the key fire data of the terrain and topography, and constructs the real-time fire information according to the above information.

[0090] The AI assisted decision-making layer 4 issues fire-fighting task planning and fire-fighting instructions for firefighters and unmanned fire-fighting equipment, and receives the real-time fire information output by the information management layer 3 and the fire-fighting strategy and instructions issued by the commander; the AI assisted decision-making layer 4 has a target allocation algorithm and a fire extinguishing planning algorithm, and generates a set of instructions for the firefighter terminal 6 and the unmanned equipment terminal 7 according to the fire-fighting strategy, the fire-fighting instructions and the real-time fire information.

[0091] The instruction layer 5 is used for the collection and generation of instructions, and receives the instruction set of the AI assisted decision-making layer 4; the instruction layer 5 respectively issues real-time fire-fighting instruction information to the firefighter terminal 6 and the unmanned fire-fighting equipment terminal 7 according to the received instruction set.

[0092] The firefighter terminal 6 receives the fire-fighting instructions issued by the instruction layer 5; at the same time, the firefighter terminal 6 also has a visualization function, and the firefighters can real-time refer to the terminal information and the fire scene intuitive information; in addition, the firefighter terminal 6 also detects the vital sign data such as heart rate and blood pressure of the firefighters, and the personnel information such as position information, which are automatically uploaded to the information management layer 3 in real time.

[0093] The unmanned fire-fighting equipment terminal 7 is composed of a fire unmanned plane and a fire unmanned vehicle, wherein the fire unmanned plane collects the fire environment from high altitude and performs high-altitude fire extinguishing agent delivery task, and the fire unmanned vehicle collects the fire environment on the ground and is equipped with high-pressure water cannon and various fire extinguishing equipment; the unmanned fire-fighting equipment terminal 7 receives the instruction information input into itself, and simultaneously uploads the fire scene images, fire scene temperature, fire scene mutation data and equipment data such as equipment power and equipment working condition collected by the unmanned equipment to the information management layer 3 in real time.

[0094] As shown in Figure 2 The display module 2 includes a tile map service 21, an intelligent map system front end 22 and an intelligent map system back end 23; wherein the tile map service 21 is used for providing map basic data support, and interacts with the intelligent map system back end 23 through HTTP protocol; the intelligent map system back end 23 realizes the coding of the overall function logic of the system, and interacts with the intelligent map system front end 22 through HTTP protocol and WebSocket protocol for data and instruction interaction.

[0095] As shown in Figure 3As shown, the AI-assisted decision-making layer 4 is composed of a target allocation module 41, a Hungarian algorithm 42, a fire extinguishing planning module 43, and a MADDPG algorithm 44. The target allocation module 41 is driven by the Hungarian algorithm 42 to optimize the fire extinguishing material allocation strategy, and generate a target fire point priority ranking table in combination with device information including device range, path accessibility, and fire extinguishing agent supply nodes. The fire extinguishing planning module 43 is driven by the MADDPG algorithm 44 to coordinate action paths, plan unmanned aerial vehicle group air drop route optimization and robot cluster breakthrough route, and generate end-to-end execution instruction set.

[0096] As shown in FIGS. 4(a)-(i), the fire emergency command method based on dynamic fire field modeling and dual algorithm cooperation of the present application includes the following steps:

[0097] 1) The information management layer 3 receives external information source data composed of GIS (Geographic Information System) map information, visible light image information, and GNSS (Global Navigation Satellite System) positioning information, synchronously acquires personnel state data (such as heart rate, body temperature, blood pressure) uploaded by the firefighter terminal 6, and fire field environment parameters (such as fire field temperature, fire field image, fire field mutation) and device operating state (such as device power, device temperature, device fire extinguishing material remaining amount) feedback by the unmanned fire-fighting device terminal 7.

[0098] 2) The information management layer 3 tracks the geographic position coordinates of the firefighters through the location information in the personnel information returned by the firefighter terminal 6, and monitors the physical condition of the firefighters in real time by using the risk index fusion model to calculate the risk index from the vital sign information (heart rate, body temperature, blood oxygen saturation). The mathematical model of the risk index fusion model is:

[0099] R = a · HR + β · SpO + γ · T norm + λ1 · HR + λ2 · SpO + λ3 · T 2norm norm D 2D D

[0100] Wherein, HR norm , T norm , SpO2 represent the numerical values of heart rate, body temperature, and blood oxygen saturation, respectively, and R is the risk index. HR D , T D , SpO 2D represent the change rates of heart rate, body temperature, and blood oxygen saturation, respectively, and are calculated by the following formula:

[0101]

[0102] t1 is n times the system refresh frequency, determined by actual demand. a, β, γ, λ1, λ2, and λ3 are obtained based on clinical data.​​​​

[0103] 3) The unmanned fire-fighting equipment terminal 7 performs self-checking on the internal unmanned fire-fighting equipment before receiving the instruction from the instruction layer 5, including the power of the unmanned equipment, the stock of fire extinguishing materials, etc. The information management layer 3 stores the equipment information returned by the unmanned fire-fighting equipment terminal 7.

[0104] 4) The information management layer 3 obtains the map information of the fire area (including the terrain, vegetation, buildings, vegetation, temperature, etc. The information that can be provided by the accessed GIS) through the GIS map information in the external information 8, takes the topographic map as the bottom layer, and marks the information such as the building, vegetation, temperature, and thermal map on it to preliminarily construct the dynamic fire field intelligent map. The number of personnel in the fire area is determined according to the GNSS positioning information in the external information 8, and the image of the fire area is further updated according to the visible light image information. The physical condition of the rescue personnel in the fire field is updated in real time through the personnel information feedback by the firefighter terminal to further update the intelligent map.

[0105] 5) As shown in FIG. 4(b), the information management layer 3 uploads the fire field environment information, external information, and personnel equipment information to the display module 2.

[0106] 6) As shown in FIG. 4(c), the display module 2 visually displays the fire field intuitive information and equipment terminal information and supports the interactive viewing of the commanders and firefighters.

[0107] 7) As shown in FIG. 4(d), the information management layer 3 first updates the fire field map through the external information 9, and then updates the equipment situation in the fire field map based on the data returned by the unmanned fire-fighting equipment terminal 7, extracts the longitude and latitude of the fire point, calculates the spread rate and high-risk area mark, and pushes them to the display module 2. In one scheme, the specific mathematical model of the calculation of the spread rate is:

[0108] The level set function φ(x, y, t) is used to implicitly represent the fire line:

[0109] Γ(t) = {(x, y) ∈ Ω | φ(x, y, t) = 0}

[0110] Where φ(x, y, t) < 0 is the burned area, and φ(x, y, t) > 0 is the unburned area.

[0111] The normal propagation speed of the fire line is:

[0112] V n = V0+ V w + V s

[0113] Where the basic rate V0is determined by the type of combustible material. The wind speed driving term V w and the slope driving term V s are:

[0114] V w = β w ·||W||·|cosθ|

[0115] V s = β s ·tanα·I {▽h·n＞0}

[0116] θ and α are the angles between the wind direction and the normal of the fire line and the slope angle β w and β s is the driving term coefficient.

[0117] The level set evolution equation is discretized by using the upwind difference scheme:

[0118]

[0119] where is the first-order upwind difference operator. The instantaneous propagation rate at the fire line point is:

[0120]

[0121] In one scheme, the specific mathematical model for calculating the high-risk area is as follows: the high-risk index is obtained by multiplying three key factors, R T is the thermal radiation risk factor, R S is the structure risk factor, and R E is the escape difficulty risk factor.

[0122] The thermal radiation cumulative exposure is defined and normalized to obtain the thermal radiation risk factor:

[0123]

[0124] The structure risk factor R S is composed of the static risk S static and the dynamic risk S dynamic :

[0125] R S = 0.3S static + 0.7S dynamic

[0126] The static risk factor of typical building types is as follows:

[0127]

[0128] The dynamic risk is represented as:

[0129]

[0130] where δ is the structure deformation amount, and C gasσ is the concentration of toxic gas p σ is the standard deviation of water pressure fluctuation.

[0131] Risk factor R of escape difficulty E is defined as follows:

[0132]

[0133] where C esc is the shortest escape cost expressed as ρ(s) is the path cost function.

[0134] 8) As shown in FIG. 4(e), the commander grasps the intuitive information of the fire scene and the terminal information of the equipment in real time through the display module 2, comprehensively studies and judges the development situation of the fire scene and the personnel and equipment situation, and then issues a fire-fighting command to the AI auxiliary decision layer 4.

[0135] 9) As shown in FIG. 4(f), the target allocation module 41 and the Hungarian algorithm 42 of the AI auxiliary decision layer 4 generate a fire-fighting task planning of the unmanned equipment and the firefighters, generate a target fire point priority ranking table in combination with the equipment information including the equipment range, path accessibility and fire extinguishing agent supply node, and deliver the table to the instruction layer 5. In one scheme, the target allocation module adopts the Hungarian algorithm, and the mathematical model is specifically:

[0136] The target value function of the fire point is regarded as f i Then the value of each fire point is comprehensively evaluated according to multiple characteristics:

[0137] f i = αV i + βT i + γP i

[0138] wherein V i is the comprehensive value score of the target, T i is the threat level of the target, P i is the protection ability of the target, and α, β and γ are weight coefficients. The comprehensive value f i of the target is calculated in a weighted sum manner, and the weight coefficients α, β and γ are adjusted by a machine learning method.

[0139] There are multiple fire locations g1, g2,..., g N and multiple combat elements e1, e2,..., e M . We will build a database containing target attributes, including threat level, target type, distance from each combat element, etc. The data set structure is as follows:

[0140] Data = {(g1, V1, T1, P1),..., (g NV N ,T N ,P N )}

[0141] The cost calculation formula between the target and the combat element is as follows:

[0142]

[0143] Wherein V gi is the comprehensive value of the target g i ; d(g i , e j ) is the distance between the fire location g i and the combat element e j . d max is the maximum distance between all targets and combat elements.

[0144] Assuming that the fire extinguishing resource limit of each participating fire fighting unit is R j , the Hungarian algorithm is used to maximize the attack effect on the fire location, with the constraint condition that the resource of each flat fire extinguishing unit cannot exceed its maximum fire extinguishing capacity, and the optimization model is as follows:

[0145]

[0146]

[0147] Wherein, r ij is the fire distribution amount of the fire fighting unit j to the target i, by solving the fire distribution weight w ij and the fire resource distribution amount r ij , the optimal allocation of fire resources is realized.

[0148] 10) The AI assisted decision making layer 4 drives the fire extinguishing planning module 43 to cooperate with the action path through the MADDPG algorithm 44, plans the unmanned aerial vehicle group air drop route optimization and robot cluster breakthrough route, and generates end-to-end execution instruction set through Http protocol and WebSocket protocol Data and instruction interaction, pass to the instruction layer 5. In one scheme, the mathematical model of the MADDPG algorithm is as follows:

[0149] The state space of each agent (fire fighting unit) includes the two-dimensional coordinate position (x t , y t ) of the agent at time step t, the speed v t of the agent, and the acceleration a t of the agent, which can be represented as a vector state t :

[0150] state t = [x t,y t ,v t ,a t ]

[0151] The action space for each agent consists of the displacement in the horizontal and vertical directions (Δx t ,Δy t ) and the change in velocity Δv t at time step t for the agent, which can be represented as a vector action t :

[0152] action t =[Δx t ,Δy t ,Δv t ]

[0153] The reward function can be defined by the following four factors,

[0154] Task completion reward: a higher reward is given when the agent reaches the goal position.

[0155] Path length reward: the shorter the path the agent travels, the higher the reward.

[0156] Collision penalty: a negative reward is given if the agent collides with other agents or obstacles.

[0157] Velocity reward: encourages the agent to choose a path with higher velocity to reduce the time needed to reach the goal.

[0158] The specific reward function can be represented as:

[0159] r t =ω1r task +ω2r path -ω3r collision -ω4r speed

[0160] where r task is the reward for task completion, which is positive if the agent reaches the goal position and zero otherwise. r path is the reward for path length, which is inversely proportional to the path length. r collision is the penalty for collision, which is negative when a collision occurs. r speed is the reward related to velocity, which is larger for higher agent velocities. w1, w2, w3, w4 are the weight coefficients of the reward function.

[0161] The dynamic modeling of the environment includes how the agent selects actions based on the state and how the state is updated through these actions. In this training environment, the state update for the agent follows the equation:

[0162]

[0163] where Δt is the time step, (x t ,y t ) is the current position of the agent, v t is the velocity, and a t is the acceleration.

[0164] 11) As shown in Figure 4(g), the instruction layer 5 issues the travel route, coordination and timing fire extinguishing instructions to different unmanned fire equipment terminals 7, and the unmanned fire equipment terminals 7 start to control the unmanned equipment to execute the fire extinguishing task; the instruction layer 5 also sends the fire extinguishing task to the firefighter terminal 6; and the instruction layer 5 pushes the issued instructions to the display module 2.

[0165] 12) As shown in Figure 4(h), the firefighters participating in the fire fighting task start to execute the fire extinguishing operation according to the multi-dimensional fire scene intuitive information displayed by the firefighter terminal 6 and the fire extinguishing task issued by the instruction layer 5, and the firefighter terminal 6 detects and uploads the vital sign data of the firefighters in the fire scene to the information management layer 3 in real time.

[0166] 13) As shown in Figure 4(i), the unmanned fire equipment terminal 7 receives the instructions of the instruction layer 5, and the fire unmanned plane starts to execute the tasks of high-altitude fire extinguishing agent delivery and high-altitude collection of fire scene environment, and real-time returns the fire scene image, the real-time situation of the fire scene such as fire scene temperature and fire scene mutation, and the equipment data such as equipment power to the information management layer 3; the fire unmanned vehicle starts multiple fire extinguishing equipment such as high-pressure water cannon and starts to execute the fire extinguishing task, and uploads the equipment data such as water pressure and range, the fire scene monitoring image, and the real-time situation of the fire scene such as fire scene temperature to the information management layer 3.

[0167] 14) The information management layer 3 further calculates the spreading rate and the high-risk area based on the algorithm described in the steps to process the information returned by the unmanned fire equipment terminal 7, and uploads the updated real-time information of the fire scene to the display module 2.

[0168] 15) The commander researches and judges the fire scene situation and the fire fighting strategy according to the updated real-time information of the fire scene of the display module 2, changes or maintains the order, to the AI auxiliary decision layer 4.

[0169] 16) The processes 9)-15) are executed in a cycle until the fire is completely controlled.

Claims

1. A fire emergency command system based on dynamic fire field modeling and double algorithm cooperation, characterized in that: The fire scene intelligent map system (1), the firefighter terminal (6), the unmanned firefighting equipment terminal (7) and external information (8) are included. The fire scene intelligent map system (1) includes a display module (2), an information management layer (3), an AI assisted decision making layer (4) and an instruction layer (5); the information management layer (3) obtains the longitude and latitude of the fire point and calculates the spread rate according to the external information and the fire scene environment and personnel equipment information returned by the unmanned firefighting equipment terminal and the firefighter terminal, analyzes the fire scene thermal radiation distribution map, analyzes the key fire scene data of the terrain and topography, and constructs the real-time information of the fire scene; the AI assisted decision making layer (4) issues the firefighting tasks and firefighting instructions of the firefighters and the unmanned firefighting equipment, and receives the real-time information of the fire scene output by the information management layer (3).

2. The fire emergency command system based on dynamic fire field modeling and double algorithm cooperation according to claim 1, characterized in that: The AI assisted decision making layer (4) generates an instruction set for the firefighter terminal (6) and the unmanned equipment terminal (7) according to the firefighting strategy, the firefighting instruction and the real-time information of the fire scene.

3. The fire emergency command system based on dynamic fire field modeling and double algorithm cooperation according to claim 1, characterized in that: The display module (2) includes a tile map service (21), an intelligent map system front end (22) and an intelligent map system back end (23); the tile map service (21) and the intelligent map system back end (23) interact through the HTTP protocol; the intelligent map system back end (23) and the intelligent map system front end (22) interact through the HTTP protocol and the WebSocket protocol.

4. The fire emergency command system based on dynamic fire field modeling and double algorithm cooperation of claim 1, characterized in that: The AI assisted decision making layer (4) is composed of a target allocation module (41), a Hungarian algorithm (42), a fire extinguishing planning module (43) and a MADDPG algorithm (44); the target allocation module (41) is driven by the Hungarian algorithm (42) and generates a target fire point priority ranking table; the fire extinguishing planning module (43) is driven by the MADDPG algorithm (44).

5. A fire emergency command method based on dynamic fire field modeling and double algorithm cooperation, characterized in that: The following steps are included: 1) The information management layer tracks the geographic position coordinates of the firefighters through the position information returned by the firefighter terminal, and monitors the physical condition of the firefighters through the risk index R: 2) The information management layer constructs a dynamic fire scene map, determines the number of personnel in the fire area according to the GNSS positioning information in the external information, updates the fire area image according to the visible light image, and updates the intelligent map according to the physical condition of the rescue personnel fed back by the firefighter terminal; 3) The information management layer uploads the fire scene environment information, external information and personnel equipment information to the display module; 4) The information management layer updates the fire scene map and equipment, extracts the longitude and latitude of the fire point, calculates the spread rate and high-risk area marking, and pushes them to the display module; the calculation process of the spread rate is: The level set function φ(x,y,t) is used to implicitly represent the fire line: Γ(t)={(x,y)∈Ω∣φ(x,y,t)=0} Where φ(x,y,t)<0 is the burned area, and φ(x,y,t)>0 is the unburned area; Firewire normal propagation speed: V n = V0+ V w + V s wherein V0 is a base rate; a wind speed driving term V w and a slope driving term V s is: V w = β w ·||W||·|cosθ| θ is the angle between the wind direction and the normal of the fire line; a is the slope angle; β w and β s are the driving term coefficients; The level set evolution equation is discretized by using the upwind difference format: wherein is a first-order upwind difference operator; the instantaneous propagation rate at a fire front point is: Determine the cumulative exposure of thermal radiation and normalize to get the thermal radiation risk factor: Structural risk factor R S consists of static risk S static and dynamic risk S dynamic : R S = 0.3S static + 0.7S dynamic Dynamic risk is expressed as: Wherein, δ is the structural deformation, C gas is the concentration of toxic gas, σ p is the standard deviation of water pressure fluctuations; Escape difficulty risk factor R E is: where C esc is the shortest escape cost expressed as p(s) is a path cost function; 5) The commander issues fire-fighting instructions to the AI-assisted decision-making layer; the target allocation module and the Hungarian algorithm of the AI-assisted decision-making layer generate fire-fighting task planning and target fire point priority ranking table for unmanned devices and firefighters, and deliver them to the instruction layer; the process is as follows: Consider the ignition point target value function as f i and determine the value of the ignition point: f i = aV i + bT i + gP i Wherein, V i is the comprehensive value score of the target, T i is the threat level of the target, P i is the defense capability of the target, and α, β, and γ are weight coefficients. The data set structure is constructed as: Data = {(g1, V1, T1, P1),..., (g N , N , N , N gN, VN, TN, PN} g1, g2,..., g N e1, e2,..., e M as a combat element; The cost calculation formula between the target and the combat element is: where V gi is the overall value of the target g i ; d(g i , e j ) is the distance between the fire location g i and the combat element e j ; d max is the maximum distance between all targets and combat elements. Assume that the fire-fighting resource limit of the participating unit is R j , the Hungarian algorithm is used to maximize the attack on the fire location, the constraint condition is that the resource of each fire-fighting unit is less than the maximum fire-fighting capacity, and the optimization model is: Wherein, r ij is the fire distribution amount of fire fighting unit j to target i, and the optimal distribution of fire resources is carried out by solving the fire distribution weight w ij and the fire resource distribution amount r ij . 6) The AI-assisted decision-making layer drives the fire-fighting planning module to generate a coordinated action path through the MADDPG algorithm, plans the unmanned aerial vehicle group air drop route optimization and the robot cluster breakthrough route, and generates an end-to-end instruction set through data and instruction interaction through the Http protocol and the WebSocket protocol, and delivers it to the instruction layer; 7) The instruction layer (5) issues the travel route and coordinated timing fire-fighting instructions to the unmanned fire-fighting device terminal (7), and the unmanned fire-fighting device terminal (7) controls the unmanned device to extinguish the fire; the instruction layer (5) sends the fire-fighting task to the firefighter terminal (6) and pushes the issued instructions to the display module (2); 8) The firefighter extinguishes the fire according to the fire scene information displayed by the firefighter terminal (6) and the fire-fighting task issued by the instruction layer (5), and the firefighter terminal (6) detects and uploads the firefighter's vital sign data to the information management layer (3); 9) The unmanned fire-fighting device terminal (7) receives the instructions from the instruction layer (5), and the fire-fighting unmanned aerial vehicle collects and returns the fire scene data to the information management layer (3); the fire-fighting unmanned vehicle extinguishes the fire and transmits the fire scene data to the information management layer (3); 10) The information management layer (3) processes the information returned by the unmanned fire-fighting device terminal (7), calculates the spread rate and high-risk area, and uploads the updated real-time fire scene information to the display module (2); 11) The commander judges the fire situation and fire-fighting strategy according to the updated real-time fire scene information of the display module (2), changes or maintains the command to the AI-assisted decision-making layer 4; 12) Steps 9)-11) are executed in a loop until the fire is controlled.

6. The fire emergency command method based on dynamic fire field modeling and double algorithm cooperation according to claim 5, characterized in that: In step 1), the information management layer receives external information source data composed of map information, visible light image information and GNSS positioning information, synchronously acquires personnel state data uploaded by the firefighter terminal and fire scene environment parameters and device operating status feedback by the unmanned fire-fighting device terminal.

7. The fire emergency command method based on dynamic fire field modeling and double algorithm cooperation according to claim 5, characterized in that: In step 1), the information management layer tracks the geographic position coordinates of the firefighters through the position information returned by the firefighter terminal, and monitors the physical condition of the firefighters through the risk index R: R = a · HR + β · SpO + γ · T norm + β · SpO 2norm + γ · T norm + λ1· HR D + λ2· SpO 2D + λ1· T D wherein HR norm represents heart rate, T norm represents body temperature, SpO2represents blood oxygen saturation, R is a risk index, HR D represents heart rate, T D represents body temperature, SpO 2D represents a rate of change of blood oxygen saturation: t1 is n times the system refresh frequency; α, β, γ, λ1, λ2, λ3 are the influence weights of heart rate, body temperature, blood oxygen saturation and their change rates.

8. The fire emergency command method based on dynamic fire field modeling and double algorithm cooperation according to claim 5, characterized in that: In step 6), the MADDPG algorithm is: the state space of each firefighting unit includes the two-dimensional coordinate position (x t ,y t ) of the firefighting unit at time step t, the speed v t , and the acceleration a t , expressed as a vector state t : state t =[x t ,y t ,v t ,a t ] The action space of each fire unit consists of a displacement (Δx t ,Δy t ) and a velocity change Δv t in the horizontal and vertical directions at time step t, denoted as the vector action t : action t =[Δx t ,Δy t ,Δv t ] The status update of the fire fighting unit is as follows: Δt is the time step, (x t ,y t ) is the current position of the fire fighting unit, v t is the velocity, a t is the acceleration.

9. The fire emergency command method based on dynamic fire field modeling and double algorithm cooperation according to claim 5, characterized in that: In step 9), the fire-fighting unmanned vehicle starts the fire-fighting equipment and uploads the device data including water pressure, range, fire scene monitoring image and fire scene temperature to the information management layer (3).

10. The fire emergency command method based on dynamic fire field modeling and double algorithm cooperation according to claim 5, characterized in that: In step 9), the fire-fighting unmanned vehicle executes high-altitude fire-fighting agent delivery and high-altitude collection tasks, and real-time returns fire scene images, fire scene temperature, fire scene mutations and device data to the information management layer (3).

Images