Fire emergency evacuation indication method applied to large indoor scene and related device

Abstract

The application discloses a fire emergency evacuation indication method applied to a large indoor scene and a related device, relates to the field of artificial intelligence, and obtains a personnel evacuation network flow model, a density matrix, a speed matrix and an evacuation time consumption matrix. The above parameters are input into a pre-constructed evacuation path prediction model, and a first escape path corresponding to each node is obtained through the evacuation path prediction model. Personnel can escape based on the first escape path; as time elapses, the fire detailed information of the nodes not passed through by the personnel in the first escape path can change, so that the first escape path is no longer the most suitable escape path, at which time the updated above parameters are obtained and input into the evacuation path prediction model, a second escape path is obtained through the evacuation path prediction model, and thus the personnel can escape through the second escape path.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a fire emergency evacuation indication method applied to a large indoor scene and a related device. BACKGROUND

[0002] Compared with other disaster accidents, fire has the characteristics of high occurrence frequency, strong disaster spreading and smoke hazard, etc., which has caused great loss to people's life and property. Large indoor scenes, such as large shopping malls, bank business outlets, especially bank business outlets with vaults, business halls, and clerk office back-end areas, have the characteristics of more flammable materials, crowded personnel, and complex building structures, etc., which belong to high-risk fire sites. Once a fire occurs, personnel may choose an unreasonable escape route due to panic, delay evacuation time, and cause serious loss of life and property.

[0003] Therefore, an emergency evacuation indication method suitable for large indoor scenes is needed to guide employees and customers in large indoor scenes to escape safely and efficiently, so as to minimize the loss of life and property. SUMMARY

[0004] In view of the above problems, the present application provides a fire emergency evacuation indication method applied to a large indoor scene and a related device to automatically generate an escape route. The specific scheme is as follows:

[0005] The first aspect of the present application provides a fire emergency evacuation indication method applied to a large indoor scene, comprising:

[0006] Obtaining a personnel evacuation network flow model, the personnel evacuation network flow model comprising a plurality of nodes and edges between two nodes; the edges between two nodes represent the passageways between two nodes for personnel walking; the plurality of nodes comprise an escape exit node, an ordinary node, and a staircase node;

[0007] Obtaining a distance matrix corresponding to the personnel evacuation network flow model; the distance matrix comprises the distances between all pairs of nodes with edges in the personnel evacuation network flow model;

[0008] If a fire is detected, obtaining a density matrix corresponding to the personnel evacuation network flow model; the density matrix comprises the person flow density of all nodes in the personnel evacuation network flow model and the person flow density of the passageways between two nodes with edges;

[0009] Determining a dangerous node and a disaster node from the plurality of nodes; the dangerous node is a node with a person flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs;

[0010] if the flow density of the path between the two nodes represented by the edge is greater than or equal to a second preset density value, determining that the edge between the two nodes is a congested road section;

[0011] updating the dangerous node, the disaster node, and the congested road section to the personnel evacuation network flow model;

[0012] based on the density matrix, obtaining a speed matrix corresponding to the personnel evacuation network flow model; the speed matrix includes the moving speed of personnel in the path between all pairs of nodes with edges in the personnel evacuation network flow model;

[0013] based on the distance matrix and the speed matrix, obtaining an evacuation time consumption matrix corresponding to the personnel evacuation network flow model;

[0014] inputting the personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix, and the evacuation time consumption matrix into a pre-constructed evacuation path prediction model, and obtaining a first escape path corresponding to each node through the evacuation path prediction model;

[0015] for each node, sending the first escape path corresponding to the node to a terminal corresponding to the node for escape;

[0016] if it is detected that the flow density of any node changes, or the flow density of the path between any two nodes changes, or the moving speed of personnel in the path between any two nodes changes, or the fire condition of any node changes, obtaining an updated personnel evacuation network flow model, an updated density matrix, an updated speed matrix, and an updated evacuation time consumption matrix;

[0017] inputting the updated personnel evacuation network flow model, the distance matrix, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix into the evacuation path prediction model, and obtaining a second escape path through the evacuation path prediction model;

[0018] for each node, sending the second escape path corresponding to the node to a terminal corresponding to the node for escape.

[0019] In an optional implementation, the method further includes:

[0020] for each node, detecting whether the node changes from a disaster occurrence location to a non-disaster occurrence location or changes from a non-disaster occurrence location to a disaster occurrence location through a smoke sensor deployed at the node, and if so, determining that the fire condition of the node changes;

[0021] For each of the nodes, a sensor deployed at the node detects a crowd density located at the node, and if the crowd density changes from being greater than or equal to the first preset density value to being less than the first preset density value, or from being less than the first preset density value to being greater than or equal to the first preset density value, it is determined that the crowd density of the node has changed;

[0022] For each edge between two nodes, a sensor deployed on a path between the two nodes detects a crowd density located on the path, and if the crowd density changes from being greater than or equal to the second preset density value to being less than the second preset density value, or from being less than the second preset density value to being greater than or equal to the second preset density value, it is determined that the crowd density of the path has changed.

[0023] For each edge between two nodes, a smoke sensor deployed on a path between the two nodes detects a smoke concentration of the path, and if a smoke concentration level corresponding to the smoke concentration changes, it is determined that a moving speed of a person in the path between the two nodes has changed.

[0024] In an optional implementation, the smoke concentration is a CO concentration, and for each edge between two nodes, the method of obtaining an updated moving speed of a path between the two nodes includes:

[0025] a coefficient a is calculated by a formula co is the CO concentration of the path between the two nodes;

[0026] and the updated moving speed is a product of the moving speed before updating and the coefficient a.

[0027] In an optional implementation, the evacuation path prediction model is a double deep Q network model, and the method of training the double deep Q network includes:

[0028] obtaining a sample density matrix corresponding to the personnel evacuation network flow model; the sample density matrix includes crowd densities of all nodes in the personnel evacuation network flow model and crowd densities of paths between two nodes with edges;

[0029] based on the sample density matrix, obtaining a sample speed matrix corresponding to the personnel evacuation network flow model; the sample speed matrix includes moving speeds of persons in paths between all two nodes with edges in the personnel evacuation network flow model;

[0030] based on the sample distance matrix and the sample speed matrix, obtaining a sample evacuation time consumption matrix corresponding to the personnel evacuation network flow model; ​

[0031] determining a dangerous node and a disaster node from the plurality of nodes; the dangerous node being a node with a people flow density greater than or equal to a first preset density value, and the disaster node being a node where a fire occurs;

[0032] if a people flow density of a path represented by an edge between two nodes is greater than or equal to a second preset density value, determining the edge between the two nodes as a congested path;

[0033] updating the dangerous node, the disaster node, and the congested path to the personnel evacuation network flow model;

[0034] obtaining state information of an agent, the state information including time consumption of moving from the first node to each of the second nodes respectively;

[0035] inputting the personnel evacuation network flow model, the distance matrix, the sample density matrix, the sample speed matrix, the sample evacuation time consumption matrix, and the state information into the double deep Q network model, and outputting predicted Q values corresponding to a plurality of actions respectively through the double deep Q model; the double deep Q network model includes a deep Q network model and a target Q network model; the action is an action of moving the agent from the first node to the second node;

[0036] obtaining a random number;

[0037] if the random number is greater than a preset threshold, randomly selecting a target action from the plurality of actions, the second node in the target action being a target second node;

[0038] if the random number is less than or equal to the preset threshold, selecting a target action corresponding to the maximum expected Q value from the plurality of actions; the second node in the target action being a target second node;

[0039] determining a first reward of moving the agent to the target second node based on a shortest distance between the target second node and the escape exit node, whether the target second node is the disaster node, whether the target second node is the dangerous node, whether the target second node is the escape exit node, and whether a target time consumption of the agent from an initial node to the target second node exceeds a preset specified time length;

[0040] inputting the first reward and the predicted Q value corresponding to the target action into the double deep Q network model to update the deep Q network model in the double deep Q network model;

[0041] if the target second node is not the disaster node and is not the escape exit node, and the target time consumption does not exceed the preset specified duration, returning to the step of acquiring the state information of the agent;

[0042] if the target second node is the disaster node or the target time consumption exceeds the preset specified duration, determining that the agent returns to the initial node, and returning to the step of acquiring the state information of the agent;

[0043] if the target second node is the escape exit node, acquiring an escape planning path, the escape planning path being a path of the agent from the initial node to the escape exit node;

[0044] determining a second reward based on a number of disaster nodes in the escape planning path, a number of dangerous nodes in the escape planning path, a number of congested road segments in the escape planning path, and a length of the escape planning path;

[0045] inputting the second reward and a predicted Q value corresponding to the target action into the deep Q network model to update the deep Q network model;

[0046] acquiring a target real Q value of an action of moving from a previous node to a next node in each pair of nodes with edges in the escape planning path;

[0047] if the target real Q values corresponding to each pair of nodes with edges in the escape planning path in an escape order are sequentially increased, determining that the escape planning path is an optimal path;

[0048] if the target real Q values corresponding to each pair of nodes with edges in the escape planning path in the escape order are not sequentially increased, determining that the agent returns to the initial node, and returning to the step of acquiring the state information of the agent;

[0049] if a number of updates of the deep Q network model reaches a preset number of updates, assigning parameters of the deep Q network model to the target Q network model;

[0050] if all optimal paths corresponding to all nodes have been acquired, determining that the training of the dual deep Q network model is completed.

[0051] In an optional implementation, the step of determining the first reward of the agent moving to the target second node based on the shortest distance between the target second node and the escape exit node, whether the target second node is the disaster node, whether the target second node is the dangerous node, whether the target second node is the escape exit node, and whether target time consumption of the agent from the initial node to the target second node exceeds a preset specified duration comprises:

[0052] If the target second node is not the disaster node and is not the escape exit node, and the target time consumption is less than the preset specified duration, R is calculated by formula t1 ; R t1 is determined as the first reward; k1 is a preset reward amplification coefficient, k1 is a positive number greater than 0, refers to the shortest distance; refers to the shortest distance between the first node and the escape exit;

[0053] If the target second node is not the escape exit node and the target time consumption is greater than the preset specified duration, or the target second node is the disaster node, a second preset reward value is determined as the first reward.

[0054] In an optional implementation, the second reward step is determined based on the number of disaster nodes in the escape planning path, the number of dangerous nodes in the escape planning path, the number of congested road segments in the escape planning path, and the length of the escape planning path, and includes:

[0055] If the target second node is the escape exit node and the target time consumption is less than or equal to the preset specified duration, R is calculated by formula -1 ; R -1 is determined as the second reward.

[0056] Wherein, k2 is a preset scaling factor, k3 is a preset time coefficient, and t2 is the total time consumption of the intelligent agent from the initial node to the escape exit node.

[0057] Wherein, , , wherein is the three-dimensional coordinates of the initial node; is the three-dimensional coordinates of the escape exit node closest to the initial node; is the three-dimensional coordinates of the jth node in the escape planning path; N is the total number of all nodes in the escape planning path. , wherein, refers to the number of dangerous nodes in the escape planning path; refers to the number of disaster nodes in the escape planning path;

[0058] , wherein, , refers to the three-dimensional coordinates of the starting node of the ith congested road segment; is a three-dimensional coordinate of the terminal node of the ith congested road segment.

[0059] In an optional implementation, the target reward is the first reward or the second reward, and the target reward and a predicted Q value corresponding to the target action are input into the dual deep Q network model to update the deep Q network model in the dual deep Q network model.

[0060] If the target second node is not the disaster node and is not the escape exit node and the target time consumption is less than the preset specified duration, a real Q value of the agent moving from the first node to the target second node is calculated by a formula

[0061] wherein, is the real Q value, is a predicted Q value of the agent moving from the first node to the target second node output by the dual Q network model; lr is a preset learning rate parameter, and γ is a preset discount factor, is a preset function.

[0062] If the target second node is the escape exit node and the target time consumption is less than or equal to the preset specified duration, it is determined that R -1 is the real Q value of the agent moving from the first node to the target second node.

[0063] If the target second node is not the escape exit node and the target time consumption is greater than the preset specified duration, or the target second node is the disaster node, it is determined that the second preset reward value is the real Q value of the agent moving from the first node to the target second node.

[0064] The deep Q network model is updated by a formula

[0065] In an optional implementation, the deep Q network model comprises a first CSP module, a first CBS module, a second CSP module, a second CBS module, a third CSP module and a third CBS module connected in sequence.

[0066] ​​The first CSP module comprises a module carrying an unfold function, a fourth CBS module connected with a first output end of the module carrying the unfold function, a first BottleNeck module connected with a second output end of the module carrying the unfold function, and a second BottleNeck module connected with a third output end of the module carrying the unfold function; a fifth CBS module connected with an output end of the fourth CBS module, an output end of the first BottleNeck module, and an output end of the second BottleNeck module, respectively; and a sixth CBS module connected with the fifth CBS module.

[0067] The first CBS module comprises a convolution layer, a batch normalization layer connected with the convolution layer, and an activation function layer connected with the batch normalization layer.

[0068] The first CSP module, the second CSP module, and the third CSP module have the same structure; and the first CBS module, the second CBS module, the third CBS module, the fourth CBS module, the fifth CBS module, and the sixth CBS module have the same structure.

[0069] The second aspect of the application provides a fire emergency evacuation indication device applied to a large indoor scene, comprising:

[0070] A first obtaining module is configured to obtain a personnel evacuation network flow model, wherein the personnel evacuation network flow model comprises a plurality of nodes and edges between two nodes; the edges between two nodes represent a passage between the two nodes for personnel walking; and the plurality of nodes comprise an escape exit node, a normal node, and a staircase node.

[0071] A second obtaining module is configured to obtain a distance matrix corresponding to the personnel evacuation network flow model; and the distance matrix comprises distances between all nodes with edges in the personnel evacuation network flow model.

[0072] A third obtaining module is configured to obtain a density matrix corresponding to the personnel evacuation network flow model if a fire is detected; and the density matrix comprises a person flow density of all nodes in the personnel evacuation network flow model and a person flow density of a passage between two nodes with edges.

[0073] A first determining module is configured to determine a dangerous node and a disaster node from the plurality of nodes; the dangerous node is a node with a person flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs.

[0074] a second determining module, configured to determine the edge between the two nodes as a congested road segment if the flow density of the path represented by the edge between the two nodes is greater than or equal to a second preset density value;

[0075] a first updating module, configured to update the dangerous node, the disaster node and the congested road segment to the personnel evacuation network flow model;

[0076] a fourth obtaining module, configured to obtain a speed matrix corresponding to the personnel evacuation network flow model based on the density matrix, wherein the speed matrix comprises the moving speed of personnel in the path between all pairs of nodes with edges in the personnel evacuation network flow model;

[0077] a fifth obtaining module, configured to obtain an evacuation time consumption matrix corresponding to the personnel evacuation network flow model based on the distance matrix and the speed matrix;

[0078] a sixth obtaining module, configured to input the personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix and the evacuation time consumption matrix into a pre-constructed evacuation path prediction model, and obtain a first escape path corresponding to each node by using the evacuation path prediction model;

[0079] a first sending module, configured to send the first escape path corresponding to each node to a terminal corresponding to the node for escape;

[0080] a seventh obtaining module, configured to obtain an updated personnel evacuation network flow model, an updated density matrix, an updated speed matrix and an updated evacuation time consumption matrix if any of the following conditions is met: the flow density of any node changes, the flow density of the path between any pair of nodes changes, the moving speed of personnel in the path between any pair of nodes changes, or the fire condition of any node changes;

[0081] an eighth obtaining module, configured to input the updated personnel evacuation network flow model, the distance matrix, the updated density matrix, the updated speed matrix and the updated evacuation time consumption matrix into the evacuation path prediction model, and obtain a second escape path by using the evacuation path prediction model;

[0082] a second sending module, configured to send the second escape path corresponding to each node to a terminal corresponding to the node for escape.

[0083] The third aspect of the present application provides a computer program product, comprising computer readable instructions, which, when executed on an electronic device, cause the electronic device to implement the fire emergency evacuation indication method for large indoor scenes of the first aspect or any implementation manner of the first aspect.

[0084] The fourth aspect of the present application provides an electronic device, comprising at least one processor and a memory connected with the processor, wherein:

[0085] The memory is configured to store a computer program;

[0086] The processor is configured to execute the computer program, so that the electronic device can implement the fire emergency evacuation indication method for large indoor scenes of the first aspect or any implementation manner of the first aspect.

[0087] The fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs, when the one or more computer programs are executed by an electronic device, can cause the electronic device to implement the fire emergency evacuation indication method for large indoor scenes of the first aspect or any implementation manner of the first aspect.

[0088] By the above technical solution, the fire emergency evacuation indication method applied to the large indoor scene is provided, the space characteristics of the actual evacuation scene can be restored through the personnel evacuation network flow model; the evacuation path of the actual evacuation scene can be restored through the distance matrix and the personnel evacuation network flow model; if the fire is detected, which nodes have larger traffic density and which paths between two nodes have larger traffic density can be judged through the density matrix, so that which nodes are the dangerous nodes, which nodes are the disaster nodes, and which paths are the congested sections can be determined; the speed matrix is obtained based on the density matrix, and it can be known through the speed matrix that the moving speed of the personnel is faster in which path and the moving speed of the personnel is slower in which path in the evacuation process. The evacuation time consumption matrix is obtained based on the distance matrix and the speed matrix; it can be known through the evacuation time consumption matrix that the personnel takes the shortest time to escape from which escape exit. The personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix and the evacuation time consumption matrix are input into the pre-constructed evacuation path prediction model, and the first escape path corresponding to all nodes is obtained through the evacuation path prediction model. The personnel can escape based on the first escape path; as time elapses, the fire detailed information of the nodes not passed through by the personnel in the first escape path can change, resulting in that the first escape path is not the most suitable escape path, at this time, the updated personnel evacuation network flow model, the updated density matrix, the updated speed matrix and the updated evacuation time consumption matrix need to be obtained; the personnel evacuation network flow model, the distance matrix, the updated density matrix, the updated speed matrix and the updated evacuation time consumption matrix are input into the evacuation path prediction model, and the second escape path is obtained through the evacuation path prediction model; in this way, the personnel can escape through the second escape path. BRIEF DESCRIPTION OF DRAWINGS

[0089] The above and other features, advantages, and aspects of the present disclosure will become more apparent by referring to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, like or similar reference numerals are used to refer to like or similar elements throughout the various figures. It should be understood that the drawings are diagrammatic and schematic representation of elements and features do not necessarily depict the actual scale or proportions.

[0090] Figure 1 A system architecture schematic diagram is provided for the present application;

[0091] Figure 2 An optional hardware structure schematic diagram of a terminal 100 is provided for the present application;

[0092] Figure 3 A structure schematic diagram of an electronic device 200 is provided for the present application;

[0093] Figure 4 A flow schematic diagram of a fire emergency evacuation indication method applied to a large indoor scene is provided for the embodiment of the present application;

[0094] Figure 5 A schematic diagram of a construction process of a personnel evacuation network flow model provided for an embodiment of the present application;

[0095] Figure 6 A schematic diagram of a personnel evacuation network flow model of a bank business site provided for an embodiment of the present application;

[0096] Figure 7 A schematic diagram of part of a distance matrix L of a second layer of a bank business site provided for an embodiment of the present application;

[0097] Figure 8 A schematic diagram of a personnel evacuation network flow model updated through step S406 provided for an embodiment of the present application;

[0098] Figure 9 A comparative schematic diagram of a static evacuation path and a dynamic evacuation path provided for an embodiment of the present application;

[0099] Figure 10 A structural schematic diagram of a deep Q network model provided for an embodiment of the present application;

[0100] Figure 11 A structural diagram of a first CBS module provided for an embodiment of the present application;

[0101] Figure 12 A schematic diagram of an interaction process between an agent and an environment of a bank business site provided for an embodiment of the present application;

[0102] Figure 13 A structural schematic diagram of a fire emergency evacuation indication device applied to a large indoor scene provided for an embodiment of the present application;

[0103] Figure 14 A schematic diagram of a fire emergency evacuation indication system applied to a large indoor scene provided for an embodiment of the present application. DETAILED DESCRIPTION

[0104] The embodiments of the present application are described below in conjunction with the accompanying drawings. The terms used in the embodiment part of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0105] The embodiments of the present application are described below in conjunction with the accompanying drawings. It is known to those skilled in the art that, as technology develops and new scenarios appear, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0106] The terms "first", "second", and the like in the description and in the claims of the present application and above-described drawings are used to distinguish similar objects and are not necessarily used to describe a specific sequential or chronological order. It should be understood that the terms so used are interchangeable under appropriate circumstances and are merely employed to distinguish one embodiment of the application from another. Furthermore, the terms "comprise", "have", and any variations thereof are intended to cover a non-exclusive inclusion, such that processes, methods, systems, products, or devices that comprise a list of elements are not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such processes, methods, systems, products, or devices.

[0107] Referring to Figure 1 , Figure 1 A system architecture diagram is shown. The system can include a plurality of terminals 100, an electronic device 200, and a plurality of sensors 300.

[0108] The plurality of sensors 300 can be deployed at corresponding positions in a large indoor scene, for example, respectively deployed in the lobby, customer waiting area, and teller customer service area of a bank business site, where employees and customers may exist.

[0109] For example, the plurality of sensors 300 include but are not limited to a laser radar, a camera, a smoke sensor, and a temperature sensor. It can be understood that if the smoke sensor or the temperature sensor is deployed in the customer waiting area, the smoke sensor or the temperature sensor can detect whether a fire occurs in the customer waiting area.

[0110] For example, if the camera is deployed in the customer waiting area, the image captured by the camera can be used to estimate the crowd density in the customer waiting area when a fire occurs.

[0111] For example, when a fire occurs, the smoke concentration may be high, and the image captured by the camera cannot obtain the crowd density. At this time, the point cloud data collected by the laser radar can be used to obtain the crowd density in the customer waiting area.

[0112] For example, the deployment of the plurality of sensors at various positions can be determined according to specific conditions, which is not limited here.

[0113] It can be understood that the information collected by the plurality of sensors 300 can be uploaded to the electronic device 200, so that the electronic device 200 can execute the fire emergency evacuation indication method applied to a large indoor scene provided by the embodiments of the present application to obtain the first escape path or the second escape path.

[0114] It can be understood that a plurality of terminals 100 can be deployed in an area capable of carrying people, and on a passageway for people to walk, the electronic device 200 sends the first escape path and the second escape path to the terminal 100. Thus prompting people how to escape.

[0115] For example, the terminal 100 can be a mobile phone, a tablet computer, a wearable device, a vehicle-mounted device, an augmented reality (AR) / virtual reality (VR) device, a notebook computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), etc. The embodiments of the present application do not make any limitation on this.

[0116] Next, the product form of the terminal 100 is described. Figure 1

[0117] Figure 2 An optional hardware structure schematic diagram of the terminal 100 is shown.

[0118] Referring to Figure 2 As shown, the terminal 100 can include a radio frequency unit 110, a memory 120, an input unit 130, a display unit 140, a camera 150 (optional), an audio circuit 160 (optional), a speaker 161 (optional), a microphone 162 (optional), an earphone jack 163 (optional), a processor 170, an external interface 180, a power supply 190, etc. Those skilled in the art can understand that the terminal 100 can include more or less components than those shown in the figure, or combine certain components, or different components. Figure 2 The above is only an example of the terminal, and does not constitute a limitation on the terminal, and the terminal can include more or less components than those shown in the figure, or combine certain components, or different components.

[0119] ​The input unit 130 can be used to receive inputted digital or character information, and to generate key signal inputs related to the user settings of the terminal and the function control. Specifically, the input unit 130 can include a touch screen 131 (optional) and / or other input devices 132. The touch screen 131 can collect touch operations of a user thereon or proximate thereto (such as operations of the user using a finger, a knuckle, a stylus, or any suitable object on or proximate to the touch screen), and drive corresponding connection devices according to a pre-set program. The touch screen can detect touch actions of a user on the touch screen, convert the touch actions into touch signals and send the touch signals to the processor 170, and can receive commands from the processor 170 and execute the commands; the touch signals at least include touch point coordinate information. The touch screen 131 can provide an input interface and an output interface between the terminal 100 and the user. In addition, the touch screen can be implemented in various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch screen 131, the input unit 130 can also include other input devices. Specifically, the other input devices 132 can include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control keys, on-off keys, etc.), trackballs, mice, joysticks, etc.

[0120] The input device 132 can receive inputted data, etc.

[0121] The display unit 140 can be used to display information inputted by a user or provided to the user, various menus of the terminal 100, interactive interfaces, file display, and / or playing of any kind of multimedia files. In the embodiments of the present application, the display unit 140 can be used to display interfaces, processing results, etc.

[0122] The storage 120 can be used to store instructions and data. The storage 120 can mainly include a storage instruction area and a storage data area. The storage data area can store various data such as multimedia files, texts, etc.; the storage instruction area can store software units such as operating systems, applications, instructions required by at least one function, etc., or their subsets, expanded sets. It can also include a non-volatile random access memory; to provide the processor 170 with software and applications that include managing hardware, software, and data resources in a computing processing device, supporting control. It is also used for the storage of multimedia files, and the storage of running programs and applications.

[0123] The processor 170 is the control center of the terminal 100, connects each part of the whole terminal 100 by various interfaces and lines, executes various functions of the terminal 100 and processes data by running or executing instructions stored in the memory 120 and calling data stored in the memory 120, thereby performing overall control on the terminal device. Optionally, the processor 170 can include one or more processing units; preferably, the processor 170 can integrate an application processor and a modem processor, wherein the application processor mainly processes operating systems, user interfaces, and application programs, and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 170. In some embodiments, the processor, the memory, and the like can be implemented on a single chip, and in some embodiments, they can also be implemented on independent chips respectively. The processor 170 can also be used to generate corresponding operation control signals to the corresponding components of the computing processing device, read and process data in the software, especially read and process data and programs in the memory 120, so that each functional module in it performs corresponding functions, thereby controlling the corresponding components to act according to the requirements of the instructions.

[0124] The radio frequency unit 110 (optional) can be used for receiving and sending signals in the process of receiving and sending information or calls, for example, receiving the downlink information of the base station and processing it by the processor 170; in addition, sending the uplink data to the base station. Generally, the RF circuit includes but is not limited to an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (LNA), a duplexer, and the like. In addition, the radio frequency unit 110 can also communicate with network devices and other devices through wireless communication. The wireless communication can use any communication standard or protocol, including but not limited to the global system for mobile communication (GSM), general packet radio service (GPRS), code division multiple access (CDMA), wideband code division multiple access (WCDMA), long term evolution (LTE), email, short message service (SMS), and the like.

[0125] In the embodiments of the present application, the radio frequency unit 110 can send data to the electronic device 200 and receive the processing result sent by the electronic device 200.

[0126] It should be understood that the radio frequency unit 110 is optional, which can be replaced by other communication interfaces, for example, can be a network interface.

[0127] The terminal 100 further includes a power supply 190 (such as a battery) for supplying power to various components. Preferably, the power supply can be logically connected to the processor 170 through a power management system, so that the power management system can realize functions such as management of charging, discharging, and power consumption management.

[0128] The terminal 100 further includes an external interface 180, which can be a standard Micro USB interface or a multi-pin connector, and can be used to connect the terminal 100 to other devices for communication or to connect a charger to charge the terminal 100.

[0129] Although not shown, the terminal 100 can further include a flash, a wireless fidelity (WiFi) module, a Bluetooth module, sensors of different functions, and the like, which will not be described here. Some or all of the methods described below can be applied to the terminal 100 as shown. Figure 2

[0130] Next, the product form of the electronic device 200 is described. Figure 1

[0131] For example, the electronic device 200 can be an edge computing box or a server deployed in a large indoor scene.

[0132] Figure 3 A structural schematic diagram of the electronic device 200 is provided, as shown in the figure, the electronic device 200 includes a bus 201, a processor 202, a communication interface 203, and a memory 204. The processor 202, the memory 204, and the communication interface 203 communicate through the bus 201. Figure 3

[0133] The bus 201 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus. Figure 3

[0134] ​​​​The processor 202 can be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, a microcontroller, a digital signal processor (DSP), or the like.

[0135] The memory 204 can include volatile memory, such as random access memory (RAM), and non-volatile memory, such as read-only memory (ROM), a floppy disk drive, a hard disk drive, or a solid state drive.

[0136] The memory 204 can be configured to store software code related to the method for indicating fire emergency evacuation in a large indoor scene, and the processor 202 can execute the steps of the method for indicating fire emergency evacuation in a large indoor scene, or can schedule other units to implement corresponding functions.

[0137] It should be understood that the terminal 100 and the electronic device 200 described above can be centralized or distributed devices, and the processors (e.g., the processor 170 and the processor 202) in the terminal 100 and the electronic device 200 can be hardware circuits (such as an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a general-purpose processor, a digital signal processor (DSP), a microprocessor, or a microcontroller, etc.), or a combination of these hardware circuits. For example, the processor can be a hardware system with an execution instruction function, such as a CPU, a DSP, etc., or a hardware system without an execution instruction function, such as an ASIC, an FPGA, etc., or a combination of the hardware system without an execution instruction function and the hardware system with an execution instruction function.

[0138] Each year, tens of thousands of people are killed or injured in fires. Large indoor scenes are numerous, widely distributed, complex environment, large flow of people, in the event of a disaster, if there is an efficient evacuation strategy can guide personnel evacuation to the safe area, can minimize the loss and casualties caused by the accident. Planning a reasonable escape planning path is a crucial step in these measures, and the traditional path planning algorithm only considers the path length as the only evaluation index, which may lead people to the fire area, which may cause secondary injury to people.

[0139] The number of people in a large indoor scene is large, which can easily cause the flow density on the passage to be too large, causing congestion or stagnation, and in severe cases, stampede accidents may occur. In the process of evacuating personnel, environmental factors are complex and change frequently, such as a fire suddenly occurring in an area that has not previously had a fire, or the flow density suddenly increasing in an area that previously had a small flow density. In related technologies, emergency evacuation path planning is static evacuation path. If a static evacuation path (assuming A area → B area → C area) has been planned, and the environmental factors change, then a region in the static evacuation path may not be suitable for personnel escape, for example, a fire occurs in the B area. If the static evacuation path is not updated in time, personnel may be injured again after escaping from the A area to the B area.

[0140] To solve the above problems, the embodiments of the present application provide a fire emergency evacuation indication method applied to a large indoor scene. The fire emergency evacuation indication method applied to a large indoor scene of the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0141] Reference Figure 4 , Figure 4 The flowchart of the fire emergency evacuation indication method applied to a large indoor scene provided by the embodiments of the present application is shown in Figure 4 The fire emergency evacuation indication method applied to a large indoor scene provided by the embodiments of the present application can include steps S401 to S413, which will be described in detail below.

[0142] Step S401: Obtain a personnel evacuation network flow model.

[0143] The personnel evacuation network flow model includes a plurality of nodes and edges between two nodes; the edges between two nodes represent the passage between two nodes for personnel walking; the plurality of nodes include an escape exit node, a normal node, and a staircase node.

[0144] The construction process of the personnel evacuation network flow model will be described below.

[0145] As Figure 5The diagram shown illustrates the construction process of the personnel evacuation network flow model provided in this embodiment of the application.

[0146] Assumption Figure 5 The left side shows a top view of the various objects on a floor. For example, the various objects (i.e., the gray-filled areas) include, but are not limited to: counters, physical stores, etc. Figure 5 The blank areas shown on the left are all passageways that people can pass through; Figure 5 The gray-filled area shown on the left is impassable for people.

[0147] Passageways can be abstracted as edges. Forking points with multiple directions can be abstracted as nodes. For example, the flow model of a people evacuation network can be characterized by the following formula.

[0148] ; ; Where V represents a set of nodes, such as set V containing n nodes, namely v1, v2, ..., v n .like Figure 5 As shown on the right, n=12, meaning n is the number of nodes; E represents the set of edges, such as those connecting nodes v. a and node v b The line segment is called the side. For example... Figure 5 As shown on the right, set E includes 17 edges.

[0149] Understandably, the personnel evacuation network flow model recreates the spatial characteristics of the actual evacuation scenario using nodes and edges. It is suitable for scenarios with large evacuation areas and a large number of people to be evacuated.

[0150] To help those skilled in the art better understand the significance of the personnel evacuation network flow model, examples are provided below.

[0151] like Figure 6 The diagram shown is a schematic of the personnel evacuation network flow model of a bank branch provided in an embodiment of this application.

[0152] For example, the bank branch includes: office area, basement, and vault. The bank branch has two floors.

[0153] Each floor of the bank branch is 42m long and 14m wide, with four fire escape staircases on each floor. This application considers all personnel reaching the three emergency exits on the first floor of the bank branch as a safe escape. Therefore, the process of constructing the personnel evacuation network flow model for the bank branch is as follows:

[0154] Step 1: Designate the lobby, customer waiting area, and teller waiting area of ​​the bank branch, excluding restricted areas such as elevator shafts and ventilation system control rooms, as nodes.

[0155] Second step: set the stairs connecting the upper and lower two floors of the bank business site as the stair nodes.

[0156] Third step: set the intersection of each layer passage of the bank business site (i.e. the branch) as the node.

[0157] Fourth step: set the road, corridor, passageway and other passages that can be passed through in the bank business site as the edges.

[0158] It can be understood that the above nodes and edges refer to areas with certain space in the bank business site, rather than just a point or an edge in the geometric sense. In order to facilitate calculation and research, the geometric center position of the intersection of the passage (i.e. the branch) is defined as the node; the geometric center position of the business hall, customer waiting area and counter is defined as the node. According to the above four steps, the personnel evacuation network flow model shown in Figure 6

[0159] Figure 6 The escape exit nodes 5, 21 and 36 are represented by white circular nodes filled with white color. Figure 6 The stair nodes are represented by gray rectangular nodes filled with gray color. Figure 6 The ordinary nodes are represented by gray circular nodes filled with gray color.

[0160] Step S402: obtain the distance matrix corresponding to the personnel evacuation network flow model.

[0161] The distance matrix includes the distance between each two nodes with edges in the personnel evacuation network flow model.

[0162] Still taking the personnel evacuation network flow model shown in Figure 6 as an example, Figure 6 there are 68 nodes in total. In order to more accurately describe the distance relationship between these nodes, the distance matrix L is introduced. The distance matrix is a 68x68 matrix. Since the node number starts from 0, the element L(i,j) in the i-th row and j-th column of the matrix represents the actual distance of the passage between node i-1 and node j-1. If node i and node j are not connected, the corresponding element value is 0.

[0163] As shown in Figure 7 , it is a schematic diagram of part of the distance matrix L of the second floor of the bank business site provided by the embodiment of the present application. In the actual bank business site evacuation process, each passage allows two-way evacuation of personnel, so the distance matrix L is a symmetric matrix.

[0164] ​Step S403: If a fire is detected, a density matrix corresponding to the personnel evacuation network flow model is obtained.

[0165] The density matrix includes the flow density of each node in the personnel evacuation network flow model and the flow density between each two nodes with an edge.

[0166] The process of obtaining the density matrix is described below.

[0167] Density matrix D t1 represents the flow density of each node and each edge at t1; if the density matrix is in the form of Figure 6 The personnel evacuation network flow model shown in the figure is taken as an example, the density matrix D t1 is also a 68x68 matrix, and the element value unit is person / m 2 . The element D t1 (i, j) in the i-th row and j-th column of the matrix represents the flow density between node i-1 and node j-1 at t1. When i=j, it represents the flow density at node i-1. If node i and node j are not directly connected, the corresponding element value of the density matrix is set to 99. If D t1 (i, j) is greater than 4, it means that the corresponding path is not passable. Since the flow density in each path of the bank business network is different at different time points, the elements in the density matrix will change over time. The calculation method of the flow density of the element in the i+1-th row and j+1-th column of the density matrix at t1 is as follows.

[0168] , , .

[0169] wherein, represents the flow density of path r ij from node i to node j at t1, represents the flow density of node i at t1. and represent the number of people in path r ij and node i at t1, which can be calculated by counting the current position information of each person through the program. and respectively represent the relative area of path r ij and node i.

[0170] Step S404: Determine the dangerous node and the disaster node from the plurality of nodes.

[0171] The dangerous node is a node with a personnel flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs.

[0172] Exemplarily, whether a fire occurs can be detected by a smoke sensor or a temperature sensor.

[0173] Exemplarily, the first preset density value can be based on actual conditions, and embodiments of the present application do not make a limitation in comparison.

[0174] Step S405: If the personnel flow density of a path represented by an edge between two nodes is greater than or equal to a second preset density value, the edge between the two nodes is determined as a congested road section.

[0175] Exemplarily, the second preset density value can be based on actual conditions, and embodiments of the present application do not make a limitation in comparison.

[0176] Step S406: The dangerous node, the disaster node, and the congested road section are updated to the personnel evacuation network flow model.

[0177] It can be understood that the dangerous node and the disaster node can be determined from a plurality of nodes; the dangerous node is a node with a personnel flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs. If the personnel flow of a path represented by an edge between two nodes is greater than or equal to a second preset density value, the edge between the two nodes is determined as a congested road section, otherwise, as a smooth road section.

[0178] As shown in FIG. 6, it is a schematic diagram of the personnel evacuation network flow model updated by step S406 provided by embodiments of the present application. Figure 8 As shown in FIG. 6, it is a schematic diagram of the personnel evacuation network flow model updated by step S406 provided by embodiments of the present application.

[0179] As shown in FIG. 6, it is a schematic diagram of the personnel evacuation network flow model updated by step S406 provided by embodiments of the present application. Figure 6 As shown in FIG. 6, it is a schematic diagram of the personnel evacuation network flow model updated by step S406 provided by embodiments of the present application. Figure 8 In FIG. 6, the circle filled with white (i.e., white circular node) represents an escape exit node, such as escape exit node 5, escape exit node 21, and escape exit node 36. Figure 8 In FIG. 6, the circle filled with grid (grid circular node) represents a disaster node, such as disaster node 7 and disaster node 46. Figure 8 In FIG. 6, the rectangle filled with gray (gray square node) represents a staircase node. It can be understood that if the staircase node is a fire node, the color of the staircase node can change. Figure 6 In FIG. 6, the circle filled with gray (gray circular node) represents a normal node, which does not have a fire and has a personnel flow density less than a first preset density value.

[0180] Figure 8 In FIG. 6, the dashed line represents a congested road section and a road section where a fire occurs.

[0181] The gray circular node, i.e., the normal node, represents a normal node through which the personnel can pass smoothly. The grid circular node, i.e., the disaster node, represents a node where a disaster occurs and can be used as a starting node for evacuation but cannot be used as a target node in the next step. The disaster node does not need to be considered in the static evacuation path planning stage. The white circular node, i.e., the escape exit node, represents an escape exit to which the personnel needs to arrive. The gray square node, i.e., the stair node, represents a stair, and there are four stairs on each floor of the building. The edges are divided into two types according to whether they are congested. The black solid line edge represents a normal path through which the personnel can pass smoothly. The dashed line edge represents a congested path or a path through which congestion is likely to occur, such as a path with an average width of 0.8 m to 1.1 m. Since congestion is likely to occur when the personnel pass through these paths, the personnel should try to avoid passing through these paths if there are better paths available.

[0182] It can be understood that, Figure 8 The dangerous nodes, fire nodes, and congested paths shown in the density matrix can be used to subsequently obtain the density matrix and the speed matrix. Figure 8 The personnel evacuation network flow model shown is the personnel evacuation network flow model after step S406 is performed.

[0183] Step S407: Based on the density matrix, a speed matrix corresponding to the personnel evacuation network flow model is obtained.

[0184] The speed matrix includes the moving speed of the personnel in the paths between all pairs of nodes with edges in the personnel evacuation network flow model.

[0185] The process of obtaining the speed matrix is described below.

[0186] Speed matrix represents the moving speed of the personnel on each path at t1. If the density matrix is represented by Figure 6 The personnel evacuation network flow model shown is an example. is also a 68x68 matrix. The element value of t1 is which can be calculated according to the following formula, where, represents the moving speed when going down the stairs, represents the moving speed on the horizontal path, and the elements in the set S represent the serial numbers of the nodes that the personnel need to pass through when going down the stairs.

[0187] ; taking Figure 6 as an example, the set .

[0188] ;

[0189] .

[0190] Step S408: obtaining a time consumption matrix corresponding to the personnel evacuation network flow model based on the distance matrix and the speed matrix.

[0191] The process of obtaining the time consumption matrix is described below.

[0192] Time consumption matrix represents the time consumed by the personnel moving from node i to node j at t1. If the time consumption matrix is divided by the distance matrix and the speed matrix, as shown in formula Figure 6 The personnel evacuation network flow model shown is taken as an example. The time consumption matrix is also a 68x68 matrix, and the calculation method is obtained by dividing the corresponding elements of the distance matrix and the speed matrix, as shown in formula If the element value in the time consumption matrix is 0, it indicates that the corresponding two nodes are not directly connected, and the time required for the transition between nodes is 0.

[0193] Step S409: inputting the personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix, and the time consumption matrix into a pre-constructed evacuation path prediction model, and obtaining a first escape path corresponding to each node through the evacuation path prediction model.

[0194] The first escape path is a static evacuation path.

[0195] The static evacuation path refers to a static evacuation stage once a fire or other emergency occurs before the detailed disaster information is obtained. In this stage, a static evacuation path is planned for all personnel with the current location as the starting point to guide their escape in the shortest time.

[0196] Illustratively, the evacuation path prediction model can be a machine learning model.

[0197] Illustratively, the training process of the evacuation path prediction model is as follows: taking the personnel evacuation network flow model, the distance matrix, the sample density matrix, the sample speed matrix, and the sample time consumption matrix as the input of the machine learning model, taking the labeled escape path as the training target, training the machine learning model to obtain the evacuation path prediction model.

[0198] It can be understood that the process of obtaining the sample density matrix, the sample speed matrix, and the sample time consumption matrix is the same as that of obtaining the density matrix, the speed matrix, and the time consumption matrix, which is not limited here.

[0199] Step S410: sending the first escape path corresponding to each node to the terminal corresponding to the node for escape.

[0200] Step S411: If it is detected that the flow density of any of the nodes changes, or the flow density of the path between any two nodes changes, or the moving speed of the personnel in the path between any two nodes changes, or the fire condition of any of the nodes changes, the updated personnel evacuation network flow model, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix are obtained.

[0201] It can be understood that if any node changes from a normal node to a disaster node, it is determined that the fire condition changes. If any node changes from a normal node to a dangerous node or from a dangerous node to a normal node, it is determined that the flow density changes. If the flow density of the path between any two nodes changes from greater than or equal to the second preset density value to less than the second preset density value, or from less than the second preset density value to greater than or equal to the second preset density value, it is determined that the flow density changes.

[0202] The updated personnel evacuation network flow model, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix can be obtained in real time.

[0203] The parameter that affects the element value in the density matrix or the speed matrix or the evacuation time consumption matrix is referred to as disaster detailed information (i.e., the flow density changes, or the flow density of the path between any two nodes changes, or the moving speed of the personnel in the path between any two nodes changes, or the fire condition of the node changes).

[0204] The process of determining the disaster detailed information in step S411 is described below.

[0205] For each of the nodes, whether the node changes from a disaster occurrence location to a non-disaster occurrence location or from a non-disaster occurrence location to a disaster occurrence location is detected by the smoke sensor deployed at the node. If yes, it is determined that the fire condition of the node changes.

[0206] For each of the nodes, the flow density of the personnel located at the node is detected by the sensor deployed at the node. If the flow density changes from greater than or equal to the first preset density value to less than the first preset density value, or from less than the first preset density value to greater than or equal to the first preset density value, it is determined that the flow density of the node changes.

[0207] For each edge between two nodes, a sensor deployed on the path between the two nodes detects the crowd density on the path, and if the crowd density changes from greater than or equal to a second preset density value to less than the second preset density value, or changes from less than the second preset density value to greater than or equal to the second preset density value, it is determined that the crowd density on the path changes.

[0208] For each edge between two nodes, a smoke sensor deployed on the path between the two nodes detects the smoke concentration on the path, and if the smoke concentration corresponding to the smoke concentration level changes, it is determined that the moving speed of the personnel on the path between the two nodes changes.

[0209] It can be understood that some factors may affect the moving speed of personnel, for example, the greater the smoke concentration, the slower the moving speed of personnel. The following takes the CO concentration as an example. For each edge between two nodes, the method for obtaining the updated moving speed corresponding to the path between the two nodes includes the following steps A1 to A2.

[0210] Step A1: calculate the coefficient a; p by the formula co is the CO concentration of the path between the two nodes.

[0211] Step A2: determine that the updated moving speed is the product of the moving speed before updating and the coefficient a.

[0212] The above formula abstractly represents the influence of fire smoke on the evacuation speed of personnel.

[0213] Step S412: input the updated personnel evacuation network flow model, the distance matrix, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix into the evacuation path prediction model, and obtain a second escape path through the evacuation path prediction model.

[0214] The second escape path is a dynamic evacuation path, and as time elapses, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix constantly change, and the second escape path also constantly changes.

[0215] The application embodiment refers to the parameter affecting the element value in the density matrix or the speed matrix or the evacuation time consumption matrix as disaster detailed information (i.e. the crowd density changes, or the crowd density on the path between two nodes changes, or the moving speed of personnel on the path between two nodes changes, or the fire condition of a node changes).

[0216] ​In actual emergency evacuation process, the influence of dynamic environmental factors such as fire spread and road congestion on the escaping personnel must be considered. In the evacuation process, once the disaster and other environmental factors change, if the static evacuation path is still used, there may be great safety hazards. Therefore, after detecting the detailed information of the changed disaster, the evacuation path must be dynamically adjusted according to the real-time environmental information, that is, the dynamic evacuation path, that is, the second escape path, is obtained.

[0217] In order for those skilled in the art to better understand the static evacuation path and the dynamic evacuation path mentioned in the embodiments of the present application, the following will be described in conjunction with Figure 9 .

[0218] As shown in Figure 9 , it is a comparison diagram of the static evacuation path and the dynamic evacuation path provided by the embodiments of the present application.

[0219] As shown in Figure 9 , when a fire occurs, the personnel are located at the initial node, and the static evacuation path can be obtained based on the current fire situation; if the detailed information of the changed disaster is obtained during the evacuation of the personnel based on the static evacuation path, the dynamic evacuation path can be obtained. The personnel can evacuate based on the dynamic evacuation path.

[0220] It can be understood that, as time goes by, the detailed information of the disaster may change constantly, and then multiple dynamic evacuation paths, that is, multiple second escape paths, can be obtained.

[0221] Step S413: For each node, the second escape path corresponding to the node is sent to the terminal corresponding to the node for escaping.

[0222] The embodiment of the application provides a fire emergency evacuation indication method applied to a large indoor scene. The space characteristics of an actual evacuation scene can be restored through a personnel evacuation network flow model. The evacuation path of the actual evacuation scene can be restored through a distance matrix and the personnel evacuation network flow model. If a fire is detected, the flow density of which nodes and the flow density of the paths between which nodes are greater can be determined through a density matrix, so that the nodes that are dangerous nodes, disaster nodes, and the paths that are congested paths can be determined. A speed matrix is obtained based on the density matrix. Through the speed matrix, it can be known that the moving speed of personnel is faster in which path and slower in which path during the evacuation process. An evacuation time consumption matrix is obtained based on the distance matrix and the speed matrix. Through the evacuation time consumption matrix, it can be known that the personnel consume the least time to escape from which escape exit. The personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix, and the evacuation time consumption matrix are input into a pre-constructed evacuation path prediction model. The first escape path corresponding to all nodes is obtained through the evacuation path prediction model. Personnel can escape based on the first escape path. As time passes, the detailed information of the nodes may change, resulting in that the first escape path is no longer the most suitable escape path. At this time, the updated personnel evacuation network flow model, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix need to be obtained. The personnel evacuation network flow model, the distance matrix, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix are input into the evacuation path prediction model. The second escape path is obtained through the evacuation path prediction model. In this way, personnel can escape through the second escape path.

[0223] It can be understood that if the disaster nodes are not suitable as nodes of the escape path, the disaster nodes will be avoided during path planning to ensure the safety of the evacuation process. As time passes, if the flow density of the paths between the nodes is higher than the second preset density value, a dynamic flow density control strategy will be implemented according to the congestion condition of the current path. For example, during path planning, the paths will be avoided, and the congestion condition of the congested path will be reduced by limiting the entry into the congested path.

[0224] It can be understood that as time passes, the nodes where the personnel are currently located will change. For example, some personnel are located in disaster nodes, some personnel are located in dangerous nodes, and some personnel are located in ordinary nodes. Regardless of the nodes where the personnel are located, the escape path is updated in real time with the nodes where the personnel are currently located as the starting point, and the escape path is sent to a terminal for escape, such as a field auxiliary device (such as a sports bracelet and a voice broadcast system), so that the personnel can escape based on the escape path.

[0225] In a possible implementation, the evacuation path prediction model is a double deep Q network model, and a method for training the double deep Q network includes the following steps B01 to step B23.

[0226] Step B01: Obtain a sample density matrix corresponding to the personnel evacuation network flow model; the sample density matrix includes the passenger flow density of all nodes in the personnel evacuation network flow model and the passenger flow density of the passage between each pair of nodes with an edge.

[0227] It can be understood that different personnel evacuation network flow models correspond to different evacuation path prediction models.

[0228] It can be understood that the scenario of a fire occurring can be simulated, and the agent can be used as personnel who need to evacuate to obtain the sample density matrix.

[0229] The agent (Agent) plays the role of evacuating personnel, needs to constantly select paths, find the optimal path for escape, and is an entity that performs actions.

[0230] Step B02: Based on the sample density matrix, obtain a sample speed matrix corresponding to the personnel evacuation network flow model; the sample speed matrix includes the moving speed of personnel in the passage between each pair of nodes with an edge in the personnel evacuation network flow model.

[0231] The process of obtaining the sample speed matrix is the same as that of obtaining the speed matrix, which will not be repeated here.

[0232] Step B03: Based on the sample distance matrix and the sample speed matrix, obtain a sample evacuation time consumption matrix corresponding to the personnel evacuation network flow model.

[0233] The process of obtaining the sample evacuation time consumption matrix is the same as that of obtaining the evacuation time consumption matrix, which will not be repeated here.

[0234] Step B04: Determine a dangerous node and a disaster node from the plurality of nodes; the dangerous node is a node with a passenger flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs.

[0235] Step B05: If the passenger flow density of the passage represented by the edge between each pair of nodes is greater than or equal to a second preset density value, determine the edge between each pair of nodes as a congested section.

[0236] Step B06: Update the dangerous node, the disaster node, and the congested section to the personnel evacuation network flow model.

[0237] Step B07: Obtain state information of the agent, the state information including the time consumption of moving from the first node to each second node.

[0238] The first node is the node where the agent is located before moving.

[0239] The sample dispersion time consumption matrix integrates the information of the sample density matrix, the distance matrix and the sample velocity matrix, and can comprehensively reflect the real-time position information of the agent in the dispersion environment. Since the agent can only move to the next node directly connected to it each time, i.e., the agent can move from the first node to any second node. Then the time consumption from the first node to the second node is the element value in the sample dispersion time consumption matrix. Assuming that the number of second nodes directly connected to the first node is e, and the serial number of the first node is ti, the serial numbers of the e second nodes are j1, j2, …, je in turn; assuming that the time consumption of the first node moving to the second node j1 is T ti j1 characterizes the time consumption of the first node moving to the second node j2 in the sample dispersion time consumption matrix as T ti j2 characterizes the time consumption of the first node moving to the second node j2 in the sample dispersion time consumption matrix as T ti je characterizes the time consumption of the first node moving to the second node j2 in the sample dispersion time consumption matrix as T .

[0240] Step B08: inputting the personnel dispersion network flow model, the distance matrix, the sample density matrix, the sample velocity matrix, the sample dispersion time consumption matrix and the state information into the double deep Q network model, and outputting the predicted Q values corresponding to a plurality of actions respectively through the double deep Q model.

[0241] The double deep Q network model comprises a deep Q network model and a target Q network model; and the action is the action of the agent moving from the first node to the second node.

[0242] In the DDQN (Double Deep Q-Network) model, the action selected by the agent at each time usually only depends on the action A t1 , i.e. . The time of the agent at the first node is time t1. Wherein, Q i refers to the predicted Q value corresponding to the action of moving from the first node to the second node ji.

[0243] However, if the agent selects the action corresponding to the maximum predicted Q value each time, although it can successfully complete the escape task, it will also lose the opportunity to explore a better escape path, i.e., the reinforcement learning falls into a local minimum value. In order to avoid this situation, the application adopts an ε-greedy strategy to determine the action taken by the agent next time, i.e., step B08 is followed by step B10.

[0244] Step B09: obtaining a random number.

[0245] Step B10: if the random number is greater than the preset threshold, randomly selecting a target action from the plurality of actions, a second node in the target action being a target second node.

[0246] Step B11: if the random number is less than or equal to the preset threshold, selecting a target action corresponding to the maximum expected Q value from the plurality of actions; a second node in the target action being a target second node.

[0247] For example, the application sets the exploration rate ε to 5%, i.e. the preset threshold is 0.95. When implementing the ε-greedy strategy, a random number between 0 and 1 is randomly generated, and if the random number is less than or equal to 0.95, the action corresponding to the maximum predicted Q2 value is selected; otherwise, a random selection is made from the optional actions.

[0248] Step B12: determining a first reward for the agent moving to the target second node based on the shortest distance between the target second node and the escape exit node, whether the target second node is the disaster node, whether the target second node is the danger node, whether the target second node is the escape exit node, and whether the target time taken by the agent to reach the target second node from the initial node exceeds a preset specified duration.

[0249] Step B13: inputting the first reward and the predicted Q value corresponding to the target action into the double deep Q network model to update the deep Q network model in the double deep Q network model.

[0250] In order to enable the agent to move towards the escape exit node when selecting an escape path, the application proposes a reward function design based on a distance estimator. To implement the distance estimator, a distance estimation list needs to be initialized. Each element in the list represents the shortest estimated distance from the node to any escape exit node. When the agent moves from the first node to the next target second node, if the target second node is closer to the escape exit node than before and the target second node is not a disaster node, the action is considered as a positive incentive, and the agent will get a first reward R t1 ; if the target second node is further away from the escape exit node than before and the target second node is not a disaster node, the action is considered as a negative incentive, and the agent will get a first reward R t1 .

[0251] wherein k1 is a preset reward amplification coefficient for controlling the optimization efficiency.

[0252] ​​In summary, if the target second node is not the disaster node and is not the escape exit node, and the target time consumption is less than the preset specified duration, the R is calculated by the formula t1 , and R t1 is determined as the first reward.

[0253] If the target second node is not the escape exit node, and the target time consumption is greater than the preset specified duration, or the target second node is the disaster node, it indicates that the path is relatively dangerous, and the second preset reward value is determined as the first reward. Exemplarily, the second preset reward value is -100.

[0254] Step B14: If the target second node is not the disaster node and is not the escape exit node, and the target time consumption does not exceed the preset specified duration, return to step B07.

[0255] Step B15: If the target second node is the disaster node or the target time consumption exceeds the preset specified duration, it is determined that the agent returns to the initial node, and returns to step B07.

[0256] Step B16: If the target second node is the escape exit node, an escape planning path is obtained, and the escape planning path is a path of the agent from the initial node to the escape exit node.

[0257] Step B17: Based on the number of disaster nodes in the escape planning path, the number of dangerous nodes in the escape planning path, the number of congested road sections in the escape planning path, and the length of the escape planning path, a second reward is determined.

[0258] In order to enable the agent to avoid disaster nodes and congested road sections and select relatively unobstructed safe road sections when selecting an escape path, the application proposes a reward function design based on a path scorer.

[0259] The path scorer is designed to evaluate the quality of the generated escape planning path, and to provide a corresponding reward for the agent at the end of each round according to the quality of the escape planning path. In terms of scoring standards, the path scorer mainly considers the length of the escape planning path, the safety of the escape planning path, and the unobstructedness of the escape planning path. The calculation method of the path total score is as follows:

[0260] .

[0261] Wherein, , , respectively represent the length score of the escape planning path, the safety score of the escape planning path and the smoothness score of the escape planning path. In the evacuation process, the escape planning path with shorter distance will also take relatively shorter time, but if there are disaster nodes in the escape planning path, it may cause secondary accidents in the process of people escaping, so the length score and safety score of the escape planning path need to be considered. In addition, the existence of congested road sections in the escape planning path may cause congestion of personnel, thereby reducing the escape efficiency, so the smoothness score of the escape planning path also needs to be considered. The three factors are equally important in the judgment of the quality of the escape planning path, so the score weights of the three factors are each 1 / 3, thereby ensuring that the sum of the maximum theoretical scores is 1.

[0262] wherein, , wherein is the three-dimensional coordinate of the initial node; is the three-dimensional coordinate of the escape exit node closest to the initial node; is the three-dimensional coordinate of the jth node in the escape planning path; N is the total number of all nodes in the escape planning path. Wherein, is the Manhattan distance.

[0263] The length factor of the escape planning path is considered, and the larger the value is, the straighter the escape planning path is, the shorter the escape planning path is, and the higher the probability of successful evacuation is.

[0264] wherein, is the number of dangerous nodes in the escape planning path; is the number of disaster nodes in the escape planning path.

[0265] The relative number of intelligent agents passing through disaster nodes and dangerous nodes in the escape process is described, and the higher the number is, the fewer the number of dangerous nodes and disaster nodes that the intelligent agent passes through on the escape planning path is, and the higher the safety of the escape planning path is.

[0266] wherein, , is the three-dimensional coordinate of the starting node of the ith congested road section; is the three-dimensional coordinate of the ending node of the ith congested road section.

[0267] The numerator represents the sum of the lengths of all congested road sections, and the denominator represents the length of the entire escape planning path. The larger the value is, the smoother the escape planning path is, and the higher the personnel escape efficiency is.

[0268] Since the entire escape planning path can only be scored after the complete escape planning path is generated, at the end of each round, the agent will obtain a second reward according to the escape planning path. At the end of the current round, that is, if the target second node is the escape exit node and the target time consumption is less than or equal to the preset specified time length, the second reward R is calculated by the formula -1 -1 .

[0269] wherein k2 is a preset scaling factor, k3 is a preset time coefficient, and t2 is the total time consumption of the agent from the initial node to the escape exit node.

[0270] wherein k2 is used to prevent the difference between the two rewards obtained after the end of the current round from being too large, and k3t2 is a time term, which has two main effects: on the one hand, it draws on the idea of batch normalization to ensure that the proportions of the rewards are reasonable; on the other hand, it considers the time index of the escape planning path, so that the agent is more inclined to select an escape planning path with a shorter distance and the least time consumption, thereby improving the quality of the escape planning path and speeding up the escape process.

[0271] Step B18: inputting the second reward and the predicted Q value corresponding to the target action into the deep Q network model to update the deep Q network model.

[0272] Step B19: obtaining the target real Q value of the action of moving from a previous node to a next node in the escape planning path.

[0273] The following is an example. Assuming that the escape planning path is: node A→node B→node C→node D, the target real Q value of the action of moving from a previous node to a next node in the escape planning path includes: the target real Q value of the action of moving from node A to node B, the target real Q value of the action of moving from node B to node C, and the target real Q value of the action of moving from node C to node D.

[0274] Step B20: if the target real Q values corresponding to the nodes with edges in the escape planning path in the escape order successively increase, determining that the escape planning path is the optimal path.

[0275] Still taking the above example, the target real Q values in the escape order successively are: the target real Q value of the action of moving from node A to node B, the target real Q value of the action of moving from node B to node C, and the target real Q value of the action of moving from node C to node D.

[0276] ​​Step B21: If the target real Q values corresponding to the two-by-two node pairs with edges in the escape planning path according to the escape order do not sequentially increase, it is determined that the agent returns to the initial node, and the process returns to step B07.

[0277] Step B22: If the number of updates of the deep Q network model reaches the preset number of updates, the parameters of the deep Q network model are assigned to the target Q network model.

[0278] Illustratively, the preset number of updates can be based on actual conditions, which is not limited here. For example, the preset number of updates is 15.

[0279] Step B23: If all nodes have been obtained corresponding to the optimal path, it is determined that the dual deep Q network model training is complete.

[0280] The process of inputting the target reward and the predicted Q value corresponding to the target action into the dual deep Q network model to update the deep Q network model in the dual deep Q network model is described below, which includes the following steps C1 to step C4.

[0281] Step C1: If the target second node is not the disaster node and is not the escape exit node, and the target time consumption is less than the preset specified duration, the real Q value of the agent moving from the first node to the target second node is calculated by the formula

[0282] wherein, is the real Q value, is the predicted Q value of the agent moving from the first node to the target second node output by the dual Q network model; lr is a preset learning rate parameter, and γ is a preset discount factor, is a preset function.

[0283] Illustratively, the preset function can be artificially obtained.

[0284] ​In the iterative updating process of the DDQN model, it is mainly divided into two parts: the updating of the real Q value and the updating of the internal parameters of the deep Q network model. When updating the real Q value, the DDQN model is updated according to the target reward given by the environment, so that the agent can obtain the maximum total reward in the escape process. Since the reward obtained by future actions cannot be predicted in the current escape process, the target Q network model is needed to indirectly provide the future reward. In addition, the final reward cannot be directly obtained when the escape exit node is not reached, so a discount factor γ is introduced to control the degree of discounting of future rewards. The larger γ is, the more the agent values the reward obtained after reaching the escape exit node; on the contrary, the more the agent values the reward of the current single step action. In order to successfully escape, more attention should be paid to the potential reward obtained after reaching the escape exit node than the reward of moving one step, so γ should take a larger value. At the same time, it is also necessary to define the learning rate parameter lr, which represents the degree to which future information covers current information. The larger lr is, the more attention is paid to future rewards. In the present application, the two parameters take fixed values γ = 0.95 and lr = 0.01 respectively. Finally, the real Q value corresponding to the selected action is updated by weighting .

[0285] Step C2: If the target second node is the escape exit node and the target time consumption is less than or equal to the preset specified time length, determine R -1 as the real Q value of the agent moving from the first node to the target second node.

[0286] Step C3: If the target second node is not the escape exit node and the target time consumption is greater than the preset specified time length, or the target second node is the disaster node, determine the second preset reward value as the real Q value of the agent moving from the first node to the target second node.

[0287] If the agent reaches the escape exit node or does not reach the escape exit node within the preset specified time length or moves to the disaster node, the current iteration round ends. If the agent reaches the escape exit node, the real Q value R -1 . If the agent does not reach the escape exit node within the preset specified time length or moves to the disaster node, the real Q value is the second preset reward value, i.e. -100.

[0288] Step C4: Update the deep Q network model through the formula .

[0289] In the DDQN model, the update of the real Q value and the internal parameters of the deep Q network model are different. For the parameter update of the deep Q network model, the error back propagation method can be used. Through several rounds of iterative training of the deep Q network by the least square loss function, the error between each predicted Q value and the real Q value is calculated, and the error is back propagated to the deep Q network model, so that the predicted Q value output by the deep Q network model at a given position is consistent with the real Q value, thereby completing the parameter update of the deep Q network model. The specific update method can minimize the loss function .

[0290] The structure of the deep Q network model is described below.

[0291] As shown in Figure 10 , it is a structure diagram of the deep Q network model provided by the embodiment of the application.

[0292] As shown in Figure 10 , the deep Q network model includes a first CSP (Cross Stage Partial) module 1001, a first CBS module 1002, a second CSP module 1003, a second CBS module 1004, a third CSP module 1005, and a third CBS module 1006 connected in sequence.

[0293] The structure of the first CSP module is described below. As shown in Figure 11 , it is a structure diagram of the first CBS module provided by the embodiment of the application.

[0294] The first CSP module includes a module 1101 carrying an unfold function, a fourth CBS module 1102 connected to the first output end of the module carrying the unfold function, a first BottleNeck module 1103 connected to the second output end of the module carrying the unfold function, and a second BottleNeck module 1104 connected to the third output end of the module carrying the unfold function; a fifth CBS module 1105 connected to the output end of the fourth CBS module, the output end of the first BottleNeck module, and the output end of the second BottleNeck module, respectively; and a sixth CBS module 1106 connected to the fifth CBS module.

[0295] The first CBS module includes a convolution layer, a batch normalization layer (BN2D) connected to the convolution layer (CONV), and an activation function layer (SWISH) connected to the batch normalization layer.

[0296] The first CSP module, the second CSP module and the third CSP module are the same in structure; the first CBS module, the second CBS module, the third CBS module, the fourth CBS module, the fifth CBS module and the sixth CBS module are the same in structure.

[0297] The second BottleNeck module and the first BottleNeck module are the same in structure.

[0298] The first BottleNeck module comprises three LBS modules connected in sequence. The LBS module comprises a fully connected layer (LINEAR), a batch normalization layer (BN2D) connected with the fully connected layer, and an activation function layer (SWISH) connected with the batch normalization layer.

[0299] The deep Q network model provided in the application comprises a plurality of modules. The LBS module is mainly used for extracting basic features related to the state sequence of each node in the personnel network flow evacuation model. The CBS module (the first CBS module to the sixth CBS module are collectively referred to as the CNS module) is mainly used for extracting cross-node features between different node input states, such as the time comparison information of the agent crossing the node in the evacuation process. The BottleNeck module (the first BottleNeck module and the second BottleNeck module are collectively referred to as the BottleNeck module) is also called a bottleneck module. It reduces the calculation complexity of the deep Q network model by reducing the dimension first and then increasing the dimension, and effectively extracts independent feature information in the node input state, such as evacuation time and predicted Q value. The CSP module (the first CSP module to the third CSP module are collectively referred to as the CSP module) comprises a plurality of BottleNeck modules and CBS modules, which can extract both independent features of the node and cross-node features between the nodes, and finally concatenate the feature vectors and change the number of paths through the CBS module.

[0300] The deep Q network model structure comprises three CSP modules and three CBS modules. Each CSP module is followed by a CBS module, and they are stacked in this way. The input of the deep Q network model is the state vector corresponding to the node (i.e. the personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix and the evacuation time matrix). The number of paths of each layer changes as 1-64-128-128-64-1. The output of the deep Q network model is a one-dimensional list, wherein each element represents the predicted Q value of the agent taking different actions after moving to the first node. These predicted Q values can be used to evaluate which action can maximize the benefit after the agent moves to the first node, so as to train the optimal path.

[0301] It can be understood that the DDQN model comprises a deep Q network model and a target Q network model, and the purpose is to find the action A corresponding to the maximum predicted Q value in the deep Q network model first, and then use the action A to calculate the target Q value in the target Q network model. Generally, weight sharing needs to be maintained between the deep Q network model and the target Q network model, so the structure design of the two should also be the same, and the network parameters of the deep Q network model will be synchronized to the target Q network model regularly.

[0302] Figure 6 Taking a bank business site as an example, a fire emergency evacuation indication method applied to a large indoor scene is introduced. First, taking a bank business site with a business hall, office area and basement as an actual evacuation scene, the network flow is converted into a personnel evacuation network flow model composed of nodes and edges, and the corresponding model parameters are set. Secondly, based on the DDQN model of deep reinforcement learning, the personnel network flow model is abstracted into a distance matrix, a density matrix, a time matrix and a speed matrix, and by designing the state and action elements, the reward function and the DDQN model, the static evacuation route is planned, aiming to provide an evacuation path plan at the first time after the disaster occurs to guide the personnel to escape quickly. Finally, according to the real-time disaster information obtained by the bank business site internal sensing detection system, fully considering the problems of smoke diffusion, disaster spread and personnel congestion, etc., the dynamic evacuation path is planned on the basis of the static evacuation path, and the on-site auxiliary equipment is used to guide the personnel to escape through voice broadcast and picture prompt, and this process is repeated until all personnel complete the evacuation.

[0303] In an optional implementation, the data center obtains real-time environmental parameters (temperature, smoke, personnel density, etc.) in the evacuation process through sensors installed inside the bank business site, and stores these parameters in a database

[0304] The fire emergency evacuation indication method applied to the large indoor scene provided by the embodiments of the present application has strong universality, is suitable for emergency evacuation indication of all bank branches including bank business outlets, and only needs to replace the corresponding personnel network flow evacuation model for different evacuation scenes without modifying the algorithm process and system structure. In the process of indicating personnel evacuation, the present application not only considers the escape planning path length factor, but also takes the evacuation time consumption as an index, focuses on the overall evacuation efficiency, and is more practical. The deep reinforcement learning DDQN model is adopted to obtain the escape planning path, which has stronger non-constant data fitting capability and faster convergence speed compared with the traditional path planning algorithm. The quality of the escape planning path is considered through the reward function design based on the path scorer, and the safety in the escape process is improved. In addition, the dynamic environmental information such as fire diffusion, smoke spread and personnel density is also considered, which can make dynamic adjustment to the escape planning path in time and is more in line with the actual personnel emergency evacuation demand of the bank business outlet.

[0305] The embodiments of the present application use the DDQN model. The DDQN model is an algorithm in the field of deep reinforcement learning, which does not have a specific training set, needs to interact with the environment continuously to obtain samples, has a deep Q network model and a target Q network model, and can fit more complex states. In order to apply the DDQN model to the personnel evacuation scene of the large indoor scene (taking the bank business outlet as an example), the definitions of several basic concepts of the DDQN model need to be combined with the evacuation path planning background of the large indoor scene (taking the bank business outlet as an example).

[0306] Agent: The agent plays the role of the evacuating personnel, needs to select the path continuously, finds the optimal escape path, and is an entity that executes actions.

[0307] Environment: The environment refers to the environmental information such as the office area, passageway intersection, bank self-service terminal device, fire staircase, vault and the like inside the bank business outlet. It interacts with the agent by providing feasible nodes and paths.

[0308] State: The state is the information directly observed by the agent from the environment, and is the only basis for decision-making, mainly including data such as the time required for the current node to reach the next adjacent node, the length of the path, the number of personnel on the path and the personnel travel speed.

[0309] Action: The action is the expression of the behavior executed by the agent, and in the bank business outlet evacuation scene mainly refers to the behavior action of the agent moving from the current node to the next adjacent node.

[0310] Reward: The reward is the feedback evaluation of the agent executing a certain action in the evacuation process.

[0311] AsFigure 12 Fig. 1 is a schematic diagram of an environment interaction process between an intelligent agent and a bank business site according to an embodiment of the present application.

[0312] Figure 12 The process of the intelligent agent interacting with the bank business site environment is shown. At system t1, the intelligent agent takes an action A t1 in the current state S t1 , obtains a reward R t1 , and enters the next state. This process of the intelligent agent interacting with the environment is repeatedly performed, and eventually a plurality of experience sequences are formed. The intelligent agent learns the optimal path from the current node to the escape exit node according to the experience sequences.

[0313] The above introduces a fire emergency evacuation indication method applied to a large indoor scene according to an embodiment of the present application. The following introduces a device for executing the above-mentioned fire emergency evacuation indication method applied to a large indoor scene.

[0314] Please refer to Figure 13 , Figure 13 Fig. 1 is a structural schematic diagram of a fire emergency evacuation indication device applied to a large indoor scene according to an embodiment of the present application. As shown in the figure, the fire emergency evacuation indication device applied to a large indoor scene includes: Figure 13

[0315] A first acquisition module 1201 is configured to acquire a personnel evacuation network flow model, the personnel evacuation network flow model including a plurality of nodes and edges between two nodes; the edges between two nodes represent a passage between two nodes for personnel walking; the plurality of nodes include an escape exit node, a normal node, and a staircase node.

[0316] A second acquisition module 1202 is configured to acquire a distance matrix corresponding to the personnel evacuation network flow model; the distance matrix includes distances between all nodes with edges in the personnel evacuation network flow model.

[0317] A third acquisition module 1203 is configured to acquire a density matrix corresponding to the personnel evacuation network flow model if a fire is detected; the density matrix includes a person flow density of all nodes in the personnel evacuation network flow model and a person flow density of the passage between two nodes with edges.

[0318] A first determination module 1204 is configured to determine a dangerous node and a disaster node from the plurality of nodes; the dangerous node is a node with a person flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs.

[0319] ​The second determining module 1205 is configured to determine that the edge between the two nodes is a congested road segment if the flow density of the path represented by the edge between the two nodes is greater than or equal to a second preset density value.

[0320] The first updating module 1206 is configured to update the dangerous node, the disaster node, and the congested road segment to the personnel evacuation network flow model.

[0321] The fourth obtaining module 1207 is configured to obtain a speed matrix corresponding to the personnel evacuation network flow model based on the density matrix, wherein the speed matrix includes the moving speed of personnel in the path between all pairs of nodes with edges in the personnel evacuation network flow model.

[0322] The fifth obtaining module 1208 is configured to obtain an evacuation time consumption matrix corresponding to the personnel evacuation network flow model based on the distance matrix and the speed matrix.

[0323] The sixth obtaining module 1209 is configured to input the personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix, and the evacuation time consumption matrix into a pre-constructed evacuation path prediction model, and obtain a first escape path corresponding to each node by using the evacuation path prediction model.

[0324] The first sending module 1210 is configured to send the first escape path corresponding to each node to a terminal corresponding to the node for escape.

[0325] The seventh obtaining module 1211 is configured to obtain an updated personnel evacuation network flow model, an updated density matrix, an updated speed matrix, and an updated evacuation time consumption matrix if any of the following conditions is detected: the flow density of any node changes, the flow density of the path between any pair of nodes changes, the moving speed of personnel in the path between any pair of nodes changes, or the fire condition of any node changes.

[0326] The eighth obtaining module 1212 is configured to input the updated personnel evacuation network flow model, the distance matrix, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix into the evacuation path prediction model, and obtain a second escape path by using the evacuation path prediction model.

[0327] The second sending module 1213 is configured to send the second escape path corresponding to each node to a terminal corresponding to the node for escape.

[0328] In an optional implementation, the method further includes:

[0329] a third determining module configured to determine, for each node, whether a fire condition of the node is changed by detecting, by a smoke sensor deployed at the node, whether the node is changed from a disaster-occurred place to a non-disaster-occurred place or from a non-disaster-occurred place to a disaster-occurred place;

[0330] a fourth determining module configured to determine, for each node, whether a crowd density of the node is changed by detecting, by a sensor deployed at the node, the crowd density of the node, if the crowd density is changed from greater than or equal to the first preset density value to less than the first preset density value or from less than the first preset density value to greater than or equal to the first preset density value;

[0331] a fifth determining module configured to determine, for each edge between two nodes, whether a crowd density of a passage between the two nodes is changed by detecting, by a sensor deployed at the passage, the crowd density of the passage, if the crowd density is changed from greater than or equal to a second preset density value to less than the second preset density value or from less than the second preset density value to greater than or equal to the second preset density value;

[0332] a sixth determining module configured to determine, for each edge between two nodes, whether a moving speed of a person in a passage between the two nodes is changed by detecting, by a smoke sensor deployed at the passage, a smoke concentration of the passage, if a smoke concentration level corresponding to the smoke concentration is changed.

[0333] In an optional implementation, the smoke concentration is a CO concentration, and the seventh obtaining module includes:

[0334] a first calculating unit configured to calculate, for each edge between two nodes, a coefficient a; p by a formula:

[0335] wherein the coefficient a; p is calculated based on a CO concentration of the passage between the two nodes. co

[0336] a first determining unit configured to determine, for each edge between two nodes, that an updated moving speed corresponding to the edge between the two nodes is a product of an un-updated moving speed corresponding to the edge between the two nodes and the coefficient a; p.

[0337] In an optional implementation, the evacuation path prediction model is a double deep Q network model, and the method further includes:

[0338] ​a ninth obtaining module, configured to obtain a sample density matrix corresponding to the personnel evacuation network flow model; the sample density matrix comprises a person flow density corresponding to each node in the personnel evacuation network flow model and a person flow density of a passage between two nodes having an edge;

[0339] a tenth obtaining module, configured to obtain a sample speed matrix corresponding to the personnel evacuation network flow model based on the sample density matrix; the sample speed matrix comprises a moving speed of a person in a passage between two nodes having an edge in the personnel evacuation network flow model;

[0340] an eleventh obtaining module, configured to obtain a sample evacuation time consumption matrix corresponding to the personnel evacuation network flow model based on the sample distance matrix and the sample speed matrix;

[0341] a seventh determining module, configured to determine a dangerous node and a disaster node from the plurality of nodes; the dangerous node is a node with a person flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs;

[0342] an eighth determining module, configured to determine an edge between two nodes as a congested section if a person flow density of a passage represented by the edge between the two nodes is greater than or equal to a second preset density value;

[0343] a second updating module, configured to update the dangerous node, the disaster node and the congested section to the personnel evacuation network flow model;

[0344] a twelfth obtaining module, configured to obtain state information of an agent, the state information comprising a time consumption of moving from the first node to each of the second nodes;

[0345] a thirteenth obtaining module, configured to input the personnel evacuation network flow model, the distance matrix, the sample density matrix, the sample speed matrix, the sample evacuation time consumption matrix and the state information to the double deep Q network model, and output a predicted Q value corresponding to each of a plurality of actions by the double deep Q network model; the double deep Q network model comprises a deep Q network model and a target Q network model; the action is an action of moving from the first node to the second node by the agent;

[0346] a fourteenth obtaining module, configured to obtain a random number;

[0347] a first selecting module, configured to randomly select a target action from the plurality of actions if the random number is greater than a preset threshold, the second node in the target action being a target second node;

[0348] The second selection module is configured to select a target action corresponding to the maximum expected Q value from the plurality of actions if the random number is less than or equal to the preset threshold value; and a second node in the target action is a target second node.

[0349] The ninth determination module is configured to determine a first reward of the agent moving to the target second node based on a shortest distance between the target second node and the escape exit node, whether the target second node is the disaster node, whether the target second node is the dangerous node, whether the target second node is the escape exit node, and whether a target time consumption of the agent from an initial node to the target second node exceeds a preset specified time length.

[0350] The third update module is configured to input the first reward and a predicted Q value corresponding to the target action into the double deep Q network model to update the deep Q network model in the double deep Q network model.

[0351] The first trigger module is configured to trigger the twelfth acquisition module if the target second node is not the disaster node and not the escape exit node, and the target time consumption does not exceed the preset specified time length.

[0352] The second trigger module is configured to trigger the twelfth acquisition module if the target second node is the disaster node or the target time consumption exceeds the preset specified time length.

[0353] The fifteenth acquisition module is configured to acquire an escape planning path if the target second node is the escape exit node, the escape planning path being a path of the agent from the initial node to the escape exit node.

[0354] The tenth determination module is configured to determine a second reward based on a number of the disaster nodes in the escape planning path, a number of the dangerous nodes in the escape planning path, a number of the congested road segments in the escape planning path, and a length of the escape planning path.

[0355] The fourth update module is configured to input the second reward and a predicted Q value corresponding to the target action into the deep Q network model to update the deep Q network model.

[0356] The sixteenth acquisition module is configured to acquire a target real Q value of an action of moving from a previous node to a next node in two nodes with an edge in the escape planning path.

[0357] The eleventh determination module is used to determine the escape planning path as the optimal path if the target true Q value corresponding to each pair of nodes with edges in the escape planning path according to the escape order increases sequentially.

[0358] The twelfth determination module is used to determine that if the target true Q value corresponding to each pair of nodes with edges in the escape planning path according to the escape order does not increase sequentially, the agent returns to the initial node and returns to the step of obtaining the state information of the agent.

[0359] The assignment module is used to assign the parameters of the deep Q-network model to the target Q-network model if the number of updates of the deep Q-network model reaches a preset number of updates.

[0360] The thirteenth determination module is used to determine that the dual-depth Q-network model has been trained if the optimal paths corresponding to all nodes have been obtained.

[0361] In one alternative implementation, the ninth determining module includes:

[0362] The second calculation unit is used to calculate the time taken to determine if the target second node is neither the disaster node nor the escape exit node, and the target time is less than the preset time limit, using the formula... R is calculated t1 Determine R t1 Here, k1 is the first reward; k1 is a preset reward amplification coefficient, where k1 is a positive number greater than 0. This refers to the shortest distance; This refers to the shortest distance between the first node and the escape exit;

[0363] The second determining unit is used to determine the second preset reward value as the first reward if the target second node is not an escape exit node and the target time is greater than the preset time limit, or if the target second node is the disaster node.

[0364] In one alternative implementation, the tenth determining module includes:

[0365] The third calculation unit is used to calculate the time taken to achieve the following: if the target second node is the escape exit node and the target time is less than or equal to the preset time, using the formula... R is calculated -1 Determine R -1 For the second reward;

[0366] Where k2 is a preset scaling factor, k3 is a preset time coefficient, and t2 is the total time taken for the agent to travel from the initial node to the escape exit node;

[0367] in, , wherein is a three-dimensional coordinate of the initial node; is a three-dimensional coordinate of the escape exit node closest to the initial node; is a three-dimensional coordinate of the jth node in the escape planning path; N is the total number of all nodes in the escape planning path; wherein, refers to the number of the dangerous nodes in the escape planning path; refers to the number of the disaster nodes in the escape planning path;

[0368] wherein, , is a three-dimensional coordinate of the starting node of the ith congested road section; is a three-dimensional coordinate of the ending node of the ith congested road section.

[0369] In an optional implementation, the target reward is the first reward or the second reward, and the third updating module or the fourth updating module comprises:

[0370] A fourth calculation unit is configured to, if the target second node is not the disaster node and is not the escape exit node and the target time consumption is less than the preset specified duration, calculate a real Q value of the agent moving from the first node to the target second node by a formula

[0371] wherein, is the real Q value, is a predicted Q value corresponding to the agent moving from the first node to the target second node output by the double Q network model; lr is a preset learning rate parameter, and γ is a preset discount factor, is a preset function;

[0372] A third determination unit is configured to, if the target second node is the escape exit node and the target time consumption is less than or equal to the preset specified duration, determine R -1 is a real Q value of the agent moving from the first node to the target second node;

[0373] A fourth determination unit is configured to, if the target second node is not the escape exit node and the target time consumption is greater than the preset specified duration, or the target second node is the disaster node, determine that the second preset reward value is a real Q value of the agent moving from the first node to the target second node;

[0374] An updating unit is configured to, if the target second node is not the escape exit node and the target time consumption is greater than the preset specified duration, or the target second node is the disaster node, update the double Q network model by a formula ​updating the deep Q network model.

[0375] In an optional implementation, the deep Q network model comprises, in sequence, a first CSP module, a first CBS module, a second CSP module, a second CBS module, a third CSP module, and a third CBS module.

[0376] The first CSP module comprises a module carrying an unfold function, a fourth CBS module connected to a first output end of the module carrying the unfold function, a first BottleNeck module connected to a second output end of the module carrying the unfold function, and a second BottleNeck module connected to a third output end of the module carrying the unfold function; a fifth CBS module connected to an output end of the fourth CBS module, an output end of the first BottleNeck module, and an output end of the second BottleNeck module, respectively; a sixth CBS module connected to the fifth CBS module.

[0377] The first CBS module comprises a convolution layer, a batch normalization layer connected to the convolution layer, and an activation function layer connected to the batch normalization layer.

[0378] The first CSP module, the second CSP module, and the third CSP module have the same structure; and the first CBS module, the second CBS module, the third CBS module, the fourth CBS module, the fifth CBS module, and the sixth CBS module have the same structure.

[0379] The application also provides a fire emergency evacuation indication system applied to a large indoor scene, which mainly comprises three parts, as shown in Figure 14 The fire emergency evacuation indication system applied to the large indoor scene provided by the application is shown in Figure 14

[0380] The first part is a personnel evacuation model, which exemplarily comprises a personnel evacuation network flow model, node attributes, and edge attributes. The node attributes refer to whether a node is a dangerous node or a disaster node, and the edge attributes refer to whether an edge is a congested road section.

[0381] The second part is to set a static evacuation strategy, which includes but is not limited to a DDQ model environmental factor, a reward function, a double Q network model, and optimal path determination.

[0382] Exemplarily, the experimental platform refers to a platform where the electronic device is located. Exemplarily, the data center is a storage space for information collected by the sensor. ​

[0383] The third part is an actual emergency evacuation system. A dynamic evacuation strategy, i.e., a dynamic evacuation path, is obtained by combining real-time environmental information through the actual emergency evacuation system.

[0384] For example, the actual emergency evacuation system includes a plurality of sensors, field auxiliary equipment, such as a terminal for escape

[0385] The embodiment of the present application also provides a computer program product including computer readable instructions, which, when executed on an electronic device, enable the electronic device to implement any of the fire emergency evacuation indication methods applied to a large indoor scene provided by the embodiments of the present application.

[0386] The embodiment of the present application also provides a computer readable storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the fire emergency evacuation indication methods applied to a large indoor scene provided by the embodiments of the present application.

[0387] In addition, it should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines.

[0388] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. Generally, functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure for implementing the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better embodiment. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer's floppy disk, U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, a training device, or a network device, etc.) execute the methods described in various embodiments of the present application.

[0389] In the embodiments described above, the entire or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, the entire or part of the embodiments can be implemented in the form of a computer program product.

[0390] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the entire or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website site, computer, training device or data center to another website site, computer, training device or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer readable storage medium can be any available medium that can be stored by the computer or data storage device such as training device, data center, etc. integrated with one or more available media sets. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD), or a semiconductor medium (for example, solid state disk (SSD)) and the like.

Claims

1. A fire emergency evacuation indication method applied to a large indoor scene, characterized in that, The method comprises the following steps: obtaining a personnel evacuation network flow model, the personnel evacuation network flow model comprising a plurality of nodes and edges between two nodes; the edges between two nodes represent a passage between two nodes for personnel walking; the plurality of nodes comprise an escape exit node, a general node, and a staircase node; obtaining a distance matrix corresponding to the personnel evacuation network flow model; the distance matrix comprises distances between all nodes with edges in the personnel evacuation network flow model; if a fire is detected, obtaining a density matrix corresponding to the personnel evacuation network flow model; the density matrix comprises a personnel flow density of all nodes in the personnel evacuation network flow model and a personnel flow density of a passage between two nodes with edges; determining a dangerous node and a disaster node from the plurality of nodes; the dangerous node is a node with a personnel flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs; if the personnel flow density of the passage represented by the edges between two nodes is greater than or equal to a second preset density value, determining that the edges between two nodes are a congested section; updating the dangerous node, the disaster node, and the congested section to the personnel evacuation network flow model; based on the density matrix, obtaining a speed matrix corresponding to the personnel evacuation network flow model; the speed matrix comprises a moving speed of personnel in a passage between all nodes with edges in the personnel evacuation network flow model; based on the distance matrix and the speed matrix, obtaining an evacuation time consumption matrix corresponding to the personnel evacuation network flow model; inputting the personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix, and the evacuation time consumption matrix into a pre-constructed evacuation path prediction model to obtain a first escape path corresponding to each node through the evacuation path prediction model; for each node, sending the first escape path corresponding to the node to a terminal corresponding to the node for escape; if a change in the personnel flow density of any node, or a change in the personnel flow density of a passage between any two nodes, or a change in the moving speed of personnel in a passage between any two nodes, or a change in the fire condition of any node is detected, obtaining an updated personnel evacuation network flow model, an updated density matrix, an updated speed matrix, and an updated evacuation time consumption matrix; inputting the updated personnel evacuation network flow model, the distance matrix, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix into the evacuation path prediction model to obtain a second escape path through the evacuation path prediction model; for each node, sending the second escape path corresponding to the node to a terminal corresponding to the node for escape; the evacuation path prediction model is a double deep Q network model, and a method for training the double deep Q network comprises the following steps: obtaining a sample density matrix corresponding to the personnel evacuation network flow model; the sample density matrix includes the flow density of each node in the personnel evacuation network flow model and the flow density of the passage between each two nodes with an edge; obtaining a sample speed matrix corresponding to the personnel evacuation network flow model based on the sample density matrix; the sample speed matrix includes the moving speed of the personnel in the passage between each two nodes with an edge in the personnel evacuation network flow model; obtaining a sample evacuation time consumption matrix corresponding to the personnel evacuation network flow model based on the sample distance matrix and the sample speed matrix; determining a dangerous node and a disaster node from the plurality of nodes; the dangerous node is a node with a flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs; if the flow density of the passage represented by the edge between each two nodes is greater than or equal to a second preset density value, determining that the edge between each two nodes is a congested section; updating the dangerous node, the disaster node, and the congested section to the personnel evacuation network flow model; obtaining state information of an agent, the state information including the time consumption of moving from a first node to each second node; inputting the personnel evacuation network flow model, the distance matrix, the sample density matrix, the sample speed matrix, the sample evacuation time consumption matrix, and the state information into the double deep Q network model, and outputting a plurality of action corresponding predicted Q values through the double deep Q network model; the double deep Q network model includes a deep Q network model and a target Q network model; the action is the action of moving the agent from the first node to the second node; obtaining a random number; if the random number is greater than a preset threshold, randomly selecting a target action from the plurality of actions, and the second node in the target action is a target second node; if the random number is less than or equal to the preset threshold, selecting a target action corresponding to the maximum predicted Q value from the plurality of actions; the second node in the target action is a target second node; determining a first reward of moving the agent to the target second node based on the shortest distance between the target second node and the escape exit node, whether the target second node is the disaster node, whether the target second node is the dangerous node, whether the target second node is the escape exit node, and whether the target time consumption of the agent from the initial node to the target second node exceeds a preset specified time length; inputting the first reward and the predicted Q value corresponding to the target action into the double deep Q network model to update the deep Q network model in the double deep Q network model; if the target second node is not the disaster node and not the escape exit node, and the target time consumption does not exceed the preset specified time length, returning to the step of obtaining the state information of the agent. If the target second node is the disaster node or the target time consumption exceeds the preset specified time length, it is determined that the agent returns to the initial node, and the step of acquiring state information of the agent is returned to. If the target second node is the escape exit node, an escape planning path is acquired, and the escape planning path is a path from the initial node to the escape exit node for the agent. Based on the number of disaster nodes in the escape planning path, the number of dangerous nodes in the escape planning path, the number of congested road sections in the escape planning path, and the length in the escape planning path, a second reward is determined. The second reward and a predicted Q value corresponding to the target action are input into the deep Q network model to update the deep Q network model. A target real Q value of an action of moving from a previous node to a next node in each pair of nodes with edges in the escape planning path is acquired. If the target real Q values corresponding to each pair of nodes with edges in the escape planning path in the escape order successively increase, it is determined that the escape planning path is an optimal path. If the target real Q values corresponding to each pair of nodes with edges in the escape planning path in the escape order do not successively increase, it is determined that the agent returns to the initial node, and the step of acquiring state information of the agent is returned to. If the number of updates of the deep Q network model reaches a preset number of updates, parameters of the deep Q network model are assigned to the target Q network model. If all the optimal paths corresponding to all the nodes have been acquired, it is determined that the double deep Q network model training is completed.

2. The fire emergency evacuation instruction method according to claim 1, characterized by, Further comprising: For each node, a smoke sensor deployed at the node is used to detect whether the node is changed from a disaster occurrence location to a non-disaster occurrence location or from a non-disaster occurrence location to a disaster occurrence location, and if so, it is determined that the fire condition of the node is changed. For each node, a sensor deployed at the node is used to detect a people flow density at the node, and if the people flow density is changed from greater than or equal to a first preset density value to less than the first preset density value or from less than the first preset density value to greater than or equal to the first preset density value, it is determined that the people flow density of the node is changed. For each edge between each pair of nodes, a sensor deployed at a passageway between the pair of nodes is used to detect a people flow density at the passageway, and if the people flow density is changed from greater than or equal to a second preset density value to less than the second preset density value or from less than the second preset density value to greater than or equal to the second preset density value, it is determined that the moving speed of personnel in the passageway between the pair of nodes is changed. For each edge between each pair of nodes, a smoke sensor deployed at a passageway between the pair of nodes is used to detect a smoke concentration of the passageway, and if a smoke concentration level corresponding to the smoke concentration changes, it is determined that the moving speed of personnel in the passageway between the pair of nodes is changed.

3. The fire emergency evacuation instruction method according to claim 2, characterized by, The smoke concentration is a CO concentration, and the method for obtaining the updated moving speed corresponding to the path between each pair of nodes includes: The coefficient a; p is calculated by the formula co is the CO concentration of the passage between the two nodes.​ The updated moving speed is determined as the product of the moving speed before updating and the coefficient a.

4. The fire emergency evacuation instruction method according to claim 1, characterized by, The first reward step of determining the agent moving to the target second node based on the shortest distance between the target second node and the escape exit node, whether the target second node is the disaster node, whether the target second node is the dangerous node, whether the target second node is the escape exit node, and whether the target time consumption of the agent from the initial node to the target second node exceeds a preset specified time length, includes: If the target second node is neither the disaster node nor the escape exit node, and the target time is less than the preset time limit, then by formula... R is calculated t1 Determine R t1 Here, k1 is the first reward; k1 is a preset reward amplification coefficient, where k1 is a positive number greater than 0. This refers to the shortest distance; This refers to the shortest distance between the first node and the escape exit; If the target second node is not an escape exit node and the target time consumption is greater than the preset specified time length, or the target second node is the disaster node, the second preset reward value is determined as the first reward.

5. The fire emergency evacuation instruction method according to claim 4, wherein The second reward step of determining based on the number of disaster nodes in the escape planning path, the number of dangerous nodes in the escape planning path, the number of congested road segments in the escape planning path, and the length of the escape planning path, includes: If the target second node is the escape exit node and the target time consumption is less than or equal to the preset specified time length, R is calculated by the formula , R -1 is determined as the second reward. -1 ​ Wherein, k2 is a preset scaling factor, k3 is a preset time coefficient, and t2 is the total time consumption of the agent from the initial node to the escape exit node; wherein, , wherein is a three-dimensional coordinate of the initial node; is a three-dimensional coordinate of the escape exit node closest to the initial node; is a three-dimensional coordinate of the jth node in the escape planning path; N is the total number of all nodes in the escape planning path; wherein, refers to the number of the dangerous nodes in the escape planning path; refers to the number of the disaster nodes in the escape planning path; wherein, , is a three-dimensional coordinate of the start node of the ith congested road segment; is a three-dimensional coordinate of the end node of the ith congested road segment.

6. The fire emergency evacuation instruction method according to claim 5, wherein The target reward is the first reward or the second reward, and the target reward and the predicted Q value corresponding to the target action are input into the double deep Q network model to update the deep Q network model in the double deep Q network model step, including: If the target second node is neither the disaster node nor the escape exit node, and the target time is less than the preset time limit, then by formula... The true Q value of the agent moving from the first node to the target second node is calculated. wherein, is the real Q value, is the predicted Q value output by the double deep Q network model corresponding to the agent moving from the first node to the target second node; lr is a preset learning rate parameter, and γ is a preset discount factor, is a preset function. if the target second node is the escape exit node and the target time consumption is less than or equal to the preset specified time length, determining R -1 a real Q value of the agent moving from the first node to the target second node. If the target second node is not an escape exit node and the target time consumption is greater than the preset specified time length, or the target second node is the disaster node, the second preset reward value is determined as the real Q value of the agent moving from the first node to the target second node; The deep Q-network model is updated by the formula .

7. The fire emergency evacuation instruction method according to claim 1, wherein The deep Q network model includes a first CSP module, a first CBS module, a second CSP module, a second CBS module, a third CSP module, and a third CBS module connected in sequence. The first CSP module includes a module carrying an unfold function, a fourth CBS module connected to a first output end of the module carrying the unfold function, a first BottleNeck module connected to a second output end of the module carrying the unfold function, and a second BottleNeck module connected to a third output end of the module carrying the unfold function; a fifth CBS module connected to the output end of the fourth CBS module, the output end of the first BottleNeck module, and the output end of the second BottleNeck module, respectively; and a sixth CBS module connected to the fifth CBS module. The first CBS module includes a convolution layer, a batch normalization layer connected to the convolution layer, and an activation function layer connected to the batch normalization layer. The first CBS module includes a convolution layer, a batch normalization layer connected to the convolution layer, and an activation function layer connected to the batch normalization layer. The first CSP module, the second CSP module, and the third CSP module are of the same structure; the first CBS module, the second CBS module, the third CBS module, the fourth CBS module, the fifth CBS module, and the sixth CBS module are of the same structure.

8. A fire emergency evacuation indicating device applied to a large indoor scene, characterized in that, Comprise: The first acquisition module is used for acquiring a personnel evacuation network flow model, the personnel evacuation network flow model comprising a plurality of nodes and edges between two nodes; the edges between two nodes represent a passage between two nodes for personnel walking; the plurality of nodes comprise an escape exit node, a normal node, and a staircase node; The second acquisition module is used for acquiring a distance matrix corresponding to the personnel evacuation network flow model; the distance matrix comprising distances between all nodes with edges in the personnel evacuation network flow model; The third acquisition module is used for acquiring a density matrix corresponding to the personnel evacuation network flow model if a fire is detected; the density matrix comprising a person flow density of all nodes in the personnel evacuation network flow model and a person flow density of a passage between two nodes with edges; The first determination module is used for determining a dangerous node and a disaster node from the plurality of nodes; the dangerous node being a node with a person flow density greater than or equal to a first preset density value, and the disaster node being a node where a fire occurs; The second determination module is used for determining an edge between two nodes as a congested section if a person flow density of the passage represented by the edge between two nodes is greater than or equal to a second preset density value; The first update module is used for updating the dangerous node, the disaster node, and the congested section to the personnel evacuation network flow model; The fourth acquisition module is used for acquiring a speed matrix corresponding to the personnel evacuation network flow model based on the density matrix; the speed matrix comprising a moving speed of personnel in a passage between all nodes with edges in the personnel evacuation network flow model; The fifth acquisition module is used for acquiring an evacuation time consumption matrix corresponding to the personnel evacuation network flow model based on the distance matrix and the speed matrix; The sixth acquisition module is used for inputting the personnel evacuation network flow model, the distance matrix, the density matrix, the speed matrix, and the evacuation time consumption matrix to a pre-constructed evacuation path prediction model, and acquiring a first escape path corresponding to each node through the evacuation path prediction model; The first sending module is used for sending the first escape path corresponding to each node to a terminal corresponding to the node for escaping; The seventh acquisition module is used for obtaining an updated personnel evacuation network flow model, an updated density matrix, an updated speed matrix, and an updated evacuation time consumption matrix if a person flow density corresponding to any node, a person flow density of a passage between any two nodes, a moving speed of personnel in a passage between any two nodes, or a fire condition of any node is changed. The eighth obtaining module is configured to input the updated personnel evacuation network flow model, the distance matrix, the updated density matrix, the updated speed matrix, and the updated evacuation time consumption matrix into the evacuation path prediction model, and obtain a second escape path through the evacuation path prediction model; The second sending module is configured to send the second escape path corresponding to each node to a terminal for escape corresponding to the node; The evacuation path prediction model is a double deep Q network model, and a method for training the double deep Q network includes: Obtaining a sample density matrix corresponding to the personnel evacuation network flow model; the sample density matrix includes the passenger flow density of all nodes in the personnel evacuation network flow model and the passenger flow density between two nodes having an edge; Based on the sample density matrix, a sample speed matrix corresponding to the personnel evacuation network flow model is obtained; the sample speed matrix includes the moving speed of personnel between two nodes having an edge in the personnel evacuation network flow model; Based on the sample distance matrix and the sample speed matrix, a sample evacuation time consumption matrix corresponding to the personnel evacuation network flow model is obtained; Determining a dangerous node and a disaster node from the plurality of nodes; the dangerous node is a node with a passenger flow density greater than or equal to a first preset density value, and the disaster node is a node where a fire occurs; If the passenger flow density of the passage represented by the edge between the two nodes is greater than or equal to a second preset density value, the edge between the two nodes is determined to be a congested section; Updating the dangerous node, the disaster node, and the congested section to the personnel evacuation network flow model; Obtaining state information of an agent, the state information including the time consumption of moving from a first node to each second node; Inputting the personnel evacuation network flow model, the distance matrix, the sample density matrix, the sample speed matrix, the sample evacuation time consumption matrix, and the state information into the double deep Q network model, and outputting a plurality of action corresponding predicted Q values through the double deep Q model; the double deep Q network model includes a deep Q network model and a target Q network model; the action is the action of moving the agent from the first node to the second node; Obtaining a random number; If the random number is greater than a preset threshold, a target action is randomly selected from the plurality of actions, and the second node in the target action is a target second node; If the random number is less than or equal to the preset threshold, a target action corresponding to the maximum predicted Q value is selected from the plurality of actions; the second node in the target action is a target second node; determine a first reward of the agent moving to the target second node based on a shortest distance between the target second node and the escape exit node, whether the target second node is the disaster node, whether the target second node is the dangerous node, whether the target second node is the escape exit node, and whether a target time consumption of the agent from an initial node to the target second node exceeds a preset specified time length; input the first reward and a predicted Q value corresponding to the target action into the double deep Q network model to update the deep Q network model in the double deep Q network model; if the target second node is not the disaster node and not the escape exit node, and the target time consumption does not exceed the preset specified time length, return to the step of obtaining the state information of the agent; if the target second node is the disaster node or the target time consumption exceeds the preset specified time length, determine that the agent returns to the initial node, and return to the step of obtaining the state information of the agent; if the target second node is the escape exit node, obtain an escape planning path, the escape planning path being a path of the agent from the initial node to the escape exit node; determine a second reward based on a number of disaster nodes in the escape planning path, a number of dangerous nodes in the escape planning path, a number of congested road segments in the escape planning path, and a length of the escape planning path; input the second reward and the predicted Q value corresponding to the target action into the deep Q network model to update the deep Q network model; obtain a target real Q value of an action of moving from a previous node to a next node in each pair of nodes with an edge in the escape planning path; if the target real Q values corresponding to each pair of nodes with an edge in the escape planning path according to an escape order are sequentially increased, determine that the escape planning path is an optimal path; if the target real Q values corresponding to each pair of nodes with an edge in the escape planning path according to the escape order are not sequentially increased, determine that the agent returns to the initial node, and return to the step of obtaining the state information of the agent; if a number of times of updating of the deep Q network model reaches a preset number of times of updating, assign parameters of the deep Q network model to the target Q network model; if all optimal paths corresponding to all nodes have been obtained, determine that training of the double deep Q network model is completed.

9. A computer program product, characterised in that, The computer readable instructions, when executed on an electronic device, cause the electronic device to implement the method for fire emergency evacuation indication in a large indoor scene according to any one of claims 1 to 7.

Images