Indoor escape route optimization method and system

Abstract

The present application relates to the technical field of path planning, in particular to an indoor escape route optimization method and system. A hierarchical road network model is constructed; when a danger occurs, Internet of Things monitoring data is received to simulate danger diffusion and predict a danger affected area; crowd distribution information is obtained by combining cameras and Wi-Fi probes to obtain a predicted personnel distribution map in a preset time zone; the hierarchical road network model is used to optimize escape routes in the preset time zone according to the predicted danger affected area and the predicted personnel distribution map, and an optimal escape plan is output; a plurality of optimal escape routes are respectively sent to a plurality of associated escape personnel through a device terminal, and route dynamic updating is performed according to danger perception data until the personnel are evacuated. The optimal escape plan that is suitable for different danger scenarios and personnel distribution situations can be quickly and accurately planned.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and in particular to a method and system for optimizing indoor escape routes. Background Technology

[0002] In the traditional fields of building safety management and emergency evacuation, early methods relied primarily on simple evacuation signs and manual guidance. With technological advancements, emergency systems based on static maps and pre-set evacuation plans have gradually emerged. While these systems can provide evacuation guidance to some extent, they have numerous limitations. Regarding indoor road network construction, previous methods may have simply drawn main passageways without fully considering the complex topology and functional zoning within buildings, resulting in inaccurate and inefficient route planning. For hazard prediction, early methods relied heavily on experience-based judgments, lacking real-time and accurate data support, making it difficult to effectively simulate the dynamic spread of hazards. In crowd monitoring, traditional methods struggle to comprehensively and in real-time grasp the distribution and movement of people, failing to provide sufficiently detailed information for evacuation plan development. Furthermore, escape route planning often considers only a single factor, such as the shortest distance, neglecting crucial elements such as safety and exit load balancing. Moreover, once escape routes are determined using traditional systems, they are difficult to dynamically adjust based on actual hazard changes and population flow, resulting in a lack of real-time responsiveness and specificity in escape routes. Summary of the Invention

[0003] This invention addresses the problem of insufficient real-time performance and specificity in existing escape route planning technologies by providing an indoor escape route optimization method and system.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0005] In a first aspect, the present invention provides an indoor escape route optimization method, comprising: constructing a hierarchical road network model based on indoor map vector data and road network topology of the target building, and marking key escape nodes; receiving IoT monitoring data to simulate the spread of danger when danger occurs, and outputting the predicted danger-affected area within a preset time zone; combining cameras and Wi-Fi probes to obtain crowd distribution information and predicting crowd movement, and outputting the predicted personnel distribution map within the preset time zone; using the hierarchical road network model, with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, optimizing escape routes within the preset time zone based on the predicted danger-affected area and the predicted personnel distribution map, and outputting the optimal escape plan, wherein the optimal escape plan includes several optimal escape routes; sending the several optimal escape routes to several associated escape personnel through a device terminal, and dynamically updating the routes according to the danger perception data until the personnel have been evacuated.

[0006] Optionally, receiving IoT monitoring data to simulate hazard diffusion and outputting a predicted hazard-affected area within a preset time zone includes: configuring an IoT monitoring array, wherein the IoT monitoring array includes at least a temperature sensor group, a smoke detector group, and a vibration sensor group, and each sensor is identified with location information; receiving IoT monitoring data through the IoT monitoring array when a hazard occurs, wherein the IoT monitoring data includes a temperature distribution sequence, a smoke concentration distribution sequence, and a vibration characteristic distribution sequence; performing hazard diffusion simulation based on the temperature distribution sequence, the smoke concentration distribution sequence, and the vibration characteristic distribution sequence, and outputting a predicted hazard-affected area within a preset time zone.

[0007] The process includes: simulating hazard diffusion based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, and outputting a predicted hazard impact area within a preset time zone. This includes: identifying hazard types based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence to determine the current hazard event type, wherein the hazard event type includes at least fire and earthquake; matching the current hazard event type to obtain an adapted hazard diffusion simulator, wherein the hazard diffusion simulator is constructed based on a deep neural network and trained to convergence using sample data; and using the adapted hazard diffusion simulator to simulate hazard diffusion based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, and outputting a predicted hazard impact area within a preset time zone.

[0008] The process involves combining cameras and Wi-Fi probes to acquire crowd distribution information, predicting crowd movement, and outputting a predicted personnel distribution map within a preset time zone. This includes: acquiring crowd distribution information at multiple consecutive monitoring time points using cameras and Wi-Fi probes to construct a crowd distribution feature sequence; collecting a sample crowd distribution feature sequence set based on the target building's historical escape records, constrained by the current dangerous event type, and obtaining a sample crowd distribution map set within the historical time zone, where the time interval of the historical time zone is the same as the preset time zone; training a deep neural network to convergence using the sample crowd distribution feature sequence set and the sample crowd distribution map set to obtain a crowd movement predictor, and analyzing the crowd distribution feature sequence to obtain a predicted personnel distribution map within the preset time zone.

[0009] Optionally, using the hierarchical road network model, with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, the escape routes within a preset time zone are optimized based on the predicted danger zone and predicted personnel distribution map, and the optimal escape plan is output. This includes: rendering the predicted danger zone and predicted personnel distribution map onto the hierarchical road network model to construct a personnel escape simulation space; within the personnel escape simulation space, using the predicted danger zone as a passage restriction condition and combining it with the predicted personnel distribution map, simulating escape routes and randomly generating multiple initial escape plans, wherein each initial escape plan includes several escape routes associated with several evacuees; simulating personnel escape based on the multiple initial escape plans, and evaluating the fitness of multiple plans with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance; and optimizing escape routes within a preset time zone based on the fitness of the multiple plans until the optimization converges, and outputting the optimal escape plan.

[0010] The process involves simulating evacuation based on multiple initial escape plans, evaluating the fitness of each plan with the objectives of minimizing evacuation time, maximizing path safety levels, and optimizing exit load balancing. This includes: randomly selecting a first initial escape plan to simulate evacuation within a preset time zone; outputting several shortest simulated evacuation times as a first evacuation duration set; outputting several straight-line distances to the danger zone as a first danger distance set; and outputting the number of simulated personnel within multiple escape exit ranges as a first exit load characteristic. The first evacuation duration set and the first danger distance set are summed to obtain a first total evacuation time and a first total danger distance. Based on the first exit load characteristic, the standard deviation of the number of simulated personnel is calculated to obtain a first exit load balancing coefficient. Based on preset index weights, the first total evacuation time, the first total danger distance, and the first exit load balancing coefficient are weighted and summed to obtain the first plan fitness. Multiple plan fitnesss are then analyzed sequentially, with plan fitness being negatively correlated with total evacuation time, total danger distance, and exit load balancing coefficient.

[0011] The process of optimizing escape routes within a preset time zone based on the fitness of multiple escape routes until convergence is achieved, and then outputting the optimal escape route, includes: First, based on the fitness of the multiple escape routes, an initial solution sequence is obtained by arranging the initial escape route solutions in descending order of fitness; the first K solutions of the initial solution sequence are designated as optimal solutions, and the last J solutions as inferior solutions, and random iso-clustering is performed on the J inferior solutions centered on the K optimal solutions to obtain K solution sets, where J is a multiple of K, and P is an integer greater than 9; within each solution set, the optimal solution is used as the optimization direction, and the optimal escape route is optimized according to the optimization... The step size is adjusted to obtain K updated solution sets. If the updated inferior solution involves the predicted danger zone, any initial escape plan is randomly selected for replacement. The K updated solution sets are identified. In each updated solution set, if the fitness of the inferior solution is greater than or equal to the fitness of the superior solution, the inferior solution replaces the superior solution. Iterative optimization continues until a preset number of optimization attempts is reached. K current updated solution sets are output, and the solution set with the largest sum of fitness among the K current updated solution sets is set as the optimal solution set. The superior solution of the optimal solution set is set as the optimal escape plan.

[0012] Secondly, the present invention provides an indoor escape route optimization system, comprising:

[0013] The layered road network construction module is used to build a layered road network model based on the indoor map vector data and road network topology of the target building, and to mark key escape nodes.

[0014] The hazard diffusion simulation module is used to receive IoT monitoring data to simulate hazard diffusion when a hazard occurs, and output the predicted hazard impact area within a preset time zone;

[0015] The personnel distribution prediction module is used to combine cameras and Wi-Fi probes to obtain crowd distribution information, predict crowd movement, and output a predicted personnel distribution map within a preset time zone.

[0016] The escape plan generation module is used to optimize the escape routes within a preset time zone based on the predicted danger zone and the predicted personnel distribution map, with the goal of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, using the hierarchical road network model. The optimal escape plan includes several optimal escape routes.

[0017] The escape route sending module is used to send the several optimal escape routes to several associated escape personnel through the device terminal, and to dynamically update the routes according to the danger perception data until the personnel have been evacuated.

[0018] By implementing this invention, a hierarchical road network model can be constructed based on the indoor map vector data and road network topology of the target building, and key escape nodes can be marked, providing a basic framework for subsequent escape route planning and transforming the complex spatial structure inside the building into an orderly and usable network.

[0019] By implementing this invention, it is possible to receive IoT monitoring data to simulate the spread of danger when danger occurs, and output the predicted danger-affected area within a preset time zone. This allows people inside the building to know in advance the range of potential danger spread so as to avoid dangerous areas in time, which greatly improves the safety of personnel escape and provides key risk information for escape route planning, making the planning more targeted and scientific.

[0020] By implementing this invention, it is possible to combine cameras and Wi-Fi probes to obtain crowd distribution information, predict crowd movement, and output a predicted personnel distribution map within a preset time zone. This helps to understand the real-time distribution and flow trends of people in the building, avoid excessive crowd gathering or congestion in escape route planning, achieve more reasonable personnel evacuation, improve evacuation efficiency, and reduce the dangers caused by crowd congestion.

[0021] By implementing this invention, it is possible to utilize the hierarchical road network model to minimize evacuation time, maximize path safety levels, and optimize exit load balance. Based on the predicted danger zone and predicted personnel distribution map, the escape routes within a preset time zone are optimized, and the optimal escape plan is output. The optimal escape plan includes several optimal escape routes, ensuring that the planned escape routes can guarantee rapid and safe evacuation of personnel while maintaining a relatively balanced flow of people at each exit. This avoids evacuation obstruction due to excessive load at a particular exit, thereby improving the reliability and effectiveness of the overall evacuation plan.

[0022] By implementing this invention, several optimal escape routes can be sent to several associated escapees via a device terminal, and the routes can be dynamically updated according to the danger perception data until the personnel have been evacuated. This provides accurate and personalized escape guidance to the escapees, and the routes can be adjusted in a timely manner according to the real-time changes in the danger situation, ensuring that the escape routes are always in the optimal state and increasing the probability of successful escape.

[0023] In summary, by implementing this invention, it is possible to quickly and accurately plan the optimal escape scheme that adapts to different dangerous scenarios and personnel distribution. Attached Figure Description

[0024] Figure 1 A flowchart illustrating the indoor escape route optimization method provided by this invention;

[0025] Figure 2 This is a schematic diagram of the indoor escape route optimization system provided by the present invention.

[0026] In the attached diagram, the components represented by each number are as follows:

[0027] The module includes: a layered road network construction module 11, a hazard diffusion simulation module 12, a personnel distribution prediction module 13, an escape plan generation module 14, and an escape route sending module 15. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0030] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0031] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for optimizing indoor escape routes, including:

[0032] S100: Based on the indoor map vector data and road network topology of the target building, construct a hierarchical road network model and mark key escape nodes;

[0033] S200: When a hazard occurs, it receives IoT monitoring data to simulate the spread of the hazard and outputs the predicted hazard impact area within a preset time zone;

[0034] S300: Combines camera and Wi-Fi probe to obtain crowd distribution information, predicts crowd movement, and outputs a predicted personnel distribution map within a preset time zone;

[0035] S400: Using the hierarchical road network model, with the goals of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, the escape routes within a preset time zone are optimized based on the predicted danger zone and predicted personnel distribution map, and the optimal escape plan is output, wherein the optimal escape plan includes several optimal escape routes.

[0036] S500: The device terminal sends the several optimal escape routes to several associated escape personnel, and dynamically updates the routes according to the danger perception data until the personnel have been evacuated.

[0037] In this embodiment of the application, to optimize indoor escape routes, it is first necessary to construct a hierarchical road network model of the target building, that is, to construct a hierarchical road network model based on the indoor map vector data and road network topology of the target building as described in step S100, and to mark key escape nodes.

[0038] For example, the indoor map vector data can be extracted from the building BIM model or CAD drawings, such as the coordinates and geometry of walls, doors, windows, and corridors, with an accuracy requirement of ≥10cm, as well as the height of each floor, the location of stairwells, and vertical connectivity. Furthermore, the data can be calibrated on-site using laser scanning or indoor positioning systems, such as UWB, to correct deviations between the drawings and the actual layout, such as narrowing of passageways caused by temporary partitions or equipment placement.

[0039] The aforementioned road network topology transforms passable areas, such as corridors and staircases, into "edges" in graph theory. Edge attributes include length (affecting evacuation time), width (affecting passage capacity, e.g., a 1.2m wide corridor allowing 2 people to pass per second), and passage restrictions (e.g., one-way doors, access control status). Furthermore, corridor intersections, stairwell entrances, and safety exits can be transformed into "nodes," which can be classified as ordinary nodes or critical nodes. Ordinary nodes include corridor intersections; critical nodes include safety exits, stairwell entrances, etc.

[0040] Next, the road network topology and indoor map vector data need to be merged. Specifically, first, geometric elements such as walls, corridors, doors, and windows are extracted from the indoor map vector data and converted into point and line coordinates. Then, the road network topology is abstracted into nodes and edges of a graph. Nodes correspond to passage intersections, stairwells, etc., and edges correspond to passable paths. Through coordinate matching, key locations in the vector data, such as stairwell entrances, are mapped to topological nodes, and corridors are converted into connecting edges, while assigning attributes such as edge length and width. During the merging process, the planar layout of the vector data is matched to the topological graph structure on a floor-by-floor basis, ensuring that node coordinates are consistent with the vector data and that the edge orientation matches the actual passageway orientation. This ultimately forms a layered road network model with spatial coordinates and topological relationships.

[0041] Furthermore, based on the above data, sub-layers can be created according to natural floors, such as F1, F2, etc., with each layer having its own road network sub-graph. Floors are connected by stair nodes: for example, stair nodes on floor F1 and F2 are connected by "vertical edges," with edge attributes recording information such as the number of steps and ascent time, such as 0.5 seconds per step. Depth-first search (DFS) can also be used to verify road network connectivity, ensuring that any room node is reachable from the safety exit node. Furthermore, Dijkstra's algorithm can be used to calculate the shortest paths between nodes, generating a distance matrix to provide foundational data for subsequent escape simulations.

[0042] Key escape points that need to be marked include legally designated safety exits, smoke-proof stairwells, entrances to enclosed stairwells, entrances to refuge floors, and fire elevator lobbies, among others.

[0043] The specific methods for constructing the aforementioned hierarchical road network model can all be implemented using existing technologies, and will not be elaborated here.

[0044] In step S200 of this application embodiment, receiving IoT monitoring data to simulate hazard diffusion and outputting the predicted hazard-affected area within a preset time zone includes:

[0045] Configure an IoT monitoring array, wherein the IoT monitoring array includes at least a temperature sensor group, a smoke detector group, and a vibration sensor group, and each sensor is identified with location information;

[0046] When a danger occurs, IoT monitoring data is received through the IoT monitoring array, wherein the IoT monitoring data includes temperature distribution sequence, smoke concentration distribution sequence and vibration characteristic distribution sequence;

[0047] Based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, a hazard diffusion simulation is performed, and the predicted hazard impact area within a preset time zone is output.

[0048] Optionally, the temperature sensor group can use an infrared temperature sensor with an accuracy of ±0.5℃. The infrared temperature sensor can be combined with a thermocouple sensor and deployed in locations such as corridor ceilings, stairwells, and room entrances, with a spacing of ≤10 meters, to ensure that the temperature field is covered without any blind spots.

[0049] The smoke detector array can use photoelectric smoke sensors with a sensitivity of 0.15-0.5dB / m. They can be installed under the ceiling to cover all public areas and rooms, with a spacing of ≤8 meters. They should be placed in ventilation openings, cable shafts and other places where smoke is likely to accumulate.

[0050] The vibration sensor array can use an accelerometer with a range of ±50g, fixed to load-bearing walls, columns, and stair structures, to monitor vibrations caused by earthquakes or structural collapses. The spacing can be adjusted according to the building structure density, such as one sensor every 20 meters in high-rise buildings.

[0051] Furthermore, each sensor needs to be assigned a unique ID, such as T-101 representing temperature sensor number 1 on the 1st floor. Based on the building's BIM model or CAD drawings, the sensor's coordinates (X, Y, Z) are entered into the management system to form a mapping table of "Sensor ID - 3D coordinates - installation location description". For example, for a certain sensor, its information identifier could be Sensor ID: S-203, meaning smoke detector number 3 on the 2nd floor, with coordinates: X = 15.2m, Y = 22.8m, Z = 4.5m, corresponding to the ceiling of the 2nd floor corridor.

[0052] Finally, IoT monitoring data needs to be received through the IoT monitoring array. Specifically, the sensors can be connected to the edge gateway via RS-485, LoRa, or Wi-Fi modules. The gateway collects sensor data in real time with a sampling frequency of ≥10Hz. Then, wired and wireless dual-link backup is used to transmit data to ensure that the network is not interrupted in dangerous scenarios.

[0053] The IoT monitoring data includes temperature distribution sequences, smoke concentration distribution sequences, and vibration characteristic distribution sequences. The temperature distribution sequence is grouped by sensor ID, recording the temperature values ​​at each point in real time to form a time series. For example, the temperature changes of T-101 from 10:30:00 to 10:30:30 are 25.8℃, 26.2℃, and 26.5℃. Similarly, the smoke concentration distribution sequence is generated by sorting the concentration values ​​of each smoke detector (in %LEL) by time to create a concentration-time series. The vibration sensor collects acceleration signals, and frequency domain features, such as the principal vibration frequency and amplitude, are extracted using FFT (Fast Fourier Transform) to form a feature vector sequence, such as [frequency 1, amplitude 1, frequency 2, amplitude 2...].

[0054] In step S200 of this embodiment, a hazard diffusion simulation is performed based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, and the predicted hazard-affected area within a preset time zone is output, including:

[0055] Hazard type identification is performed based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence to determine the current hazardous event type, wherein the hazardous event type includes at least fire and earthquake;

[0056] An adaptive hazard diffusion simulator is obtained based on the matching of the current hazard event type. The hazard diffusion simulator is constructed based on a deep neural network and trained to convergence using sample data.

[0057] Using the adaptive hazard diffusion simulator, the hazard diffusion simulation is performed based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, and the predicted hazard impact area within the preset time zone is output.

[0058] In this embodiment of the application, to simulate the diffusion of danger based on the temperature distribution sequence, smoke concentration distribution sequence and vibration characteristic distribution sequence, it is first necessary to extract temperature sequence features, smoke concentration features and vibration features based on the temperature distribution sequence, smoke concentration distribution sequence and vibration characteristic distribution sequence.

[0059] Among these features, temperature sequence characteristics include calculating the rate of temperature rise (ΔT / Δt) and spatial temperature gradient, and extracting the spatiotemporal distribution characteristics of abnormal temperature rise areas, such as a temperature rise of ≥10℃ in a certain area within 5 minutes. The spatial temperature gradient refers to the temperature difference between adjacent sensors. Smoke concentration characteristics involve analyzing the rate of concentration rise and spatial diffusion patterns, such as the concentration of smoke detected simultaneously exceeding 5% LEL at three or more smoke detectors on a certain floor. Vibration characteristics are determined through frequency domain analysis of vibration sequences, such as analyzing whether the dominant frequency is within the seismic characteristic frequency band of 0.1–10Hz, and amplitude abrupt change points, such as whether the instantaneous acceleration is ≥0.2g, to determine the vibration type.

[0060] Hazard type identification can be achieved using a hazard type classification model. This hazard type classification model can use a hybrid architecture of convolutional neural network (CNN) + long short-term memory network (LSTM), and its main structure consists of three layers: input layer, feature extraction layer, and classification layer.

[0061] Its input is used to transform temperature, smoke, and vibration sequences into multi-dimensional feature vectors, such as the time-space matrix of the temperature sequence and the vibration spectrum. The feature extraction layer uses CNN to extract spatial features, such as the shape of the temperature anomaly area, and uses LSTM to capture the dynamic changes of the time series, such as the concentration growth trend over time. The classification layer is used to output the probability of hazard type through the fully connected layer, such as the probability of hazard type being fire, earthquake, other disasters, or false alarm, and sets a threshold, such as determining the corresponding hazard type when the probability of hazard type is ≥0.7.

[0062] By inputting the aforementioned temperature sequence characteristics, smoke concentration characteristics, and vibration characteristics, the corresponding output will be the type of hazardous event, such as fire or earthquake.

[0063] Next, an adapted hazard diffusion simulator needs to be obtained based on the current hazard event type. The hazard diffusion simulator is built based on a deep neural network and trained to convergence using sample data.

[0064] The hazard spread simulator may include two parts: a fire spread simulator and an earthquake impact simulator.

[0065] The fire spread simulator is built by integrating physical models (such as the FDS fire dynamics model) with deep learning. By inputting temperature and smoke concentration, it can output the spatiotemporal distribution of flame spread and smoke diffusion.

[0066] The earthquake impact simulator is built using a graph neural network (GNN). It uses the building structure topology, i.e., a layered road network model, as a graph structure. By inputting vibration feature sequences, it can predict structural weak points, such as wall cracks and the probability of staircase collapse.

[0067] The training samples for the fire spread simulator are collected historical fire case data, such as temperature-time curves and smoke spread videos of a shopping mall fire. Combined with FDS simulation, a multi-scenario training set is generated, which includes different fire source locations and ventilation conditions.

[0068] The training samples for the earthquake impact simulator are derived from earthquake wave simulation data, such as the vibration acceleration at various points on a building under different magnitudes, combined with the hazardous areas marked by structural mechanics analysis, such as the locations where the vibration of load-bearing columns exceeds the limit.

[0069] Then, by combining manual and algorithmic methods, dangerous areas such as flame spread outlines and structural damage areas are marked in the above physical simulation results or monitoring videos and converted into polygon coordinates or mesh masks as supervisory data.

[0070] The training samples for the fire spread simulator shall be no less than 1,000 scenarios, and the training samples for the earthquake impact simulator shall be no less than 500 scenarios. Each scenario shall contain complete sensor sequence data and corresponding hazard area annotations.

[0071] The core architecture of the fire spread simulator includes an LSTM layer and a 3D convolutional layer. The LSTM layer contains 64 neurons and is used to process the trends of temperature and smoke concentration changes over time. The 3D convolutional layer is used to capture the spatial diffusion patterns of temperature and smoke concentration. Its loss function combines binary cross-entropy (BCE) and the Dice coefficient to optimize the prediction accuracy for discrete regions.

[0072] The core architecture of the earthquake impact simulator consists of a graph structure and a GNN layer. First, the nodes and edges of the hierarchical road network model form the graph structure. Node features include building material parameters, such as wall compressive strength, while edge features include connection stiffness. The GNN layer employs the GraphSAGE algorithm to aggregate the vibration features of adjacent nodes and predict the probability of node damage. Its loss function can use mean squared error (MSE) + focal loss to balance the prediction accuracy between normal and hazardous areas.

[0073] For training the hazard spread simulator, the training batch size for the fire simulator was set to 32, and the training batch size for the earthquake simulator was set to 16. The optimizer used was AdamW, with an initial learning rate of 1e-4, which decayed by 0.8 times every 50 training rounds. The training rounds were preset to 200, and the early termination condition was that the validation set loss did not decrease for 10 consecutive rounds.

[0074] The evaluation metric can be the overlap rate between the predicted hazard area and the actual hazard area. The hazard diffusion simulator is considered converged when the overlap rate is ≥0.7 across 10 consecutive predictions. The overlap rate is calculated as follows: Overlap Rate = Predicted Hazard Area / Actual Hazard Area.

[0075] Finally, the real-time monitored temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence are standardized, that is, normalized to the [0,1] interval. According to the time window, such as 5 minutes, the sequence is sliced ​​and processed. The result is input into the corresponding simulator and the predicted danger zone within the preset time zone can be output. For example, the danger zone in the next 5 minutes or 10 minutes can be output.

[0076] In step S300 of this embodiment, crowd distribution information is obtained by combining a camera and a Wi-Fi probe, and crowd movement prediction is performed to output a predicted personnel distribution map within a preset time zone, including:

[0077] By combining cameras and Wi-Fi probes to acquire crowd distribution information at multiple consecutive monitoring time points, a crowd distribution feature sequence is constructed.

[0078] Based on the historical escape records of the target building, and constrained by the current dangerous event type, a sample population distribution feature sequence set is collected, and a sample population distribution map is obtained within the historical time zone, thus obtaining a sample population distribution map set. The time interval of the historical time zone is the same as the preset time zone.

[0079] Using the sample population distribution feature sequence set and sample population distribution map set, a deep neural network is trained until convergence to obtain a population movement predictor, and a predicted population distribution map within a preset time zone is obtained based on the population distribution feature sequence analysis.

[0080] In this embodiment of the application, the combination of cameras and Wi-Fi probes to obtain crowd distribution information and to predict crowd movement is intended to provide a key basis for escape route planning in order to avoid route congestion.

[0081] Among them, the cameras used to obtain crowd distribution information at multiple consecutive monitoring time points can be wide-angle smart cameras, deployed in key areas such as corridor intersections, stairwells, and lobbies, covering all personnel activity spaces, with a frame rate of 10fps, supporting real-time video streaming and human detection functions; while the Wi-Fi probes can be probe devices that support the IEEE802.11ac protocol, installed on the ceiling or wall, with a coverage radius of ≤30 meters, scanning the MAC addresses of surrounding Wi-Fi devices in real time, obtaining signal strength (RSSI) and connection status, with a sampling frequency of 1 time / second.

[0082] Furthermore, object detection algorithms such as YOLOv8 can be used to identify human targets in video captured by cameras in real time, calculate the population density in each area (e.g., people per square meter), and generate a two-dimensional heatmap coordinate system (e.g., (X, Y, people)). Perspective transformation is also needed to correct camera view distortion and ensure that the location coordinates are consistent with the actual building coordinate system. For Wi-Fi probe data, the mapping relationship between RSSI values ​​and signal source locations, such as triangulation, can be used to estimate the location range of people carrying Wi-Fi devices, with an error range of ≤5m.

[0083] Furthermore, by time windows, such as 10 seconds, the camera and Wi-Fi data are aggregated to form a continuous feature sequence, forming a feature vector for each time point: [number of people in area A, number of people in area B, ..., probability of movement direction of people in each area]. For example, at time t1, the feature vector could be [15 people (hall), 8 people (corridor), movement direction: from hall to corridor with a probability of 60%].

[0084] The training data for the crowd movement predictor is a set of sample crowd distribution feature sequences. Specifically, this data can be obtained by selecting scenarios from historical escape records that match the current type of dangerous event; for example, a fire scenario would require data from 50 fire drills. Each scenario contains a continuous crowd distribution feature sequence of at least 30 minutes. After being segmented according to a preset time zone (e.g., 10 minutes), at least 500 sample pairs are generated, forming the sample crowd distribution feature sequence set. The samples in this set are then divided into a training set and a validation set in an 8:2 ratio for training the crowd movement predictor.

[0085] The input features of the crowd movement predictor are the sequence of crowd distribution features, and the output features are the predicted distribution map of people. It can be built using a hybrid architecture of LSTM+CNN, where the LSTM layer is used to capture the dynamics of the time series and the CNN layer is used to extract the spatial distribution pattern.

[0086] Specifically, the crowd movement predictor has a 5-layer structure: an input layer, an LSTM layer, a CNN layer, an upsampling layer, and an output layer. The input layer receives the crowd distribution feature sequence; the LSTM layer contains 128 neurons to capture crowd movement trends in the time series, such as the increase or decrease in the number of people in a certain area over time; the CNN layer uses 3×3 convolutional kernels with 64 channels to extract spatial crowd clustering patterns, such as the spatial feature of congestion at stairwells; and the upsampling layer is used to increase the resolution of the feature map to match the population distribution. Figure 1 For example, a 1m×1m grid; the output layer is used to output a predicted personnel distribution map within a preset time zone through a fully connected layer.

[0087] For training the crowd movement predictor, the training batch size is 32. The initial learning rate is set to 1×10⁻⁶. -4 Training is performed at a rate of 0.8x every 50 epochs; the optimizer used is Adam; the loss function is mean squared error. The training epochs are preset to 200, and an early stopping mechanism is employed to avoid overfitting.

[0088] The crowd movement predictor is considered converged when the mean squared error (MSE) on the validation set is ≤0.05 and the intersection-over-union (IoU) ratio between the predicted and actual population distribution maps is ≥0.85. By inputting a population distribution feature sequence, the predicted population distribution map within a preset time zone can be obtained.

[0089] In step S400 of this embodiment, the hierarchical road network model is used to optimize evacuation time, maximize path safety level, and achieve optimal exit load balancing. Based on the predicted danger zone and predicted personnel distribution map, escape routes within a preset time zone are optimized, and the optimal escape plan is output, including:

[0090] The predicted danger zone and predicted personnel distribution map are rendered onto the hierarchical road network model to construct a personnel escape simulation space.

[0091] Within the simulated personnel escape space, the predicted danger zone is used as the access restriction condition. Combined with the predicted personnel distribution map, the escape route is simulated, and multiple initial escape plans are randomly generated. Each initial escape plan includes several escape routes associated with several escaped personnel.

[0092] Based on the multiple initial escape plans, personnel escape simulations were conducted, and the fitness of the multiple plans was evaluated with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance.

[0093] Based on the fitness of the multiple schemes, the escape route within the preset time zone is optimized until the optimization converges, and the optimal escape scheme is output.

[0094] In this embodiment, the predicted danger zone and predicted population distribution map are rendered onto the layered road network model to construct a population escape simulation space. Specifically, the predicted danger zone, such as a fire smoke-covered area or an earthquake structural collapse area, can be superimposed onto the layered road network model in polygon coordinates and marked as a "no-entry zone". The predicted population distribution map is converted into grid density data, such as 5 people per square meter in a certain area, and mapped to road network nodes as a population density constraint. Furthermore, the above data can be visualized, for example, the no-entry zone can be marked with a red semi-transparent layer, and road sections with high population density can be represented by yellow gradient lines, so that the simulation space can intuitively reflect the danger zone and population distribution.

[0095] Furthermore, it is necessary to use the predicted danger zone as a passage restriction condition, and combine it with the predicted personnel distribution map to simulate escape routes, randomly generating multiple initial escape plans. For example, the edges of the road network within the aforementioned restricted area, such as corridors and staircases, are set as impassable. The nodes of the road network within the aforementioned restricted area, such as room doors and stairwells, are set as blocked. Then, based on the predicted personnel distribution map, dynamic passage capacity is set for each road network edge; for example, if a corridor currently has a high density, the passage speed is reduced by 30%.

[0096] Next, escape route simulation needs to be performed using the predicted personnel distribution map to randomly generate multiple initial escape plans. All initial escape plans must avoid restricted areas. Each initial plan includes: personnel grouping information (area + number of people); a set of escape routes for each group (e.g., 30 people in the 1st-floor office area corresponding to 30 routes); and exit allocation results (the expected number of people to be received at each exit). For example, a specific escape plan could be for 30 people on the west side of the 3rd floor: 20 people go from the S2 staircase to the west exit on the 2nd floor, and 10 people go from the S2 staircase to the 4th floor, then to the east exit on the 4th floor.

[0097] In step S400 of this application embodiment, the characteristic is that personnel escape simulation is performed based on the multiple initial escape plans, and the fitness of multiple plans is evaluated with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, including:

[0098] Randomly select the first initial escape plan to simulate the escape of personnel within a preset time zone, output several shortest simulated evacuation times as the first evacuation duration set, output several straight-line distances to the danger zone as the first danger distance set, and output the number of multiple simulated personnel within the range of multiple escape exits as the first exit load feature.

[0099] Summing the first evacuation duration set and the first danger distance set respectively yields the first total evacuation duration and the first total danger distance;

[0100] Based on the first outlet load characteristics, the standard deviation of multiple simulated personnel numbers is calculated to obtain the first outlet load balancing coefficient.

[0101] Based on the preset index weights, the first total evacuation time, the first total danger distance, and the first exit load balancing coefficient are weighted and summed to obtain the first scheme fitness. Then, multiple scheme fitnesss are analyzed in sequence. Among them, the scheme fitness is negatively correlated with the total evacuation time, the total danger distance, and the exit load balancing coefficient.

[0102] In this embodiment, the personnel escape simulation uses a discrete time step model to simulate the movement of personnel along the escape route. The personnel movement speed can be dynamically adjusted according to the degree of congestion on the route, such as 1.5 m / s in open sections and 0.8 m / s in congested sections.

[0103] The first evacuation time set records the time each person takes from the starting point to the safety exit. For example, person A takes 45 seconds, and person B takes 52 seconds, forming a set containing N time values. The first danger distance set calculates the minimum straight-line distance between each route and the danger zone. For example, the closest distance from the fire source to the middle of the route is 10 meters, forming a set containing M distance values, where M is the total number of routes. The exit load characteristic is the cumulative number of people received at each safety exit at the end of the simulation. For example, safety exit 1 received 80 people, and safety exit 2 received 120 people.

[0104] Furthermore, the first total evacuation time = the sum of all time values ​​in the first evacuation time set, such as 5000 seconds for 100 people. The first total danger distance = the sum of all distance values ​​in the first danger distance set, such as 300 meters for 20 routes.

[0105] In this embodiment, the first exit load balancing coefficient is calculated by first calculating the average number of people receiving each security exit, such as 80, 120, and 100 people receiving security exits 1-3; then calculating the standard deviation, such as 16.33. Further, this standard deviation is used as the first exit load balancing coefficient, as shown in the example where the value is 16.33. The smaller the value, the better the exit load balancing.

[0106] Furthermore, a weighted calculation of the suitability of the proposed solutions is required. First, the first total evacuation time, the first total danger distance, and the first exit load balancing coefficient need to be considered. For example, the suitability of the first proposed solution can be calculated as: First Solution Suitability = 1 / (α × First Total Evacuation Time + β × First Total Danger Distance + γ × First Exit Load Balancing Coefficient), where α, β, and γ are weight values, all positive numbers less than 1, and their sum is 1. The specific weight values ​​can be determined based on the actual implementation requirements. A higher suitability indicates a better solution.

[0107] Based on the above method, the fitness of multiple initial escape plans is obtained by analyzing them sequentially.

[0108] In step S400 of this embodiment, the escape route is optimized within a preset time zone based on the fitness of the multiple schemes until the optimization converges, and the optimal escape scheme is output, including:

[0109] Based on the fitness of the multiple schemes, the initial escape schemes are arranged in descending order of fitness to obtain an initial solution sequence.

[0110] The first K solutions of the initial solution sequence are designated as excellent solutions, and the last J solutions are designated as inferior solutions. Then, with the K excellent solutions as the center, the J inferior solutions are randomly clustered with equal value to obtain K solution sets, where J is P times K and P is an integer greater than 9.

[0111] Within each solution set, the optimal solution is used as the optimization direction, and the inferior solution is adjusted according to the optimization step size to obtain K updated solution sets. If the updated inferior solution involves the predicted danger zone, any initial escape plan is randomly selected for replacement.

[0112] Identify the K updated solution sets. Within each updated solution set, if the fitness of the inferior solution is greater than or equal to the fitness of the superior solution, then replace the superior solution with the inferior solution.

[0113] Continue iterative optimization until the preset number of optimization attempts is reached. Output K currently updated solutions and set the solution set with the largest sum of fitness among the K currently updated solutions as the optimal solution set. Set the best solution in the optimal solution set as the optimal escape plan.

[0114] First, the initial escape plans need to be sorted in descending order of fitness (higher fitness indicates a better plan), forming an initial solution sequence S = [s1, s2, ..., s...]. n ], where s1 is the current optimal solution. Set the number of optimal solutions K, such as K = 10, and take the first K solutions in the sequence as the optimal solution set G = {g1, g2, ..., g...}. k The number of inferior solutions J = K × P (P > 9, e.g., J = 100 when P = 10), and the last J solutions are taken as the set of inferior solutions B = {b1, b2, ..., b}. j}

[0115] Then, using each excellent solution gi as the cluster center, the inferior solution B is randomly assigned to K clusters, ensuring that each cluster contains J / K inferior solutions. For example, when K=10 and J=100, each cluster contains 10 inferior solutions. Then, by calculating the Hamming distance between inferior and excellent solutions, i.e., the route difference, the assignment strategy is adjusted to make the average difference between inferior solutions and their corresponding excellent solutions within each cluster similar, avoiding any inferior solution being too good or too bad. For example, the solution set C1 = {g1,b1,b2,...,b...} 10}, centered at g1, contains 10 inferior solutions; solution set C2 = {g2, b} 11 ,b 12 ,...,b 20}, with g2 as the center, and so on; finally generating K solution sets {C1,C2,...,C}. k}

[0116] Next, using the optimal solution as the optimization direction, the suboptimal solutions are adjusted according to the optimization step size, resulting in K updated solution sets. The step size parameter δ represents the adjustment magnitude of the suboptimal solution route; for example, δ can be set to 0.1. For instance, during the optimization process, one non-critical node in the suboptimal solution route can be randomly replaced with a probability of δ, or 5% of the personnel in the suboptimal solution can be reassigned to the target exit with a probability of δ.

[0117] For each inferior solution bj in the solution set Ci, it needs to be adjusted towards the superior solution gi. First, the route differences between bj and gi need to be calculated, such as different choices of a certain corridor. Then, according to the optimization step size δ, the solution moves closer to the route characteristics of gi, such as replacing the difference nodes of bj with the corresponding nodes of gi. If the adjusted inferior solution route crosses the predicted danger zone, an unused initial escape plan is randomly selected to replace the inferior solution, ensuring the feasibility of updating the solution set.

[0118] Next, for each solution set Ci, compare the fitness of all inferior solutions bj with that of the superior solution gi. If the fitness of bj is greater than or equal to that of gi, then replace gi with bj to form a new superior solution gi'; otherwise, gi remains unchanged.

[0119] The optimization count is preset to T, such as T = 50 times. After each iteration, the above clustering, adjustment and replacement process is repeated. Then the sum of the fitness of each solution set is recorded. When the number of iterations reaches T, the optimization is terminated.

[0120] Finally, the sum of the fitness of the K updated solution sets is calculated, and the solution set with the largest sum is selected as the optimal solution set. The best solution in the optimal solution set is set as the optimal escape plan.

[0121] The final optimal escape plan should include a set of escape routes for people in each area and a plan for allocating safe exits.

[0122] In step S500 of this embodiment, the device terminal sends the several optimal escape routes to several associated escapees, and performs dynamic route updates according to the danger perception data until the evacuation is completed. Specifically, the real-time location of each escapee is determined based on the positioning data of the terminal device carried by the person, and the optimal escape route is obtained according to the above calculation method. For example, route R1 corresponds to person A in the lobby on the 1st floor, and route R2 corresponds to person B in the corridor on the 2nd floor.

[0123] During the evacuation process, the IoT monitoring array continues to collect temperature, smoke concentration, and vibration data, updating the danger zone every 5 seconds. If a new danger is found on the original route, such as a new smoke cover in a corridor, the route is immediately replanned, and the optimal escape route that avoids the new danger is sent to several related evacuees.

[0124] Furthermore, when personnel arrive at a safety exit, their identities can be identified using RFID readers or cameras at the exit, and they can be marked as "evacuated." The system can then update the number of personnel who have not yet evacuated in real time, so that firefighters can carry out further rescue operations.

[0125] Example 2, as Figure 2 As shown, based on the same inventive concept as the indoor escape route optimization method provided in Embodiment 1, this embodiment of the invention also provides an indoor escape route optimization system, including:

[0126] The layered road network construction module 11 is used to construct a layered road network model based on the indoor map vector data and road network topology of the target building, and to mark key escape nodes.

[0127] The hazard diffusion simulation module 12 is used to receive IoT monitoring data to simulate hazard diffusion when a hazard occurs, and output the predicted hazard impact area within a preset time zone;

[0128] The personnel distribution prediction module 13 is used to combine the camera and Wi-Fi probe to obtain crowd distribution information, predict crowd movement, and output a predicted personnel distribution map within a preset time zone.

[0129] The escape plan generation module 14 is used to utilize the layered road network model to optimize the escape routes within a preset time zone based on the predicted danger zone and the predicted personnel distribution map, with the goals of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance. The optimal escape plan includes several optimal escape routes.

[0130] The escape route sending module 15 is used to send the several optimal escape routes to several associated escape personnel through the device terminal, and to dynamically update the routes according to the danger perception data until the personnel have been evacuated.

[0131] Furthermore, the hazard diffusion simulation module 12 includes the following execution steps: configuring an IoT monitoring array, wherein the IoT monitoring array includes at least a temperature sensor group, a smoke detector group, and a vibration sensor group, and each sensor is identified with location information; receiving IoT monitoring data through the IoT monitoring array when a hazard occurs, wherein the IoT monitoring data includes a temperature distribution sequence, a smoke concentration distribution sequence, and a vibration characteristic distribution sequence; performing hazard diffusion simulation based on the temperature distribution sequence, the smoke concentration distribution sequence, and the vibration characteristic distribution sequence, and outputting the predicted hazard impact area within a preset time zone.

[0132] The process includes: simulating hazard diffusion based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, and outputting a predicted hazard impact area within a preset time zone. This includes: identifying hazard types based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence to determine the current hazard event type, wherein the hazard event type includes at least fire and earthquake; matching the current hazard event type to obtain an adapted hazard diffusion simulator, wherein the hazard diffusion simulator is constructed based on a deep neural network and trained to convergence using sample data; and using the adapted hazard diffusion simulator to simulate hazard diffusion based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, and outputting a predicted hazard impact area within a preset time zone.

[0133] Furthermore, the personnel distribution prediction module 13 includes the following execution steps: combining cameras and Wi-Fi probes to acquire crowd distribution information at multiple consecutive monitoring time points, and constructing a crowd distribution feature sequence; based on the historical escape records of the target building, and constrained by the current dangerous event type, collecting a sample crowd distribution feature sequence set, and obtaining a sample crowd distribution map of the sample crowd distribution feature sequence within the historical time zone, and obtaining a sample crowd distribution map set, wherein the time interval of the historical time zone is the same as the preset time zone; using the sample crowd distribution feature sequence set and the sample crowd distribution map set, training a deep neural network until convergence to obtain a crowd movement predictor, and analyzing the crowd distribution feature sequence to obtain a predicted personnel distribution map within the preset time zone.

[0134] Furthermore, the escape plan generation module 14 includes the following execution steps: rendering the predicted danger zone and the predicted personnel distribution map onto the layered road network model to construct a personnel escape simulation space;

[0135] Within the simulated evacuation space, using the predicted danger zone as a passage restriction, and combining it with the predicted personnel distribution map, evacuation routes are simulated, and multiple initial evacuation plans are randomly generated. Each initial evacuation plan includes several evacuation routes associated with several evacuees. Based on these multiple initial evacuation plans, evacuation simulations are performed, and the fitness of multiple plans is evaluated with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance. Based on the fitness of the multiple plans, evacuation routes within a preset time zone are optimized until convergence, and the optimal evacuation plan is output.

[0136] The process involves simulating evacuation based on multiple initial escape plans, evaluating the fitness of each plan with the objectives of minimizing evacuation time, maximizing path safety levels, and optimizing exit load balancing. This includes: randomly selecting a first initial escape plan to simulate evacuation within a preset time zone; outputting several shortest simulated evacuation times as a first evacuation duration set; outputting several straight-line distances to the danger zone as a first danger distance set; and outputting the number of simulated personnel within multiple escape exit ranges as a first exit load characteristic. The first evacuation duration set and the first danger distance set are summed to obtain a first total evacuation time and a first total danger distance. Based on the first exit load characteristic, the standard deviation of the number of simulated personnel is calculated to obtain a first exit load balancing coefficient. Based on preset index weights, the first total evacuation time, the first total danger distance, and the first exit load balancing coefficient are weighted and summed to obtain the first plan fitness. Multiple plan fitnesss are then analyzed sequentially, with plan fitness being negatively correlated with total evacuation time, total danger distance, and exit load balancing coefficient.

[0137] The process of optimizing escape routes within a preset time zone based on the fitness of multiple escape routes until convergence is achieved, and then outputting the optimal escape route, includes: First, based on the fitness of the multiple escape routes, an initial solution sequence is obtained by arranging the initial escape route solutions in descending order of fitness; the first K solutions of the initial solution sequence are designated as optimal solutions, and the last J solutions as inferior solutions, and random iso-clustering is performed on the J inferior solutions centered on the K optimal solutions to obtain K solution sets, where J is a multiple of K, and P is an integer greater than 9; within each solution set, the optimal solution is used as the optimization direction, and the optimal escape route is optimized according to the optimization... The step size is adjusted to obtain K updated solution sets. If the updated inferior solution involves the predicted danger zone, any initial escape plan is randomly selected for replacement. The K updated solution sets are identified. In each updated solution set, if the fitness of the inferior solution is greater than or equal to the fitness of the superior solution, the inferior solution replaces the superior solution. Iterative optimization continues until a preset number of optimization attempts is reached. K current updated solution sets are output, and the solution set with the largest sum of fitness among the K current updated solution sets is set as the optimal solution set. The superior solution of the optimal solution set is set as the optimal escape plan.

[0138] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0139] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0140] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0141] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0142] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0143] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0144] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. An indoor escape route optimization method, characterized in that, The methods include: Based on the indoor map vector data and road network topology of the target building, a hierarchical road network model is constructed, and key escape nodes are marked. When a danger occurs, it receives IoT monitoring data to simulate the spread of the danger and outputs the predicted danger-affected area within a preset time zone. By combining cameras and Wi-Fi probes to obtain crowd distribution information, and predicting crowd movement, a predicted personnel distribution map within a preset time zone is output. Using the hierarchical road network model, with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, the escape routes within a preset time zone are optimized based on the predicted danger zone and predicted personnel distribution map, and the optimal escape plan is output. The optimal escape plan includes several optimal escape routes. The device terminal sends the several optimal escape routes to several associated escapees, and dynamically updates the routes according to the danger perception data until all personnel have been evacuated. Using the aforementioned hierarchical road network model, with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, the system optimizes escape routes within a preset time zone based on the predicted danger zone and predicted personnel distribution map, and outputs the optimal escape plan, including: The predicted danger zone and predicted personnel distribution map are rendered onto the hierarchical road network model to construct a personnel escape simulation space. Within the simulated personnel escape space, the predicted danger zone is used as the access restriction condition. Combined with the predicted personnel distribution map, the escape route is simulated, and multiple initial escape plans are randomly generated. Each initial escape plan includes several escape routes associated with several escaped personnel. Based on the multiple initial escape plans, personnel escape simulations were conducted, and the fitness of the multiple plans was evaluated with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance. Based on the fitness of the multiple schemes, the escape route within the preset time zone is optimized until the optimization converges, and the optimal escape scheme is output.

2. The indoor escape route optimization method according to claim 1, characterized in that, It receives IoT monitoring data, simulates hazard spread, and outputs the predicted hazard impact area within a preset time zone, including: Configure an IoT monitoring array, wherein the IoT monitoring array includes at least a temperature sensor group, a smoke detector group, and a vibration sensor group, and each sensor is identified with location information; When a danger occurs, IoT monitoring data is received through the IoT monitoring array, wherein the IoT monitoring data includes temperature distribution sequence, smoke concentration distribution sequence and vibration characteristic distribution sequence; Based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, a hazard diffusion simulation is performed, and the predicted hazard-affected area within a preset time zone is output.

3. The indoor escape route optimization method according to claim 2, characterized in that, Based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, a hazard diffusion simulation is performed, outputting the predicted hazard-affected area within a preset time zone, including: Hazard type identification is performed based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence to determine the current hazardous event type, wherein the hazardous event type includes at least fire and earthquake; An adaptive hazard diffusion simulator is obtained based on the current hazard event type matching. The hazard diffusion simulator is constructed based on a deep neural network and trained to convergence using sample data. Using the adaptive hazard diffusion simulator, the hazard diffusion simulation is performed based on the temperature distribution sequence, smoke concentration distribution sequence, and vibration characteristic distribution sequence, and the predicted hazard impact area within the preset time zone is output.

4. The indoor escape route optimization method according to claim 3, characterized in that, By combining camera and Wi-Fi probes to acquire crowd distribution information and predict crowd movement, a predicted population distribution map within a preset time zone is output, including: By combining cameras and Wi-Fi probes to acquire crowd distribution information at multiple consecutive monitoring time points, a crowd distribution feature sequence is constructed. Based on the historical escape records of the target building, and constrained by the current dangerous event type, a sample population distribution feature sequence set is collected, and a sample population distribution map is obtained within the historical time zone, thus obtaining a sample population distribution map set. The time interval of the historical time zone is the same as the preset time zone. Using the sample population distribution feature sequence set and sample population distribution map set, a deep neural network is trained until convergence to obtain a population movement predictor, and a predicted population distribution map within a preset time zone is obtained based on the population distribution feature sequence analysis.

5. The indoor escape route optimization method according to claim 1, characterized in that, Based on the aforementioned initial escape plans, personnel escape simulations were conducted, and the fitness of multiple plans was evaluated with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balancing. These included: Randomly select the first initial escape plan to simulate the escape of personnel within a preset time zone, output several shortest simulated evacuation times as the first evacuation duration set, output several straight-line distances to the danger zone as the first danger distance set, and output the number of multiple simulated personnel within the range of multiple escape exits as the first exit load feature. Summing the first evacuation duration set and the first danger distance set respectively yields the first total evacuation duration and the first total danger distance; Based on the first outlet load characteristics, the standard deviation of multiple simulated personnel numbers is calculated to obtain the first outlet load balancing coefficient. Based on the preset index weights, the first total evacuation time, the first total danger distance, and the first exit load balancing coefficient are weighted and summed to obtain the first scheme fitness. Then, multiple scheme fitnesss are analyzed in sequence. Among them, the scheme fitness is negatively correlated with the total evacuation time, the total danger distance, and the exit load balancing coefficient.

6. The indoor escape route optimization method according to claim 1, characterized in that, Based on the fitness of the multiple schemes, the escape route within a preset time zone is optimized until convergence, and the optimal escape scheme is output, including: Based on the fitness of the multiple schemes, the initial escape schemes are arranged in descending order of fitness to obtain an initial solution sequence. The first K solutions of the initial solution sequence are designated as excellent solutions, and the last J solutions are designated as inferior solutions. Then, with the K excellent solutions as the center, the J inferior solutions are randomly clustered with equal value to obtain K solution sets, where J is P times K and P is an integer greater than 9. Within each solution set, the optimal solution is used as the optimization direction, and the inferior solution is adjusted according to the optimization step size to obtain K updated solution sets. If the updated inferior solution involves the predicted danger zone, any initial escape plan is randomly selected for replacement. Identify the K updated solution sets. Within each updated solution set, if the fitness of the inferior solution is greater than or equal to the fitness of the superior solution, then replace the superior solution with the inferior solution. Continue iterative optimization until the preset number of optimization attempts is reached. Output K currently updated solutions and set the solution set with the largest sum of fitness among the K currently updated solutions as the optimal solution set. Set the best solution in the optimal solution set as the optimal escape plan.

7. An indoor escape route optimization system, characterized in that, The system includes: The layered road network construction module is used to build a layered road network model based on the indoor map vector data and road network topology of the target building, and to mark key escape nodes. The hazard diffusion simulation module is used to receive IoT monitoring data to simulate hazard diffusion when a hazard occurs, and output the predicted hazard impact area within a preset time zone; The personnel distribution prediction module is used to combine cameras and Wi-Fi probes to obtain crowd distribution information, predict crowd movement, and output a predicted personnel distribution map within a preset time zone. The escape plan generation module is used to optimize the escape routes within a preset time zone based on the predicted danger zone and the predicted personnel distribution map, with the goal of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, using the hierarchical road network model. The optimal escape plan includes several optimal escape routes. The escape route sending module is used to send the several optimal escape routes to several associated escape personnel through the device terminal, and to dynamically update the routes according to the danger perception data until the personnel have been evacuated. Using the aforementioned hierarchical road network model, with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance, the system optimizes escape routes within a preset time zone based on the predicted danger zone and predicted personnel distribution map, and outputs the optimal escape plan, including: The predicted danger zone and predicted personnel distribution map are rendered onto the hierarchical road network model to construct a personnel escape simulation space. Within the simulated personnel escape space, the predicted danger zone is used as the access restriction condition. Combined with the predicted personnel distribution map, the escape route is simulated, and multiple initial escape plans are randomly generated. Each initial escape plan includes several escape routes associated with several escaped personnel. Based on the multiple initial escape plans, personnel escape simulations were conducted, and the fitness of the multiple plans was evaluated with the objectives of minimizing evacuation time, maximizing path safety level, and optimizing exit load balance. Based on the fitness of the multiple schemes, the escape route within the preset time zone is optimized until the optimization converges, and the optimal escape scheme is output.

Images