Hospital fire emergency evacuation method and system based on internet of things

Abstract

The application discloses a hospital fire emergency evacuation method and system based on Internet of Things, and relates to the technical field of emergency evacuation methods. A plurality of emergency evacuation paths are determined according to the indoor distribution map of the hospital and the regional position of the fire area. The best emergency evacuation path is determined based on the plurality of emergency evacuation paths, the current position of the hospitalized patient and the hospital fire state map. Therefore, the evacuation action instruction of the hospitalized patient is determined according to the node position of the plurality of emergency evacuation nodes, the corresponding fire smoke area and the action state of the hospitalized patient. The corresponding fire fighting event is determined according to the evacuation behavior of the hospitalized patient and the position of the fire fighting personnel. The plurality of fire fighting projects are determined according to the identification of the fire fighting event. The multi-level rescue event of the hospitalized patient is determined according to the plurality of fire fighting projects, the current position of the hospitalized patient and the fire change event of each fire area. The accuracy of the multi-level rescue event of the hospitalized patient is improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of emergency evacuation methods, in particular to a hospital fire emergency evacuation method and system based on Internet of Things. BACKGROUND

[0002] With the development of science and technology, inpatient patients are in a targeted treatment area of a hospital, in order to further improve the safety of the hospital, fire drills of the hospital need to be carried out, in the prior art, the hospital is in a fire drill scene, a fire position is collected, and the evacuation direction of the inpatient patients is regulated according to the fire position, a preset emergency evacuation path is used as a single-dimensional path, and the best emergency evacuation path is not the best emergency evacuation path, so that the accuracy of the best emergency evacuation path is low, and the evacuation action instruction of the inpatient patients cannot be realized, and the accuracy of the multistage rescue event of the inpatient patients is affected. SUMMARY

[0003] The application aims at overcoming the defects of the prior art, and provides a hospital fire emergency evacuation method and system based on Internet of Things.

[0004] The application provides a hospital fire emergency evacuation method based on Internet of Things, which comprises the following steps:

[0005] An indoor image is determined based on a plurality of cameras of the hospital and Internet of Things, and a plurality of abnormal object features are determined according to the recognition of the indoor image;

[0006] A corresponding fire warning level is determined based on the feature position and feature form of the plurality of abnormal object features, a fire area is determined according to the warning level of the plurality of abnormal object features and corresponding smoke data, and a corresponding fire dynamic map is determined, which comprises the following steps: corresponding smoke data is determined based on smoke detection of the feature position of each abnormal object feature, and at the same time, a first fire information combination is determined according to the relative position of the plurality of abnormal object features and the corresponding smoke data; a second fire information combination is determined according to the fire warning level of the plurality of abnormal object features and the corresponding smoke data, the fire area is determined based on the first fire information combination and the second fire information combination, and a corresponding fire dynamic map is determined based on the monitoring of each fire area in the fire area; the fire dynamic map contains the following information: boundary change of the fire area, smoke diffusion direction, risk level change;

[0007] A plurality of emergency evacuation paths are determined according to the indoor distribution map of the hospital and the area position of the fire area, and the best emergency evacuation path is determined based on the plurality of emergency evacuation paths, the current position of the inpatient patient and the hospital fire state map;

[0008] A plurality of emergency evacuation nodes are determined according to the identification of the optimal emergency evacuation path, the evacuation action instructions of the inpatients are determined according to the node positions of the plurality of emergency evacuation nodes, the corresponding fire smoke areas and the movement states of the inpatients, and the emergency evacuation behaviors of the inpatients are guided based on the Internet of Things;

[0009] The corresponding fire fighting events are determined according to the evacuation behaviors of the inpatients and the positions of the fire fighters, a plurality of fire fighting items are determined according to the identification of the fire fighting events, and the multi-level rescue events of the inpatients are determined according to the plurality of fire fighting items, the current positions of the inpatients and the fire change events of the respective fire areas.

[0010] The embodiment of the present application provides a hospital fire emergency evacuation system based on the Internet of Things, which is applied to the hospital fire emergency evacuation method based on the Internet of Things.

[0011] The abnormal object feature module is used for determining indoor images based on a plurality of cameras of the hospital and the Internet of Things, and determining a plurality of abnormal object features according to the identification of the indoor images.

[0012] The fire dynamic map module is used for determining the corresponding fire warning levels based on the feature positions and feature morphologies of the plurality of abnormal object features, determining the fire areas according to the warning levels of the plurality of abnormal object features and the corresponding smoke data, and determining the corresponding fire dynamic maps, including: determining the corresponding smoke data based on the smoke detection of the feature positions of the respective abnormal object features, and simultaneously determining the first fire information combination according to the relative positions of the plurality of abnormal object features and the corresponding smoke data; determining the second fire information combination according to the warning levels of the plurality of abnormal object features and the corresponding smoke data, determining the fire areas based on the first fire information combination and the second fire information combination, and determining the corresponding fire dynamic maps based on the monitoring of the respective fire areas in the fire areas; the fire dynamic map contains the following information: the boundary change of the fire area, the smoke diffusion direction and the risk level change.

[0013] The emergency evacuation path module is used for determining a plurality of emergency evacuation paths according to the indoor distribution map of the hospital and the area positions of the fire areas, and determining the optimal emergency evacuation path based on the plurality of emergency evacuation paths, the current positions of the inpatients and the hospital fire state map.

[0014] The evacuation action instruction module is used for determining a plurality of emergency evacuation nodes according to the identification of the optimal emergency evacuation path, determining the evacuation action instructions of the inpatients according to the node positions of the plurality of emergency evacuation nodes, the corresponding fire smoke areas and the movement states of the inpatients, and guiding the emergency evacuation behaviors of the inpatients based on the Internet of Things.

[0015] A multi-stage rescue event module is used to determine corresponding fire fighting events according to evacuation behaviors of inpatients and positions of fire fighters, to determine a plurality of fire fighting items according to identification of the fire fighting events, and to determine multi-stage rescue events of the inpatients according to the plurality of fire fighting items, current positions of the inpatients, and fire change events of each fire area.

[0016] Compared with the prior art, the present application has the following beneficial effects:

[0017] In the embodiment of the present application, the method in the embodiment of the present application is used to determine a fire area according to a warning level of a plurality of abnormal object features and corresponding smoke data, and to determine a corresponding fire dynamic diagram; a plurality of emergency evacuation paths are determined according to an indoor distribution diagram of a hospital and a region position of the fire area, and a best emergency evacuation path is determined based on the plurality of emergency evacuation paths, current positions of inpatients, and a hospital fire state diagram; the fire area is introduced, the overall consideration of the plurality of emergency evacuation paths, the current positions of the inpatients, and the hospital fire state diagram is compatible, and the accuracy of the best emergency evacuation path is improved.

[0018] Therefore, a plurality of emergency evacuation nodes are determined according to identification of the best emergency evacuation path, evacuation action instructions of the inpatients are determined according to node positions of the plurality of emergency evacuation nodes, corresponding fire smoke areas, and action states of the inpatients, and emergency evacuation behaviors of the inpatients are guided based on the Internet of Things; corresponding fire fighting events are determined according to evacuation behaviors of the inpatients and positions of fire fighters, a plurality of fire fighting items are determined according to identification of the fire fighting events, multi-stage rescue events of the inpatients are determined according to the plurality of fire fighting items, current positions of the inpatients, and fire change events of each fire area, the evacuation action instructions of the inpatients are introduced, the overall consideration of the plurality of fire fighting items, the current positions of the inpatients, and the fire change events of each fire area is realized, the accuracy of the multi-stage rescue events of the inpatients is improved, and the full-process rescue of the inpatients is realized. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of the hospital fire emergency evacuation method based on the Internet of Things in the embodiment of the present application;

[0020] Figure 2 is a flowchart of step S11 in the hospital fire emergency evacuation method based on the Internet of Things in the embodiment of the present application;

[0021] Figure 3 is a flowchart of step S12 in the hospital fire emergency evacuation method based on the Internet of Things in the embodiment of the present application;

[0022] Figure 4This is a flowchart illustrating step S13 in the Internet of Things-based emergency evacuation method for hospital fires in an embodiment of the present invention.

[0023] Figure 5 This is a flowchart illustrating step S14 of the Internet of Things-based emergency evacuation method for hospital fires in an embodiment of the present invention.

[0024] Figure 6 This is a flowchart illustrating step S15 of the Internet of Things-based emergency evacuation method for hospital fires in an embodiment of the present invention.

[0025] Figure 7 This is a schematic diagram of the structural composition of a hospital fire emergency evacuation system based on the Internet of Things in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] Please see Figures 1 to 7 An IoT-based emergency evacuation method for hospital fires, applied to emergency evacuation scenarios; the IoT-based emergency evacuation method for hospital fires includes:

[0028] Step S11: Determine indoor images based on multiple cameras and the Internet of Things in the hospital, and identify multiple abnormal object features based on the recognition of indoor images;

[0029] Step S12: Determine the corresponding fire warning level based on the feature location and feature shape of multiple abnormal objects, determine the fire area based on the warning level of multiple abnormal objects and the corresponding smoke data, and determine the corresponding fire dynamic map.

[0030] Step S13: Determine multiple emergency evacuation routes based on the hospital's indoor layout map and the location of the fire area. Based on the multiple emergency evacuation routes, the current location of hospitalized patients, and the hospital fire status map, determine the optimal emergency evacuation route.

[0031] Step S14: Based on the identification of the optimal emergency evacuation route, determine multiple emergency evacuation nodes, determine the evacuation action instructions for hospitalized patients based on the node locations of multiple emergency evacuation nodes, the corresponding fire smoke areas, and the movement status of hospitalized patients, and guide the emergency evacuation behavior of hospitalized patients based on the Internet of Things.

[0032] Step S15: Determine the corresponding fire-fighting events based on the evacuation behavior of hospitalized patients and the location of firefighters. Determine multiple fire-fighting items based on the identification of fire-fighting events. Determine multi-level rescue events for hospitalized patients based on multiple fire-fighting items, the current location of hospitalized patients, and fire change events in various fire areas.

[0033] refer to Figure 2 In step S11, the specific steps are as follows:

[0034] S111: Cameras are installed in different locations in the hospital. Multiple cameras communicate with each other through the Internet of Things. Multiple cameras dynamically capture images of the hospital's interior space and collect images of multiple sub-rooms. The interior image is determined based on the shooting location of multiple sub-room images, the images of two adjacent sub-rooms, and the image synthesis mechanism corresponding to the Internet of Things.

[0035] S112: Based on indoor image detection, multiple object monitoring areas are determined. In the multiple object monitoring areas, the shape of the object is determined according to the recognition of each object monitoring area. Based on the shape of the object, the corresponding abnormal location and the surrounding environmental features, the corresponding abnormal object features are determined to collect multiple abnormal object features.

[0036] In the embodiments of this application, corresponding cameras are installed at different locations in the hospital. Multiple cameras communicate with each other through the Internet of Things (IoT). The multiple cameras dynamically capture images of the hospital's interior space and collect multiple sub-indoor images. The indoor image is determined based on the shooting positions of the multiple sub-indoor images, the images of two adjacent sub-indoor images, and the image synthesis mechanism corresponding to the IoT. This approach takes into account the overall consideration of the shooting positions of the multiple sub-indoor images, the images of two adjacent sub-indoor images, and the image synthesis mechanism corresponding to the IoT, thus ensuring the accuracy of the indoor image.

[0037] At this time, high-definition network cameras are deployed in key areas of the hospital (such as ward corridors, stairwells, pharmacies, ICUs, etc.). Each camera is equipped with a unique IP address and connects to the central server through IoT protocols (such as MQTT, CoAP). The cameras are powered by PoE (Power over Ethernet) to ensure that they can continue to work through UPS when there is a power outage. The system achieves low-latency communication (<100ms) between devices through IoT gateways and uses TLS encryption to ensure data transmission security.

[0038] Each camera uses H.265 encoding to capture video streams in real time (bitrate 4Mbps), and extracts 10 keyframes per second as sub-indoor images using the OpenCV algorithm; each sub-image contains: timestamp (accurate to milliseconds); camera ID; spatial location parameters (latitude, longitude, floor, height); lens parameters (focal length, aperture, distortion coefficient); the system ensures that the time synchronization error of all cameras is <50ms through the NTP protocol.

[0039] Furthermore, multiple object monitoring areas are determined based on indoor image detection. Within these multiple object monitoring areas, the shape of the object is determined based on the identification of each monitoring area. Based on the object's shape, corresponding abnormal location, and surrounding environmental features, corresponding abnormal object features are determined to collect multiple abnormal object features. This approach integrates the identification of each object monitoring area to determine the object's shape. By considering the object's shape, corresponding abnormal location, and surrounding environmental features as a whole, the accuracy of the corresponding abnormal object features is ensured.

[0040] At this point, the system adopts a grid-based partitioning method, dividing the indoor images into 5m×5m monitoring units, and further dividing the areas into Level 1 monitoring zones (high-risk: pharmacy, oxygen station, power distribution room), Level 2 monitoring zones (medium-risk: wards, nurses' station, corridors), and Level 3 monitoring zones (low-risk: waiting area, administrative office area) according to risk level. In addition, the system supports a dynamic adjustment mechanism, such as automatically increasing the monitoring priority between the pharmacy and equipment at night, and strengthening the monitoring of corridors and waiting areas during peak hours. In terms of recognition algorithms, the YOLOv5 object detection model is used. This model has been trained on more than 10,000 hospital scene images, achieving a recognition accuracy of 95.2% and a single-frame processing latency of less than 200 milliseconds, laying a solid technical foundation for subsequent object recognition and anomaly detection.

[0041] After the monitoring area is divided, the system enters the object recognition and analysis phase. First, the object's shape is classified, including solids (such as equipment, furniture, and debris), liquids (such as leaked liquids and stagnant water), gases (such as smoke and steam), and special shapes (such as flames and electric sparks). Features of each shape are extracted, such as the aspect ratio, volume estimation, and material identification of solids; the diffusion area, flow direction, and color characteristics of liquids; the diffusion speed, concentration gradient, and color intensity of gases; and the irregularity and dynamic change rate of special shapes. Second, the system accurately locates abnormal locations using absolute coordinates (based on the X and Y coordinates of the hospital's floor plan) and relative coordinates (such as relative to fire hydrants), and determines whether a location is abnormal based on criteria such as the presence of prohibited areas, abnormal clustering, or abnormal distribution. Finally, the system performs a comprehensive analysis based on surrounding environmental characteristics, including collecting temperature, humidity, light intensity, and ventilation data through thermal imagers, humidity sensors, light intensity sensors, and wind speed sensors. Based on this data, flammability, obstruction, and hazard are assessed, providing comprehensive environmental evidence for the identification of abnormal objects.

[0042] After completing the analysis of object morphology, location, and environmental characteristics, the system enters the stage of identifying and collecting abnormal object features. Each abnormal object is assigned a unique feature ID (e.g., AO-20240917-001), and its object type, location information, morphological features, dynamic features (e.g., change trend, change rate, and predicted trend), and correlation features (e.g., spatial relationship with other objects, correlation with environmental parameters, and possible chain reactions) are recorded. The collection process includes preliminary detection (the system automatically identifies suspicious objects and records basic features), and in-depth analysis (calling high-precision algorithms combined with environmental data for comprehensive analysis). The system involves comprehensive assessment and feature confirmation (optional manual review or automatic system confirmation), ultimately storing all features in an abnormal object feature database and establishing a time-series record. Furthermore, the system categorizes anomalies into three levels based on their urgency: Level 1 anomalies (such as open flames or flammable liquid leaks) require a response within 10 seconds; Level 2 anomalies (such as equipment overheating or partial passageway blockage) require a response within 30 seconds; and Level 3 anomalies (such as improperly placed items) require a response within 5 minutes. Through this rigorous process, the system can efficiently and accurately collect multiple abnormal object features, providing crucial data support for subsequent fire warnings and emergency evacuations.

[0043] refer to Figure 3 In step S12, the specific steps are as follows:

[0044] S121: Real-time monitoring of the feature locations of multiple abnormal object features, and determination of the corresponding fire warning level based on the feature locations, corresponding feature shapes, and relative distance between two adjacent abnormal object features, so as to mark the fire warning level of each abnormal object feature.

[0045] S122: Based on the smoke detection of the feature positions of each abnormal object, the corresponding smoke data is determined. At the same time, the first fire information combination is determined according to the relative positions of multiple abnormal object features and the corresponding smoke data.

[0046] S123: Determine the second fire information combination based on the fire warning level and corresponding smoke data of multiple abnormal object characteristics, determine the fire area based on the first fire information combination and the second fire information combination, and determine the corresponding fire dynamic map based on the monitoring of each fire area in the fire area.

[0047] In the embodiments of this application, the feature positions of multiple abnormal object features are monitored in real time. The corresponding fire warning level is determined based on the feature positions of multiple abnormal object features, their corresponding feature shapes, and the relative distance between two adjacent abnormal object features. This marks the fire warning level of each abnormal object feature. It takes into account the overall consideration of the feature positions of multiple abnormal object features, their corresponding feature shapes, and the relative distance between two adjacent abnormal object features, thus ensuring the accuracy of the corresponding fire warning level.

[0048] At this time, the system collects image data in real time through cameras and IoT sensors deployed in the hospital, and extracts the location information of abnormal objects (such as smoke, flames, high-temperature equipment, etc.) through image recognition algorithms; the location of abnormal objects in the image is mapped to the coordinate system of the hospital building floor plan. For example: camera number: CAM-03 (located in the eastern section of the ICU corridor on the 3rd floor); image coordinates: (x=320, y=240); actual coordinate transformation: based on the camera calibration parameters, it is mapped to the actual building coordinates (X=15.6m, Y=22.3m, Z=3F).

[0049] A lightweight CNN model (such as MobileNetV3) is used to classify abnormal objects in images by shape: Input: an image patch of the abnormal object region (64×64 pixels); Output: shape category probability (flame, smoke, high temperature reflection, others); Feature vector extraction: a 128-dimensional feature vector is output through the intermediate layer of the model for subsequent similarity comparison; Shape confidence threshold: Flame: confidence > 0.85; Smoke: confidence > 0.75; Others: those below the above threshold are marked as "to be confirmed".

[0050] The Euclidean distance formula is used to calculate the spatial distance between any two abnormal objects. The influence of hospital building structures (such as walls and fire doors) on the actual connecting paths is considered, and a connectivity coefficient is introduced (e.g., straight corridor = 1.0, detour required = 1.5). Proximity determination: a threshold is set (e.g., 5 meters), and abnormal objects smaller than this value are considered as "neighbor groups".

[0051] The fire warning level is comprehensively assessed using the formula: Warning Level = w1 × Location Risk + w2 × Morphological Confidence + w3 × Proximity Density; where the weights are w1 = 0.4, w2 = 0.4, and w3 = 0.2 (which can be adjusted according to the actual situation of the hospital). A mapping table of warning levels is collected, as shown in Table 1.

[0052] Table 1 Mapping Relationship of Warning Levels

[0053]

[0054] Specifically, for the fire drill at Hospital A's B campus, the scenario was set as follows: Time: June 15, 2025, 14:30; Location: Corridor near the ICU on the 3rd floor of the inpatient department; Event: Simulating localized smoke caused by a short circuit in medical equipment; Camera CAM-03 detected an abnormal object A (suspected smoke): Image coordinates: (300, 250); Converted to actual coordinates: (15.2m, 22.1m, 3F); Camera CAM-04 detected an abnormal object B (suspected high-temperature reflection): Image coordinates: (640, 200); Converted to actual coordinates: (18.5m, 22.5m, 3F); Object A: CNN output: Smoke probability 0.91; Feature vector: Similarity to smoke template 0.89; Judgment: Confirmed as smoke; Object B: CNN output: High-temperature reflection probability 0.78; Feature vector: Similarity to flame template 0.65; Judgment: Suspected high-temperature source (not flame).

[0055] Euclidean distance between A and B: D is 3.3 meters; Connectivity coefficient: straight corridor, take 1.0; Effective distance: 3.3 meters < 5-meter threshold → judged as a neighboring group; Warning level assessment: Location risk: near ICU → 0.9; Morphological confidence: smoke 0.91 + high temperature 0.78 → average 0.845; Neighbor density: 1 neighboring group → 0.2; Overall score: 0.4 × 0.9 + 0.4 × 0.845 + 0.2 × 0.2 = 0.798; Warning level: Level II (close to Level I threshold).

[0056] System Response: A Level II warning was sent to the fire control center, indicating the location coordinates and morphological type; real-time images from CAM-03 / CAM-04 were automatically retrieved for on-duty personnel to confirm; abnormal areas were highlighted (yellow markers) in the hospital BIM model; warning information was simultaneously pushed to security personnel's mobile terminals; Exercise Results: The system took 4.2 seconds from detection to issuing the warning; manual confirmation matched the system's judgment, confirming it as initial smoke caused by a short circuit in the equipment; the smoke extraction system was subsequently activated and nearby non-critical patients were evacuated; the exercise was evaluated as "efficient and accurate".

[0057] Furthermore, smoke data is determined based on the smoke detection of the feature positions of each abnormal object. At the same time, a first fire information combination is determined based on the relative positions of multiple abnormal object features and the corresponding smoke data. This comprehensive consideration of the relative positions of multiple abnormal object features and the corresponding smoke data ensures the accuracy of the first fire information combination.

[0058] At this point, an improved YOLOv5-Seg model is used, combined with optical flow for dynamic smoke detection; input: a sequence of 5 consecutive frames (from cameras CAM-03 and CAM-04); output: smoke region mask and confidence score; smoke data parameters: smoke concentration (grayscale mean): 0~1 (0 for no smoke, 1 for dense smoke); diffusion rate (pixels / second): calculated by optical flow vector field; area ratio (the proportion of the image area occupied by the smoke region); updated once per second and transmitted to the edge computing node in real time.

[0059] Using CAM-03 as the reference point, calculate the Euclidean distance of other anomalous objects in the planar coordinate system. If the distance is less than a threshold (e.g., 5 meters), it is determined as a "nearby anomalous feature". The DBSCAN algorithm is used to cluster the anomalous features to form "nearby groups". At the same time, the region combination determination logic is as follows: if ≥2 anomalous objects in the same nearby group detect smoke and the average smoke concentration is >0.4, it is determined as the "first fire information combination". The region center point is the geometric center of the coordinates of all anomalous objects in the group. The region radius is based on the distance of the farthest anomalous object, with an additional 1-meter buffer.

[0060] Specifically, CAM-03 (door to the drug warehouse): Smoke area detected: image coordinates (300,200) to (340,260); smoke concentration: 0.58; diffusion speed: 1.2 m / s; CAM-04 (inside the warehouse): Smoke area detected: image coordinates (150,180) to (190,240); smoke concentration: 0.66; diffusion speed: 1.5 m / s; Anomalous object F01 (detected by CAM-03): coordinates (15.6,22.3); Anomalous object F02 (detected by CAM-04): coordinates (16.1,22.7); Euclidean distance: D is 0.64 meters; judgment result: F01 and F02 are "nearby anomalies".

[0061] First fire information combination judgment: Average smoke concentration: (0.58+0.66) / 2=0.62>0.4;

[0062] The center point of the region is: [(15.6+16.1) / 2,(22.3+22.7) / 2]=[15.85,22.5];

[0063] The radius of the area is 1.64 meters;

[0064] System Response: The FR01 area was marked with a red circle on the fire command center's large screen; the smoke extraction system in that area was automatically activated; an instruction was sent to security personnel: "Immediately proceed to the 3rd floor medicine warehouse area to confirm the fire situation"; Drill Results: The system took 6.3 seconds from detection to generating the first fire information combination; manual confirmation matched the system's judgment, confirming it as an initial electrical fire; in subsequent drills, this area combination became a key input parameter for evacuation route planning.

[0065] Therefore, a second fire information combination is determined based on the fire warning level and corresponding smoke data of multiple abnormal object characteristics. The fire area is determined based on the first and second fire information combinations. Within the fire area, the corresponding fire dynamic map is determined based on the monitoring of each fire area. This approach combines the overall considerations of the first and second fire information combinations, ensuring the accuracy of the fire area.

[0066] At this point, the second fire information combination is determined based on the fire warning levels of multiple abnormal object features and the corresponding smoke data. Input data: Fire warning levels of each abnormal object feature (e.g., output of S121): F01 (pharmacy warehouse): warning level L2; F02 (electrical distribution room): warning level L3; F03 (nurse station): warning level L1; Smoke data (e.g., output of S122): F01: smoke concentration 0.62, diffusion velocity 0.3m / s; F02: smoke concentration 0.81, diffusion velocity 0.5m / s; F03: smoke concentration 0.25, diffusion velocity 0.1m / s; Combination judgment logic: A weighted scoring model (Weighted Risk Score, WRS) is adopted: WRS = α × warning level + β × smoke concentration + γ × diffusion velocity; where α = 0.5, β = 0.3, γ = 0.2 (the weights can be adjusted according to the actual situation of the hospital); if WRS ≥ 2.0, then the feature is included in the second fire information combination.

[0067] Area combination merging: First fire information combination (S122 output): FR01 (containing F01 and F02); Second fire information combination (current step output): FR02 (containing only F02); Merging logic: If the two area combinations overlap (such as sharing F02), they are merged into one fire area.

[0068] Specifically, F01: 0.5×2+0.3×0.62+0.2×0.3=1.246;

[0069] F02: 0.5×3+0.3×0.81+0.2×0.5=1.843;

[0070] F03: 0.5×1+0.3×0.25+0.2×0.1=0.595;

[0071] Conclusion: Only F02 satisfies WRS≥2.0 and is included in the second fire information combination.

[0072] FR-merged: Center point: Coordinates of F02 (16.2, 23.1); Radius: Take the maximum radius of FR01 and FR02 (1.64 meters); Included features: F01, F02, F03 (F03 is included in the influence range because F03 is less than 3 meters away from F02).

[0073] A spatiotemporal graph convolutional network is used to model the changes in fire zones. Input data includes the center point coordinates, radius, smoke concentration, and diffusion rate of each fire zone; time-series data (updated every second). Output: a dynamic fire graph containing the following information: changes in the fire zone boundaries (expanding with smoke diffusion); smoke diffusion direction (arrow markings); risk level changes (color coding: green → yellow → red). The fire zone boundaries are recalculated every second based on the latest smoke data. Kalman filtering is used to predict the smoke diffusion trend for the next 30 seconds. Simultaneously, the following is displayed on the fire command center's large screen: the dynamic boundary of the FR-merged fire zone (red circle, expanding over time); the smoke diffusion direction (from F02 to F01 and F03); and the predicted boundary (dashed circle, indicating the possible range for the next 30 seconds).

[0074] Specifically, the second fire information combination determination: F02's WRS is 1.843 (close to the threshold), so the system includes it in the second fire information combination FR02; F01 and F03 have lower WRS and are not included; fire area merging: FR01 (containing F01 and F02) overlaps with FR02 (containing F02) and is merged into FR-merged; F03 is included in the affected area because it is less than 3 meters away from F02; fire dynamic map generation: the system displays the dynamic boundary of FR-merged in real time (initial radius 1.64 meters); predicts that smoke will spread to the nurse station (F03) in the next 30 seconds, and the boundary will expand to 2.5 meters; the high-risk area is marked in red on the command center screen and an evacuation alarm is issued; exercise results: the system took 8.7 seconds from detection to generating the fire dynamic map; manual confirmation is consistent with the system prediction, and the smoke spread path matches the actual situation; firefighters can quickly locate the fire source based on the dynamic map, and the evacuation efficiency is improved by 40%.

[0075] refer to Figure 4 In step S13, the specific steps are as follows:

[0076] S131: Collect the location and database of the hospital, and determine the indoor distribution map of the hospital based on the location and database; determine the fire-affected area based on the indoor distribution map of the hospital and the regional location of the fire area;

[0077] S132: Collect fire dynamic maps, determine the fire change trend area based on the hospital's indoor distribution map and the corresponding fire dynamic map, and determine multiple emergency evacuation routes based on the fire impact area, fire change trend area and fire area. Each emergency evacuation route can avoid the fire area along the corresponding direction.

[0078] S133: Construct a hospital fire status map based on the fire dynamic map and the fire area. At the same time, determine the first path event based on multiple emergency evacuation routes and the current location of hospitalized patients. Determine the second path event based on the current location of hospitalized patients and the hospital fire status map. Determine the optimal emergency evacuation route based on the first path event, the second path event and multiple emergency evacuation routes.

[0079] In the embodiments of this application, the location and database of the hospital are collected, and the indoor distribution map of the hospital is determined based on the location and database of the hospital. The fire impact area corresponding to the location of the fire area is determined based on the indoor distribution map of the hospital and the regional location of the fire area, which takes into account the overall consideration of the location and database of the hospital and ensures the accuracy of the indoor distribution map of the hospital.

[0080] At this point, the hospital location and database are collected: the system obtains the hospital's spatial layout data through the hospital's geographic information system (GIS) and building information model (BIM); the database includes: building floor plans (floors, rooms, corridors, stairs, etc.); locations of fixed facilities (such as fire extinguishers, fire hydrants, and emergency exits); and personnel distribution data (such as the number of beds and the location of medical staff).

[0081] Indoor distribution map generation: Using topological modeling, the hospital is divided into multiple functional areas (such as wards, operating rooms, corridors, etc.); each area is assigned a unique identifier (such as "ICU-03", "Corridor-2F-East Section"); Fire impact area determination: Based on the location of the fire area (such as FR-merged output by S123), its impact range is calculated: Initial impact radius: Dynamically adjusted based on smoke diffusion speed and concentration (such as FR-merged initial radius of 1.64 meters); Impact area expansion: Combining indoor ventilation conditions and building structure, the smoke diffusion path is predicted.

[0082] Specifically, the system loads the BIM model of the 3rd floor of Campus B, including areas such as the ICU, pharmacy, and power distribution room; the influence radius of FR-merged (including F01 and F02) is extended to 2.5 meters, covering the pharmacy and part of the corridor; the system marks the affected area as a "high-risk area" and prohibits personnel from entering.

[0083] Furthermore, fire dynamic maps are collected, and fire trend areas are determined based on the hospital's indoor layout map and the corresponding fire dynamic maps. Multiple emergency evacuation routes are determined based on the fire impact area, fire trend area, and fire area. Each emergency evacuation route can avoid the fire area along its corresponding direction, taking into account the overall consideration of the fire impact area, fire trend area, and fire area, and ensuring the accuracy of multiple emergency evacuation routes.

[0084] At this point, the fire dynamic map is generated by step S123, including: fire source location (e.g., FR-merged); smoke concentration distribution (e.g., area F01: 0.8 mg / m³); fire spread direction (e.g., spreading from the medicine warehouse to the power distribution room); temperature change trend (e.g., temperature rise rate in area F02: 2.5℃ / min); data processing: time series analysis: perform time series analysis on 10 consecutive frames of the fire dynamic map to extract fire change characteristics; spatial interpolation: use Kriging interpolation to fill sensor blind zone data;

[0085] LSTM (Long Short-Term Memory) network is used to predict the fire situation in the next 5 minutes. Input: 10 frames of historical fire dynamic data; Output: prediction results for the next 5 frames; Trend region division: High-risk area: temperature > 60℃ or smoke concentration > 1.0 mg / m³; Medium-risk area: temperature 40-60℃ or smoke concentration 0.5-1.0 mg / m³; Low-risk area: temperature < 40℃ and smoke concentration < 0.5 mg / m³.

[0086] Specifically, the fire drill at Hospital A's B campus included the following: Fire Dynamic Map Acquisition: The system received a fire dynamic map from the 3rd-floor pharmacy warehouse (F01), showing: Fire source location: northeast corner of the pharmacy warehouse; Smoke concentration: 0.8 mg / m³ (threshold: 0.5 mg / m³); Spread direction: spreading towards the southwest corridor; Temperature rise: 2.5℃ / min; Trend prediction: The LSTM model predicted that after 5 minutes: the high-risk area would expand to the west corridor of the pharmacy warehouse (an additional 15㎡); the medium-risk area would cover the area north of the nurses' station (an additional 20㎡); Trend area marking: The system marked the risk areas on the indoor distribution map with red, yellow, and green colors.

[0087] Path planning algorithm: Improved Algorithm A, considering: avoiding fire areas (weight: 0.8); path length (weight: 0.5); congestion level (weight: 0.3); path generation rules: starting point: the current location of the hospitalized patient (e.g., ICU ward); ending point: the nearest safe exit; constraints: cannot cross high-risk areas; avoid medium-risk areas as much as possible; prioritize wide corridors; simultaneously, perform spatial overlay analysis of the path and the fire area; calculate the minimum distance between the path and the high-risk area (requirement > 2 meters); check if the path crosses a medium-risk area (crossing is allowed but time factor is increased); dynamic adjustment mechanism: if a potential conflict is detected, automatically replan the path; prioritize vertical evacuation (stairs) over horizontal evacuation.

[0088] Specifically, the input conditions are: Starting point: ICU ward (3rd floor, east side); End point: West safety exit; Fire area: Drug warehouse (high risk), north side of nurse station (medium risk); Path generation: Path 1: ICU → East corridor → North staircase → 2nd floor → West exit (length: 85m, estimated time: 3.5 minutes); Path 2: ICU → South corridor → West staircase → 1st floor → West exit (length: 92m, estimated time: 4 minutes); Path 3: ICU → East corridor → South staircase → 1st floor → East exit (length: 110m, estimated time: 4.5 minutes).

[0089] Route verification: Route 1: Minimum distance to high-risk area is 2.3 meters (safe), does not cross medium-risk area; Route 2: Minimum distance to high-risk area is 1.8 meters (unsafe), needs adjustment; Route 3: Crosses medium-risk area by 15 meters (time factor × 1.2); Route adjustment: Route 2 is modified to: ICU → South corridor → West staircase → 1st floor → West exit (after adjustment, distance to high-risk area is 2.1 meters).

[0090] Path execution results: Path 1 was selected as the optimal path, and 5 inpatients in the ICU were successfully evacuated; Actual time: 3 minutes and 42 seconds, with an error of less than 7% compared to the prediction; Safety verification: A safe distance was maintained from the fire source throughout the process, and no contact with smoke was encountered; System performance: Path planning time: 1.2 seconds; Path adjustment response time: 0.8 seconds; Success rate: 100% (all 10 drills were successful).

[0091] Therefore, a hospital fire status map is constructed based on the fire dynamic map and fire area. Simultaneously, a first route event is determined based on multiple emergency evacuation routes and the current location of hospitalized patients. A second route event is determined based on the current location of hospitalized patients and the hospital fire status map. The optimal emergency evacuation route is then determined based on the first route event, the second route event, and multiple emergency evacuation routes. This approach considers the first route event, the second route event, and multiple emergency evacuation routes holistically, ensuring the accuracy of the optimal emergency evacuation route. Furthermore, the introduction of fire areas, which considers multiple emergency evacuation routes, the current location of hospitalized patients, and the hospital fire status map holistically, further improves the accuracy of the optimal emergency evacuation route.

[0092] At this point, a hospital fire status map is constructed based on the fire dynamic map and fire areas. Input data: Fire dynamic map (from S123): including fire source location, smoke concentration, temperature changes, etc.; Fire areas (from S122): such as FR-merged (pharmacy warehouse + power distribution room); Fusion algorithm: using graph neural network (GNN) to connect discrete fire areas into a continuous status map; Nodes: each fire area (such as FR-merged); Edges: fire propagation relationships between areas (such as propagation probability, time delay); Status map features: Spatial dimension: including building structure information (such as corridors, stairs, fire doors); Temporal dimension: dynamically updated fire development (refreshed every 10 seconds); Risk level: represented by color coding (red: high risk, yellow: medium risk, green: safe).

[0093] Specifically, the fire status diagram is constructed as follows: Node 1: FR-merged (3rd floor medicine warehouse + power distribution room); Node 2: FR-corridor (3rd floor southwest corridor); Edge: FR-merged → FR-corridor (propagation probability 0.8, delay 2 minutes); Risk level: FR-merged: Red (temperature > 80℃, smoke concentration > 1.0mg / m³); FR-corridor: Yellow (temperature 60℃, smoke concentration 0.6mg / m³).

[0094] The first path event is determined based on multiple emergency evacuation routes and the current location of hospitalized patients. The first path event is the initial evacuation plan based on static path planning. Input data: multiple emergency evacuation routes (from S132): such as route 1, route 2, and route 3; current location of hospitalized patients: obtained through the hospital's positioning system (such as WiFi positioning and RFID tags); event generation algorithm: multi-objective optimization: objective 1: shortest path length (Dijkstra's algorithm); objective 2: avoid fire areas (an improved version of the A algorithm); objective 3: consider the patient's movement speed (such as the speed of a stretcher patient at 0.5 m / s).

[0095] Specifically, the inpatient's location is: ICU ward on the 3rd floor (coordinates: X=120, Y=85); First path event generation: Path 1: ICU → East Corridor → Staircase 1 → East Exit on the 1st floor; Length: 85 meters; Estimated time: 2 minutes and 50 seconds (calculated based on stretcher speed); Avoidance area: Completely avoid FR-merged and FR-corridor; Path 2: ICU → South Corridor → Staircase 2 → South Exit on the 1st floor; Length: 92 meters; Estimated time: 3 minutes and 4 seconds; Avoidance area: Requires crossing the edge of the FR-corridor (risk factor 0.3).

[0096] The second path event is determined based on the current location of the inpatient and the hospital fire status map. The second path event is a dynamic adjustment scheme based on the real-time fire status map. Input data: current location of the inpatient (updated in real time); hospital fire status map (refreshed every 10 seconds). Adjustment algorithm: real-time risk assessment: calculate the risk index (RI) of each point on the path: RI = smoke concentration × 0.4 + temperature × 0.3 + propagation probability × 0.3; dynamic replanning: trigger path replanning when RI > 0.7; incremental path update is performed using the DLite algorithm.

[0097] Specifically, real-time status update: T+2 minutes: FR-corridor risk level rises to red (RI=0.85); patient location: moved to the middle of the east corridor (X=150, Y=85); second path event generation: original path 2 assessment: RI=0.85>threshold, needs adjustment; new path 2': ICU→east corridor→stairs 3→1st floor north exit; length: 98 meters; estimated time: 3 minutes 16 seconds; risk index: RI=0.2 (safe).

[0098] The optimal emergency evacuation route is determined based on the first path event, the second path event, and multiple emergency evacuation routes. The evaluation dimensions are: safety (weight 0.5): risk index, number of fire areas to avoid; timeliness (weight 0.3): estimated evacuation time; feasibility (weight 0.2): path width, slope, and obstacle conditions. The decision algorithm adopts the TOPSIS method: constructing a decision matrix (path × evaluation index); calculating the closeness of each path to the ideal solution; and selecting the path with the highest closeness as the optimal solution.

[0099] Specifically, a schematic diagram of the path evaluation matrix is ​​shown in Table 2:

[0100] Table 2. Schematic diagram of the path evaluation matrix

[0101]

[0102] Optimal route selection: Route 1 had the highest overall score (0.905); Final decision: Route 1 was selected as the optimal emergency evacuation route; Exercise results and verification: 5 hospitalized patients were successfully evacuated via Route 1; Actual time: 2 minutes and 55 seconds (error from prediction <2%); Safety indicators: Overall risk index <0.3; System performance: State diagram update frequency: 10 seconds / time; Route replanning response time: <1 second; Decision accuracy: 95% (optimal route was selected 19 times out of 20 exercises).

[0103] refer to Figure 5 In step S14, the specific steps are as follows:

[0104] S141: Based on the detection of the best emergency evacuation route, multiple emergency evacuation routes are determined, and corresponding emergency evacuation nodes are determined according to the multiple emergency evacuation routes, the corresponding fire areas, and the indoor distribution map of the hospital, so as to collect multiple emergency evacuation nodes.

[0105] S142: Among multiple emergency evacuation nodes, the node position of each emergency evacuation node is determined based on the detection of each emergency evacuation node; multiple fire smoke features are determined based on the smoke detection of the node positions of each emergency evacuation node; and the corresponding fire smoke area is determined based on the feature position, corresponding feature shape and emergency evacuation route of the multiple fire smoke features.

[0106] S143: Collect the movement status of hospitalized patients, determine the first level of evacuation action instructions based on the movement status of hospitalized patients, their corresponding current location and multiple fire smoke areas, determine the second level of evacuation action instructions based on the node locations of multiple emergency evacuation nodes and their corresponding fire smoke areas, determine the evacuation action instructions for hospitalized patients based on the first level of evacuation action instructions and the second level of evacuation action instructions, and transmit the evacuation action instructions along the Internet of Things to the corresponding emergency response device to guide the emergency evacuation behavior of hospitalized patients.

[0107] In the embodiments of this application, multiple emergency evacuation routes are determined based on the detection of the optimal emergency evacuation route. The corresponding emergency evacuation nodes are determined according to the multiple emergency evacuation routes, the corresponding fire areas, and the indoor distribution map of the hospital. This method collects multiple emergency evacuation nodes, taking into account the overall consideration of multiple emergency evacuation routes, the corresponding fire areas, and the indoor distribution map of the hospital, thus ensuring the accuracy of the corresponding emergency evacuation nodes.

[0108] At this point, input: Optimal emergency evacuation route (from S133); Analysis dimensions: Route length: Divide long routes into manageable shorter segments; Building structure: Based on natural dividing points such as rooms, corridors, and staircases; Safety requirements: Ensure each segment is within a fire compartment; Key point identification: Automatically identify structural transition points on the route (such as doors and corners); Identify potential risk points (such as narrow passages or areas near flammable materials); Segment division criteria: Length control: Each segment should be 15-20 meters (meeting hospital evacuation standards); Safety: Avoid crossing fire compartment boundaries; Identifiability: Each segment should have clear start and end point markers.

[0109] Emergency evacuation nodes are determined based on emergency evacuation routes, fire areas, and indoor distribution maps. Input data includes: emergency evacuation routes; fire area information (from S122); hospital indoor distribution map (from S131). Node type definitions: Decision nodes: points requiring directional selection (e.g., corridor intersections); Safety nodes: temporary refuge points (e.g., fire compartment boundaries); Destination nodes: final safe areas (e.g., outdoor assembly points). Node attribute calculations include: Spatial coordinates: precise location based on the indoor distribution map; Safety level: calculated based on distance from the fire area; Accessibility: assessed based on width and historical data. Node selection criteria include: Safety: node location must be within a safe area; Accessibility: ensures accessibility from adjacent routes; Identifiability: node location should be clearly marked or easily described.

[0110] Node data acquisition methods: Automatic acquisition: IoT sensors: environmental sensors deployed at each node; camera monitoring: real-time acquisition of node status images; RFID tags: precise identification of node location; Data content: Basic information: node ID, name, type; precise spatial coordinates (x, y, z); Real-time status: current personnel density; environmental parameters (temperature, smoke concentration, etc.); traffic status (whether congested); Related information: list of neighboring nodes; estimated time to reach neighboring nodes; available auxiliary facilities (such as wheelchairs, stretchers).

[0111] Furthermore, among multiple emergency evacuation nodes, the node location of each emergency evacuation node is determined based on the detection of each node; multiple fire smoke features are determined based on the smoke detection of the node locations of each emergency evacuation node; and the corresponding fire smoke area is determined based on the feature location, corresponding feature shape, and emergency evacuation route of the multiple fire smoke features. This comprehensive consideration of the feature location, corresponding feature shape, and emergency evacuation route of multiple fire smoke features ensures the accuracy of the corresponding fire smoke area.

[0112] At this point, multi-sensor fusion positioning is achieved through: camera visual positioning (using indoor distribution maps and real-time images for feature matching); Bluetooth beacon positioning (using Bluetooth beacons deployed in the hospital for triangulation); inertial navigation (using accelerometers and gyroscopes to assist positioning in areas with weak signals); position accuracy control (visual positioning accuracy: ±0.3 meters; Bluetooth positioning accuracy: ±1 meter; fusion positioning accuracy: ±0.5 meters); and coordinate system (adopting the unified coordinate system of the hospital's BIM model).

[0113] Based on smoke detection at the locations of various emergency evacuation nodes, multiple fire smoke characteristics are determined, including: visible light image analysis (detecting visual features of smoke, color, texture, transparency); infrared thermal imaging (detecting smoke temperature distribution); and laser scattering (measuring smoke particle concentration). Feature types include: location features (coordinates of the smoke center); morphological features (smoke shape, diffusion type, rising type, vortex type); concentration features (smoke density level, 1-5); and motion features (smoke diffusion speed and direction).

[0114] Based on the characteristic locations, corresponding characteristic shapes, and emergency evacuation routes of multiple fire smoke features, the corresponding fire smoke areas are determined. For the calculation of smoke areas, the following spatial interpolation algorithms are used: Inverse Distance Weighted Method (IDW): This method extrapolates unknown areas based on known smoke points; Kriging interpolation: This is the optimal interpolation method considering spatial correlation; For area boundaries, a dynamic contour algorithm is used: This algorithm generates area boundaries based on smoke concentration contour lines; The influence of ventilation conditions and building structure on smoke diffusion is considered. Risk level classification: High-risk area: Smoke concentration > 3, visibility < 5 meters; Medium-risk area: Smoke concentration 2-3, visibility 5-10 meters; Low-risk area: Smoke concentration < 2, visibility > 10 meters.

[0115] Therefore, the system collects the movement status of hospitalized patients, determines the first level of evacuation action instructions based on the patients' movement status, their current location, and multiple fire smoke areas, and determines the second level of evacuation action instructions based on the node locations of multiple emergency evacuation nodes and their corresponding fire smoke areas. Based on the first and second level evacuation action instructions, the system determines the evacuation action instructions for hospitalized patients. At the same time, the evacuation action instructions are transmitted along the Internet of Things to the corresponding emergency response devices, guiding the emergency evacuation behavior of hospitalized patients. This system takes into account both the first and second level evacuation action instructions, ensuring the accuracy of the evacuation action instructions for hospitalized patients.

[0116] At this time, the patient status is collected through wearable devices: heart rate monitoring: real-time collection of patient heart rate data; activity monitoring: detection of patient movement status through accelerometer; vital signs: monitoring of key indicators such as blood oxygen and blood pressure; status classification criteria: fully autonomous: able to walk independently (such as patients with mild symptoms); partially assisted: requires assistance or the use of assistive devices (such as canes); fully dependent: requires a stretcher or wheelchair (such as patients with severe symptoms).

[0117] The first level of evacuation action is determined based on the inpatient's activity status, corresponding current location, and multiple fire smoke areas. Input parameters: patient activity status; patient current location (from S142); fire smoke area distribution (from S142); decision logic: safety assessment: calculate the shortest path from the patient's current location to the safe area; assess the smoke risk level along the path; consider the impact of the patient's mobility on the path; instruction generation rules: fully autonomous patients: select the fastest path, avoiding high-risk areas; partially assisted patients: select the safest path, considering the passage of assistive devices; fully dependent patients: select the shortest path, prioritizing the use of elevators (when safe).

[0118] The second level of evacuation action instructions is determined based on the location of multiple emergency evacuation nodes and their corresponding fire smoke areas. Input parameters: location of emergency evacuation nodes (from S142); fire smoke area corresponding to each node (from S142); decision logic: node risk assessment: assess the smoke risk of each evacuation node; calculate the passage safety between nodes; consider node capacity and congestion; instruction generation rules: high-risk nodes: detour or wait; medium-risk nodes: pass quickly; low-risk nodes: pass normally.

[0119] Evacuation instructions for hospitalized patients are determined based on the first and second levels of evacuation action instructions, with the following weighting: First level instruction weight: 60% (considering individual patient circumstances); Second level instruction weight: 40% (considering environmental risks). Conflict handling: When the two levels of instructions conflict: patient safety is prioritized; the lowest-risk option is selected; and a safety margin is extended. Simultaneously, the evacuation action instructions are transmitted via the Internet of Things (IoT) to the corresponding emergency response device, guiding the emergency evacuation behavior of hospitalized patients. Wireless network: 5G / WiFi 6 dual backup; IoT protocol: MQTT / CoAP low-power protocol; Response device types: Ward terminal: displaying text instructions; Wearable device: vibration alert; Broadcast system: voice announcement; Indicator light: LED directional guidance; Guidance mechanism: Real-time feedback: patient location tracking; instruction execution confirmation; Dynamic path adjustment.

[0120] refer to Figure 6 In step S15, the specific steps are as follows:

[0121] S151: Collect the current location of hospitalized patients and monitor their evacuation behavior in real time. At the same time, firefighters enter the hospital to carry out simultaneous firefighting in various fire areas and collect the location of firefighters. Based on the evacuation behavior of hospitalized patients, the location of firefighters, and the amount of firefighting in each fire area, determine the corresponding fire and fire-fighting events.

[0122] S152: Based on the identification of fire-fighting events, identify multiple fire-fighting items and mark the corresponding item content. Determine the first rescue event based on the item content of multiple fire-fighting items and the current location of the hospitalized patient.

[0123] S153: Monitor each fire zone in real time and collect fire change events in each fire zone. Determine the second rescue event based on the current location of the hospitalized patient and the fire change events in each fire zone. Determine the multi-level rescue event for the hospitalized patient based on the first rescue event, the second rescue event and the patient's action status.

[0124] In the embodiments of this application, the current location of hospitalized patients is collected, and their evacuation behavior is monitored in real time. At the same time, firefighters enter the hospital to carry out simultaneous firefighting in various fire areas, and the location of the firefighters is collected. Based on the evacuation behavior of hospitalized patients, the location of firefighters, and the amount of firefighting in each fire area, the corresponding fire and fire-fighting events are determined. This approach takes into account the overall consideration of the evacuation behavior of hospitalized patients, the location of firefighters, and the amount of firefighting in each fire area, ensuring the accuracy of the corresponding fire and fire-fighting events.

[0125] At this time, the current location of the hospitalized patient is collected, and Bluetooth beacon-assisted positioning provides supplementary positioning in areas with signal obstruction (such as elevator shafts and equipment rooms); visual recognition assistance uses cameras to identify patient identifiers (such as wristband color and bed number) in real time; wearable device positioning uses the smart bracelet worn by the patient to upload location data in real time.

[0126] Real-time monitoring of inpatient evacuation behavior, monitoring content includes: movement speed (calculated using an accelerometer on a wearable device); movement direction (determined by combining continuous location data); dwell time (recording the duration of stay at key points); abnormal behavior (such as turning back, prolonged stay, deviation from the planned route, etc.); behavior analysis algorithm: speed threshold judgment.

[0127] Normal speed: 0.5-1.5 m / s (stretcher transfer); Abnormal speed: <0.2 m / s (may be obstructed); Directional consistency check: Calculate the angle between the actual direction of movement and the predetermined route; if the angle is >45°, it is marked as a deviation from the route; Dwell time analysis: Single point dwell time >30 seconds triggers an alarm; Dwell time at critical nodes (such as stairwells) >60 seconds triggers an advanced alarm.

[0128] Firefighters, upon entering the hospital, simultaneously extinguished fires in various fire areas and collected firefighter locations. Firefighter deployment included: group entry (dividing firefighters into 3-5 groups based on fire area distribution); simultaneous firefighting (each group simultaneously entering different fire areas); equipment configuration (each group equipped with fire extinguishers, water hoses, breaching tools, etc.); firefighting operation procedures: area assessment (firefighters quickly assess the fire situation upon arrival); firefighting plan selection: small fire: use fire extinguishers; medium fire: use water hoses; large fire: use fire monitors; firefighting execution: execute firefighting operations according to the plan; effectiveness evaluation: assess firefighting effectiveness using temperature sensors and cameras; firefighters were equipped with explosion-proof positioning terminals; inertial navigation-assisted positioning was used in complex environments; location accuracy was verified through cameras in the command center. Optionally, firefighters could also serve as personnel conducting fire rescue operations within the hospital.

[0129] Fire events are determined based on the evacuation behavior of hospitalized patients, the location of firefighters, and the amount of fire extinguishing in each fire zone. Input parameters include: patient evacuation behavior (speed, direction, dwell time); firefighter location (coordinates, direction of movement); and fire extinguishing volume (water flow rate, extinguishing agent dosage, extinguishing effect). Judgment rules are as follows: Emergency level determination: High emergency: patient speed < 0.2 m / s and firefighter distance > 50 meters; Medium emergency: patient speed 0.2-0.5 m / s and firefighter distance 30-50 meters; Low emergency: patient speed > 0.5 m / s and firefighter distance < 30 meters. Extinguishing effect determination: Effective: temperature drop > 20℃; Partially effective: temperature drop 10-20℃; Ineffective: temperature drop < 10℃. Comprehensive judgment: event level is determined by combining emergency level and extinguishing effect. Event generation process: Collect all input parameters; analyze the current status of each parameter; match the parameters with the judgment rules; generate the corresponding event based on the matching results; and publish the event to relevant systems.

[0130] Furthermore, multiple fire-fighting items are identified based on the identification of fire-fighting events, and the corresponding item contents are marked. The first rescue event is determined based on the item contents of multiple fire-fighting items and the current location of the hospitalized patient. This comprehensive consideration of the item contents of multiple fire-fighting items and the current location of the hospitalized patient ensures the accuracy of the first rescue event.

[0131] At this point, based on the identification of fire and fire-fighting events, multiple fire and fire-fighting items are determined and their corresponding contents are marked. Fire and fire-fighting event input: Source: Fire and fire-fighting event generated in step S151, for example: Event ID: FE-2024-001; Event type: Routine fire event; Event description: "Patient evacuation in ICU-205 area is normal, firefighters have arrived, and the fire extinguishing effect is good"; Handling suggestion: "Continue evacuation according to plan, maintain current fire-fighting intensity"; Fire and fire-fighting item identification logic: The system extracts key elements through a Natural Language Processing (NLP) model based on the event description and handling suggestion to identify the corresponding fire and fire-fighting items; Fire and fire-fighting items include, but are not limited to: extinguishing fires... Fire progress items (e.g., "Fire in progress", "Extinguished", "Fire spreading"); evacuation support items (e.g., "Passage clear", "Passage blocked", "Smoke spreading"); medical support items (e.g., "Oxygen support needed", "Stretcher transport needed", "Abnormal vital signs"); resource allocation items (e.g., "Reinforcements for firefighters", "Request for medical support"). Additionally, a project labeling mechanism is in place: each item is assigned a unique project number (e.g., FP-001) and a description. Project content includes: project type (firefighting, evacuation, medical, resource); priority (high, medium, low); responsible department (firefighting team, medical team, command center); and execution status (pending execution, in progress, completed).

[0132] Specifically, the input event is FE-2024-001; key elements are extracted: "Patient evacuation in ICU-205 area is normal" → evacuation support project; "Firefighters have arrived" → resource scheduling project; "Firefighting effect is good" → firefighting progress project; generated projects are: FP-001: firefighting progress project, medium priority, status is in progress; FP-002: evacuation support project, high priority, status is in progress; FP-003: resource scheduling project, low priority, status is completed.

[0133] The system determines the first rescue event based on the project content of multiple fire-fighting projects and the current location of the hospitalized patient. Input information includes: multiple fire-fighting projects (e.g., FP-001, FP-002, FP-003); the current location of the hospitalized patient (e.g., ICU-205 ward, coordinates (X:1350, Y:760)). The event generation logic is as follows: the system assesses the current rescue needs based on the project content and the patient's location, generating the first rescue event. Assessment dimensions include: whether the patient is in a high-risk area (e.g., near a fire source, in a smoke-filled area); whether the patient requires immediate medical intervention (e.g., abnormal vital signs); whether evacuation routes are safe (e.g., FP-002 status); and whether fire-fighting resources are sufficient (e.g., FP-003 status). The event structure includes: event number (e.g., RE-001); event type (e.g., "emergency rescue," "routine transfer," "continuous observation"); rescue recommendations (e.g., "immediate transfer," "provide oxygen," "wait in place"); executing entity (e.g., "medical team A," "stretcher team B"); and target location (e.g., "safe zone A," "temporary rescue point B").

[0134] Specifically, the input items are: FP-001: Firefighting progress in progress, fire in ICU-205 is under control; FP-002: Evacuation is underway, passageways are clear; FP-003: Resource allocation completed, firefighters are in place; Patient location: ICU-205, coordinates (X:1350, Y:760); Assessment results: Patient is not in a high-risk area; passageways are clear, safe transfer is possible; fire is under control, no emergency evacuation is required; First rescue event is generated: Event number: RE-001; Event type: routine transfer; Rescue suggestion: "Transfer to temporary rescue point along the safe passage"; Executing entity: Stretcher team B; Target location: Temporary rescue point B (coordinates (X:1500, Y:900)); Release time: 09:16:00.

[0135] Therefore, real-time monitoring of each fire zone and collection of fire change events in each fire zone are used to determine the second rescue event based on the current location of the hospitalized patient and the fire change events in each fire zone. Based on the first rescue event, the second rescue event, and the patient's action status, multi-level rescue events for the hospitalized patient are determined. This comprehensive consideration of the first rescue event, the second rescue event, and the patient's action status ensures the accuracy of the multi-level rescue events for the hospitalized patient. At the same time, the introduction of evacuation action instructions for the hospitalized patient realizes the comprehensive consideration of multiple fire protection projects, the current location of the hospitalized patient, and the fire change events in each fire zone, improving the accuracy of the multi-level rescue events for the hospitalized patient and realizing full-process rescue for the hospitalized patient.

[0136] At this time, real-time monitoring of each fire area is conducted, and fire change events in each fire area are collected. Based on the current location of hospitalized patients and the fire change events in each fire area, a second rescue event is determined. The monitoring system architecture is as follows: multi-source data fusion: temperature sensors: deployed on the ceiling and walls to monitor changes in ambient temperature in real time; smoke concentration sensors: update smoke concentration data every 30 seconds; infrared thermal imager: captures the location of the fire source and the range of heat radiation in real time; camera system: monitors the spread of flames and smoke through image recognition technology.

[0137] The system is configured with threshold triggering mechanisms: Temperature threshold: >60°C triggers a high-temperature alarm; Smoke concentration: >0.5mg / m³ triggers a smoke alarm; Flame recognition: Detection of flame features for 3 consecutive frames triggers a fire alarm; Event acquisition content: Event ID (e.g., CE-001); Occurrence time (accurate to the second); Location coordinates (X, Y coordinates); Change type (new fire point, fire spread, smoke diffusion, sudden temperature rise); Change severity (slight, moderate, severe); Data update frequency: Regular monitoring: updated every 30 seconds; Alarm status: updated every 5 seconds; Emergency status: updated in real time (1-second interval).

[0138] Input data: Real-time patient location (from S151); List of fire event changes; Spatial analysis: Calculate the straight-line distance between the patient and each fire event point; Analyze the ventilation and smoke diffusion direction of the patient's area; Assess potential risk points on the patient's evacuation path; Second rescue event generation rules: Risk level determination: High risk: Patient is located within the fire-affected area (<10 meters); Medium risk: Patient is located on the smoke diffusion path (10-30 meters); Low risk: Patient is located in a safe area (>30 meters); Rescue decision logic: High risk: Evacuate immediately, highest priority; Medium risk: Prepare for evacuation, strengthen monitoring; Low risk: Continue observation, prepare contingency plans; Event generation format: Event number: SE-001; Event type: Emergency evacuation / Preparing for evacuation / Continuing observation; Rescue suggestions: Specific action instructions; Estimated impact time: The possible duration of the event; Suggested action time: The suggested time window for carrying out rescue.

[0139] Multi-level rescue events are determined based on the first rescue event, the second rescue event, and the patient's movement status. Input data: First rescue event (from S152): rescue suggestions based on fire project; Second rescue event (from sub-step 2): rescue suggestions based on changes in the fire; Person's movement status: movement speed, direction, and changes in vital signs; Decision matrix: Priority weight allocation: Life safety: 40%; Medical needs: 30%; Evacuation efficiency: 20%; Resource consumption: 10%; Comprehensive score calculation: Score = Σ(Score of each factor × weight).

[0140] Multi-level rescue event structure: Level 1 Rescue (Immediate Execution): Triggering condition: Overall score > 80 points; Characteristics: Highest priority, must be executed immediately; Example: Emergency CPR + rapid evacuation; Level 2 Rescue (Ready to Execute): Triggering condition: 60 points ≤ Overall score ≤ 80 points; Characteristics: Requires preparation, but not necessarily immediate execution; Example: Prepare oxygen equipment, plan evacuation routes; Level 3 Rescue (Contingency Plan Preparation): Triggering condition: Overall score < 60 points; Characteristics: Develop contingency plans, not yet executed; Example: Develop backup evacuation routes, prepare emergency supplies; Dynamic adjustment mechanism: Real-time update: Reassessed every 10 seconds; Status tracking: Records the execution status of each rescue event; Automatic escalation: Automatically escalates the rescue level when conditions worsen.

[0141] Specifically, at 09:20:00 on March 15, 2024, a fire drill was officially launched in the ICU area on the 3rd floor of the inpatient department of Hospital B in Area A. The system first detected a sharp rise in ambient temperature to 85°C through temperature sensors deployed in the equipment room, and at the same time, the smoke sensor showed a concentration of 0.8 mg / m³, triggering the early warning mechanism. Immediately afterwards, the camera identified an open flame in the equipment room in real time, and the system immediately generated a fire change event, with event ID CE-003, located at coordinates (X:1200, Y:650) in the equipment room, the change type was "fire expansion", and the degree of change was assessed as "severe". This event marked that the fire had entered a substantial development stage, and the system then initiated the subsequent rescue event generation process.

[0142] Within 5 seconds of the fire change event being generated, the system quickly located critically ill patient P001 in ICU-205, with coordinates (X:1250, Y:680), only 15 meters from the fire and directly in the smoke diffusion path. Using a risk assessment model, the system determined that the patient faced a moderate risk, with smoke expected to reach the area shortly. Based on this, the system generated a second rescue event, SE-002, typed "Prepare for Evacuation," with a rescue recommendation explicitly stating "Immediately prepare a mobile ventilator and evacuate within 5 minutes," an estimated impact time of 10 minutes, and a recommended action time of "immediate." This event provided crucial information for subsequent multi-level rescue decisions.

[0143] At 09:20:10, the system comprehensively assessed the patient P001's condition by integrating the first rescue event (routine transfer to a temporary rescue point) from step S152 and the currently generated second rescue event (preparing for evacuation, requiring ventilator support). The assessment results showed that the patient's vital signs were stable but movement was slow, with a life safety score of 85 (requiring ventilator support), a medical needs score of 70 (requiring basic life support), an evacuation efficiency score of 60 (slow movement), and a resource consumption score of 50 (requiring 2 medical personnel). The overall score was 72 points. Based on this, the system generated multiple levels of rescue events: Level 1 rescue: none; Level 2 rescue event ID: ME-001, rescue content: "Prepare ventilator, 2 medical staff assist, evacuate along the safety passage", execution time: before 09:25:00; Level 3 rescue event ID: ME-002, rescue content: "Prepare backup ventilator, contact the emergency department for reception", execution time: before 09:30:00. This series of events ensured the safe evacuation of patients and the continuity of medical support in the event of a fire.

[0144] During the drill, medical staff responded swiftly to the multi-level rescue events generated by the system, completing the preparation of the mobile ventilator at 09:22:00. With the assistance of two medical staff members, patient P001 was successfully evacuated to the temporary rescue point along the safety passage at 09:24:30. Simultaneously, the backup ventilator was ready, and the emergency department was prepared to receive the patient. The entire drill verified the accuracy and timeliness of the system's generation of rescue events for critically ill patients in fire situations, providing valuable experience for actual fire emergency response. This drill further optimized the rescue event generation process and improved the hospital's emergency response capabilities in sudden events such as fires.

[0145] Please see Figure 7 , Figure 7 This is a schematic diagram of the structural composition of an IoT-based hospital fire emergency evacuation system according to an embodiment of the present invention; the IoT-based hospital fire emergency evacuation system includes:

[0146] Anomaly feature module 21 is used to determine indoor images based on multiple cameras and the Internet of Things in the hospital, and to determine multiple anomalous object features based on the recognition of indoor images;

[0147] The fire dynamic map module 22 is used to determine the corresponding fire warning level based on the feature location and feature shape of multiple abnormal objects, determine the fire area based on the warning level of multiple abnormal objects and the corresponding smoke data, and determine the corresponding fire dynamic map.

[0148] The emergency evacuation route module 23 is used to determine multiple emergency evacuation routes based on the hospital's indoor distribution map and the regional location of the fire area, and to determine the optimal emergency evacuation route based on the multiple emergency evacuation routes, the current location of hospitalized patients, and the hospital fire status map.

[0149] The evacuation action instruction module 24 is used to determine multiple emergency evacuation nodes based on the identification of the best emergency evacuation route, determine the evacuation action instructions of the hospitalized patients based on the node location of the multiple emergency evacuation nodes, the corresponding fire smoke area and the action status of the hospitalized patients, and guide the emergency evacuation behavior of the hospitalized patients based on the Internet of Things.

[0150] The multi-level rescue event module 25 is used to determine the corresponding fire-fighting events based on the evacuation behavior of hospitalized patients and the location of firefighters, to determine multiple fire-fighting items based on the identification of fire-fighting events, and to determine the multi-level rescue events for hospitalized patients based on multiple fire-fighting items, the current location of hospitalized patients, and fire change events in various fire areas.

[0151] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A hospital fire emergency evacuation method based on the Internet of Things, characterized in that, include: Indoor images are determined based on multiple cameras and the Internet of Things in the hospital. Based on the recognition of indoor images, the characteristics of multiple abnormal objects are identified, including smoke, flames, and high-temperature equipment. The fire warning level is determined based on the characteristic location and shape of multiple anomalous objects. The fire area is determined based on the fire warning level of the multiple anomalous objects and the corresponding smoke data, and a corresponding fire dynamic map is generated. This includes: determining the corresponding smoke data based on smoke detection of the characteristic locations of each anomalous object; simultaneously, determining a first fire information combination based on the relative positions of multiple anomalous objects and the corresponding smoke data; determining a second fire information combination based on the fire warning level of the multiple anomalous objects and the corresponding smoke data; determining the fire area based on the first and second fire information combinations; and within the fire area, determining a corresponding fire dynamic map based on monitoring of each fire area. The fire dynamic map includes the following information: changes in the fire area boundaries, smoke diffusion direction, and changes in risk level. Multiple emergency evacuation routes are determined based on the hospital's indoor layout map and the location of the fire area. The optimal emergency evacuation route is then determined based on these routes, the current location of hospitalized patients, and the hospital's fire status map. A graph neural network is used to connect discrete fire areas into a continuous fire status map, where nodes represent each fire area and edges represent the fire propagation relationship between fire areas. Multiple emergency evacuation nodes are identified based on the optimal emergency evacuation route. Evacuation action instructions for hospitalized patients are determined based on the node locations of multiple emergency evacuation nodes, the corresponding fire smoke areas, and the movement status of hospitalized patients. Emergency evacuation behavior of hospitalized patients is guided based on the Internet of Things. In the fire smoke area, the boundary of the fire smoke area is generated based on the smoke concentration contour lines. Fire and fire-fighting events are determined based on the emergency evacuation behavior of hospitalized patients, the location of firefighters, and the fire extinguishing capacity of each fire zone. Multiple fire and fire-fighting items are identified based on these events. Multi-level rescue events for hospitalized patients are then determined based on these multiple fire and fire-fighting items, the current location of hospitalized patients, and fire change events in each fire zone. The fire and fire-fighting event generation process involves: collecting all input parameters, analyzing the current state of each parameter, matching each parameter with judgment rules, and generating corresponding fire and fire-fighting events based on the matching results. The input parameters include the emergency evacuation behavior of hospitalized patients, the location of firefighters, and the fire extinguishing capacity of each fire zone. Each fire and fire-fighting event includes an event ID, event type, event description, and handling suggestions.

2. The hospital fire emergency evacuation method based on the Internet of Things according to claim 1, characterized in that, The system uses multiple cameras and the Internet of Things (IoT) in the hospital to determine indoor images, and identifies multiple abnormal object features based on the recognition of these images, including: Cameras are installed in different locations in the hospital. Multiple cameras communicate with each other through the Internet of Things (IoT). The multiple cameras dynamically capture images of the hospital's interior space and collect images of multiple sub-rooms. The interior image is determined based on the shooting location of the multiple sub-room images, the images of two adjacent sub-rooms, and the image synthesis mechanism corresponding to the IoT. Multiple object monitoring areas are determined based on indoor image detection. Within these areas, the shape of the object is determined by identifying each monitoring area. Based on the object's shape, its corresponding abnormal location, and surrounding environmental features, corresponding abnormal object features are determined to collect multiple abnormal object features.

3. The hospital fire emergency evacuation method based on the Internet of Things according to claim 1, characterized in that, The method of determining the corresponding fire warning level based on the feature location and feature shape of multiple abnormal objects includes: The system monitors the location of multiple anomalous object features in real time, and determines the corresponding fire warning level based on the feature location, corresponding feature shape, and relative distance between two adjacent anomalous object features, thereby marking the fire warning level of each anomalous object feature.

4. The hospital fire emergency evacuation method based on the Internet of Things according to claim 1, characterized in that, The process involves determining multiple emergency evacuation routes based on the hospital's indoor layout map and the location of the fire area. Then, based on these multiple emergency evacuation routes, the current locations of hospitalized patients, and the hospital fire status map, the optimal emergency evacuation route is determined, including: The system collects the hospital's location and database, and determines the hospital's indoor layout based on the hospital's location and database. Based on the hospital's indoor layout and the regional location of the fire area, it determines the corresponding fire-affected area. Specifically, based on the regional location of the fire area, it calculates its impact range to determine the fire-affected area. Collect dynamic fire maps, determine the fire trend areas based on the hospital's indoor layout map and the corresponding dynamic fire maps, and determine multiple emergency evacuation routes based on the fire impact area, fire trend areas, and fire areas. Each emergency evacuation route can avoid the fire area along its corresponding direction. In particular, predict future fire changes based on historical dynamic fire map data to determine the fire trend areas.

5. The hospital fire emergency evacuation method based on the Internet of Things according to claim 4, characterized in that, The process of determining multiple emergency evacuation routes based on the hospital's indoor layout map and the location of the fire area, and determining the optimal emergency evacuation route based on these multiple routes, the current location of hospitalized patients, and the hospital fire status map, also includes: A hospital fire status map is constructed based on the fire dynamic map and the fire area. At the same time, a first path event is determined based on multiple emergency evacuation routes and the current location of hospitalized patients. A second path event is determined based on the current location of hospitalized patients and the hospital fire status map. The optimal emergency evacuation route is determined based on the first path event, the second path event, and multiple emergency evacuation routes. The first path event is the initial evacuation plan based on static path planning. The second path event is the dynamic adjustment plan based on the real-time fire status map.

6. The hospital fire emergency evacuation method based on the Internet of Things according to claim 1, characterized in that, The process involves identifying multiple emergency evacuation nodes based on the optimal emergency evacuation route, determining evacuation action instructions for hospitalized patients based on the node locations of these nodes, the corresponding fire smoke areas, and the movement status of hospitalized patients, and guiding the emergency evacuation behavior of hospitalized patients based on the Internet of Things, including: Multiple emergency evacuation routes are determined based on the detection of the optimal emergency evacuation routes. Based on the multiple emergency evacuation routes, the corresponding fire areas, and the indoor distribution map of the hospital, the corresponding emergency evacuation nodes are determined to collect multiple emergency evacuation nodes. Among multiple emergency evacuation nodes, the node location of each emergency evacuation node is determined based on the detection of each emergency evacuation node; multiple fire smoke features are determined based on the smoke detection of the node locations of each emergency evacuation node; and the corresponding fire smoke area is determined based on the feature location, corresponding feature shape and emergency evacuation route of the multiple fire smoke features.

7. The hospital fire emergency evacuation method based on the Internet of Things according to claim 6, characterized in that, The process of identifying multiple emergency evacuation nodes based on the identification of optimal emergency evacuation routes, determining evacuation action instructions for hospitalized patients based on the node locations of multiple emergency evacuation nodes, corresponding fire smoke areas, and the movement status of hospitalized patients, and guiding the emergency evacuation behavior of hospitalized patients based on the Internet of Things, further includes: The system collects the movement status of hospitalized patients. Based on the patients' movement status, their current location, and multiple fire smoke areas, a first-level evacuation action instruction is determined. Based on the node locations of multiple emergency evacuation nodes and their corresponding fire smoke areas, a second-level evacuation action instruction is determined. Based on the first and second-level evacuation action instructions, the evacuation action instructions for hospitalized patients are determined. Simultaneously, the evacuation action instructions are transmitted along the Internet of Things to the corresponding emergency response devices, guiding the emergency evacuation behavior of hospitalized patients. Specifically, the first-level evacuation action instruction is generated by calculating the shortest path from the hospitalized patient's current location to the safe area, assessing the smoke risk level along the path, and considering the impact of the patient's mobility on the path. The second-level evacuation action instruction is generated by assessing the smoke risk of each emergency evacuation node, calculating the passage safety between emergency evacuation nodes, and considering node capacity and congestion.

8. The hospital fire emergency evacuation method based on the Internet of Things according to claim 1, characterized in that, The determination of corresponding fire and fire-fighting events based on the emergency evacuation behavior of hospitalized patients, the location of firefighters, and the fire-fighting volume in each fire area includes: The system collects the current location of hospitalized patients and monitors their emergency evacuation behavior in real time. Simultaneously, firefighters enter the hospital to conduct simultaneous firefighting operations in various fire areas and collect their locations. Based on the emergency evacuation behavior of hospitalized patients, the locations of firefighters, and the amount of firefighting done in each fire area, the system determines the corresponding fire and fire incident.

9. The hospital fire emergency evacuation method based on the Internet of Things according to claim 8, characterized in that, The process involves identifying multiple fire-fighting items based on the identification of fire-fighting events, and determining multi-level rescue events for hospitalized patients based on these multiple fire-fighting items, the current location of hospitalized patients, and fire change events in various fire zones. This includes: Based on the identification of fire and fire-fighting events, multiple fire and fire-fighting items are identified and their corresponding contents are marked. The first rescue event is determined based on the contents of multiple fire and fire-fighting items and the current location of the hospitalized patient. Real-time monitoring of each fire zone and collection of fire change events in each fire zone; determination of second rescue events based on the current location of hospitalized patients and fire change events in each fire zone; determination of multi-level rescue events for hospitalized patients based on the first rescue event, the second rescue event, and the movement status of hospitalized patients.

10. A hospital fire emergency evacuation system based on the Internet of Things, characterized in that, The IoT-based hospital fire emergency evacuation system is applied to the IoT-based hospital fire emergency evacuation method as described in any one of claims 1-9, and the IoT-based hospital fire emergency evacuation system includes: An abnormal object feature module is used to determine indoor images based on multiple cameras and the Internet of Things in the hospital, and to determine multiple abnormal object features based on the recognition of indoor images, wherein the abnormal objects include smoke, flames, and high-temperature equipment; The fire dynamic map module is used to determine the corresponding fire warning level based on the feature location and feature shape of multiple abnormal objects, determine the fire area based on the fire warning level of multiple abnormal object features and the corresponding smoke data, and determine the corresponding fire dynamic map. This includes: determining the corresponding smoke data based on smoke detection of the feature location of each abnormal object feature; simultaneously, determining a first fire information combination based on the relative position of multiple abnormal object features and the corresponding smoke data; determining a second fire information combination based on the fire warning level of multiple abnormal object features and the corresponding smoke data; determining the fire area based on the first fire information combination and the second fire information combination; and determining the corresponding fire dynamic map within the fire area based on monitoring of each fire area. The fire dynamic map includes the following information: changes in the boundary of the fire area, the direction of smoke diffusion, and changes in risk level. The emergency evacuation route module is used to determine multiple emergency evacuation routes based on the hospital's indoor layout map and the regional location of the fire area. Based on multiple emergency evacuation routes, the current location of hospitalized patients, and the hospital fire status map, the optimal emergency evacuation route is determined. In this module, a graph neural network is used to connect discrete fire areas into a continuous fire status map, where nodes represent each fire area and edges represent the fire propagation relationship between fire areas. The evacuation action instruction module is used to determine multiple emergency evacuation nodes based on the identification of the optimal emergency evacuation route, and to determine the evacuation action instructions for hospitalized patients based on the node locations of multiple emergency evacuation nodes, the corresponding fire smoke areas, and the movement status of hospitalized patients, and to guide the emergency evacuation behavior of hospitalized patients based on the Internet of Things; wherein, in the fire smoke area, the fire smoke area boundary is generated based on the smoke concentration contour lines. The multi-level rescue event module is used to determine corresponding fire-fighting events based on the emergency evacuation behavior of hospitalized patients, the location of firefighters, and the fire extinguishing capacity of each fire area. Multiple fire-fighting items are identified based on the fire-fighting event identification. Multi-level rescue events for hospitalized patients are determined based on these multiple fire-fighting items, the current location of hospitalized patients, and fire change events in each fire area. The fire-fighting event generation process involves: collecting all input parameters, analyzing the current state of each parameter, matching each parameter with judgment rules, and generating corresponding fire-fighting events based on the matching results. The input parameters include the emergency evacuation behavior of hospitalized patients, the location of firefighters, and the fire extinguishing capacity of each fire area. Each fire-fighting event includes an event ID, event type, event description, and handling suggestions.

Images