A data processing system based on a multi-modal fire prevention and control emergency evacuation large model

Abstract

The application relates to a data processing system based on a multimodal fire prevention and control emergency evacuation large model, which comprises multiple cameras, multiple direction indicators and a multimodal fire prevention and control emergency evacuation large model; wherein each branch intersection in a target site is provided with a direction indicator; each direction indicator is bound with a direction determination task; when a fire evacuation alarm of the target site is triggered, each direction determination task is executed immediately and is repeatedly executed at a preset time interval; the global optimal evacuation path determination task is decomposed into multiple relatively lightweight and distributed local direction determination tasks, so that the response delay of single decision is reduced; and the direction determination task is repeatedly executed at a preset time interval, so that the evacuation path formed by the target directions displayed by the direction indicators can timely reflect the optimal evacuation path conforming to the current actual situation of the target site, and the overall evacuation efficiency can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a data processing system based on a multimodal fire prevention and emergency evacuation large model. Background Technology

[0002] With the development of large-scale model technology, existing fire emergency evacuation systems introduce multimodal fire prevention and emergency evacuation large-scale models. By integrating multimodal data such as text and images, these models comprehensively evaluate each possible evacuation route to determine the globally optimal evacuation route. However, in large venues, the amount of multimodal data related to the venue is large, and due to the large number of safety evacuation exits, there are many possible evacuation routes. The multimodal fire prevention and emergency evacuation model needs to process massive amounts of data and has a large computational load when comprehensively judging the possible evacuation routes, resulting in a slow overall inference speed. Furthermore, in determining the globally optimal evacuation route, the multimodal fire prevention and emergency evacuation model needs to globally traverse all possible evacuation routes and independently execute a complete route planning process at each fork in the road, which can easily lead to response delays. Especially during evacuation, if an emergency occurs (such as a new fire point or a sudden increase in the number of people in a local area), global planning needs to be re-performed, that is, a comprehensive judgment of each possible evacuation route needs to be re-performed to determine the globally optimal evacuation route. This process is time-consuming, further leading to response delays. Therefore, determining the globally optimal evacuation route based on the above methods is prone to response delays, resulting in the obtained globally optimal evacuation route lagging behind the dynamic changes on site, failing to reflect the globally optimal evacuation route that conforms to the current situation in a timely manner, and affecting the overall evacuation efficiency. Summary of the Invention

[0003] To achieve the objectives of this invention, the technical solution adopted is as follows: a data processing system based on a multimodal fire prevention and emergency evacuation large-scale model. The data processing system includes: multiple cameras, multiple direction indicators, and a multimodal fire prevention and emergency evacuation large-scale model; each fork in the road in the target site is equipped with a direction indicator; each direction indicator is bound to a direction determination task; when a fire evacuation alarm in the target site is triggered, each direction determination task is immediately executed and repeated at preset time intervals; when the direction determination task is executed, the following steps are implemented: S1. Obtain the initial grid map corresponding to the direction indicator bound to the direction determination task; the initial grid map contains at least two evacuation routes starting from the location of the direction indicator and ending at the safety evacuation exit of the target site; each evacuation route consists of several continuous grids and each actual area corresponding to each grid is equipped with a camera.

[0004] S2. Mark at least one ignition point location in the initial grid map as a key grid to obtain a key grid map.

[0005] S3. Mark the current number of people in the actual area corresponding to each grid in each evacuation route in the key grid map to obtain the target grid map; the current number of people in the actual area corresponding to the grid is obtained based on the image currently captured by the camera corresponding to that grid.

[0006] S4. Input the target grid map into the multimodal fire prevention and emergency evacuation model so that the multimodal fire prevention and emergency evacuation model can output the target direction corresponding to the current direction determination task.

[0007] S5: Control the direction indicator to display the target direction.

[0008] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention decomposes the task of determining the global optimal evacuation route into multiple relatively lightweight, distributed local direction determination tasks. For each fork in the road, global path planning is no longer performed; instead, the optimal direction to be selected is determined based on local relevant data. Due to the small computational load of the direction determination task, the response delay of a single decision is significantly reduced. Furthermore, the direction determination task is repeatedly executed at preset time intervals, which can continuously output target directions that conform to the current actual situation of the target site. This allows the evacuation path, which is formed by the target directions displayed by each direction indicator, to be dynamically adjusted according to changes in the actual situation of the target site. It can promptly reflect the optimal evacuation path that conforms to the current actual situation of the target site, which is conducive to improving the overall evacuation efficiency. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Example: like Figure 1As shown, this invention provides a technical solution: a data processing system based on a multimodal fire prevention and emergency evacuation large-scale model. The data processing system includes: multiple cameras, multiple directional indicators, and a multimodal fire prevention and emergency evacuation large-scale model; each fork in the road in the target site is equipped with a directional indicator; each directional indicator is bound to a directional determination task; when a fire evacuation alarm in the target site is triggered, each directional determination task is immediately executed and repeated at preset time intervals; when the directional determination task is executed, the following steps are implemented: S1. Obtain the initial grid map corresponding to the direction indicator bound to the direction determination task; the initial grid map contains at least two evacuation routes starting from the location of the direction indicator and ending at the safety evacuation exit of the target site; each evacuation route consists of several continuous grids and each actual area corresponding to each grid is equipped with a camera.

[0012] Specifically, the target site can be understood as the site where the data processing system based on the multimodal fire prevention and emergency evacuation big data model is used.

[0013] Specifically, the forked intersection is an intersection with at least three directions of traffic.

[0014] Specifically, the directional indicators at the forks in the road are equipped with directional lights that correspond one-to-one with the direction of travel at the fork in the road. At any given time, only one directional light is in the display state, while the other directional lights are in the off state. This is used to guide people to evacuate from inside the target site to the safe evacuation exit of the target site along the direction of travel corresponding to the directional light that is in the display state.

[0015] Furthermore, the direction of travel corresponding to the indicator light that is in the display state in the direction indicator is the direction indicated by the direction indicator.

[0016] Specifically, the safety evacuation exits of the target site are those that comply with fire safety regulations and allow personnel to safely evacuate to the outside of the target site in an emergency.

[0017] Optionally, at least two evacuation routes starting from the location of the directional indicator and ending at the safe evacuation exit of the target site are marked in the grid map corresponding to the target site to obtain the initial grid map corresponding to the directional indicator.

[0018] Specifically, the preset time interval is not less than 0.5 seconds and not more than 2 seconds, preferably 1 second.

[0019] Specifically, the camera corresponding to the grid is used to capture images of the actual area corresponding to the grid in real time; the actual area corresponding to the grid is the physical area; the physical area is the actual physical space area.

[0020] S2. Mark at least one ignition point location in the initial grid map as a key grid to obtain a key grid map.

[0021] Specifically, the data processing system also includes multiple sensors for fire detection.

[0022] Specifically, sensors used for fire detection include smoke sensors, temperature sensors, and humidity sensors.

[0023] In one specific embodiment, the sensors used for fire detection also include carbon monoxide detectors and flame detectors.

[0024] Specifically, sensors for fire detection are distributed throughout the target site; furthermore, within the actual area corresponding to each grid in the target site, sensors of the same type, quantity, and function for fire detection are deployed to ensure consistency in sensing capabilities.

[0025] Specifically, the location of the ignition point is determined based on the detection data from the plurality of sensors used for fire detection.

[0026] In one specific embodiment, the location of the ignition point can be determined by those skilled in the art using any existing technology, and will not be elaborated here.

[0027] Optionally, the data processing system inputs the detection data from the multiple sensors used for fire detection into the multimodal fire prevention and emergency evacuation model, so that the multimodal fire prevention and emergency evacuation model can determine whether a fire has occurred based on the received data; when it is determined that a fire has occurred, a fire evacuation alarm is triggered at the target site.

[0028] Specifically, when a grid in the initial grid map is marked as a critical grid, it means that there is an ignition point in the actual area corresponding to the grid marked as a critical grid.

[0029] S3. Mark the current number of people in the actual area corresponding to each grid in each evacuation route in the key grid map to obtain the target grid map; the current number of people in the actual area corresponding to the grid is obtained based on the image currently captured by the camera corresponding to that grid.

[0030] S4. Input the target grid map into the multimodal fire prevention and emergency evacuation model so that the multimodal fire prevention and emergency evacuation model can output the target direction corresponding to the current direction determination task.

[0031] Specifically, the target direction is the direction of travel at the location of the direction indicator for the optimal evacuation path determined by the multimodal fire prevention and emergency evacuation model based on the target grid map; the optimal evacuation path is the evacuation path with the lowest overall risk determined by the multimodal fire prevention and emergency evacuation model from all evacuation paths included in the target grid map.

[0032] Specifically, the Qwen2.0-plus large language model or the DeepSeek-R1 671B large language model was selected as the base model; collected fire emergency evacuation-related documents were integrated; the text in the fire emergency evacuation-related documents was cleaned, segmented, and key entity identified, and combined with multimodal information such as building floor plans and evacuation procedures to construct a basic training corpus; the base model was trained for the first round based on the basic training corpus; among them, the fire emergency evacuation-related documents include, but are not limited to: "Guidelines for the Preparation and Implementation of Fire Extinguishing and Emergency Evacuation Plans of Social Units", "Fire Safety Engineering Part 9: Guidelines for Personnel Evacuation Assessment", "Code for Fire Protection Design of Buildings" and other relevant fire protection regulations, emergency plan templates, evacuation drill cases, etc.

[0033] Furthermore, based on real or typical building layouts (such as shopping malls), a simulation system is used to simulate different fire simulation scenarios. The space of each fire simulation scenario is divided into a target grid map of uniform scale, and virtual directional indicators are deployed at key nodes (such as corridor intersections and stairwells). Each virtual directional indicator corresponds to a grid map. The grid map marks all passageways with the location of its corresponding virtual directional indicator as the starting point and the safe evacuation exit as the midpoint. Under each simulation condition, experts manually annotate the location of each directional indicator according to internationally accepted evacuation assessment criteria, marking the next passageway that should be indicated at the starting point of the globally lowest risk evacuation path. Based on this annotation result, a high-quality command-response pair dataset is constructed, including the grid map and the optimal direction. Based on the command-response pair dataset, the basic large model that has completed the first round of corpus training is subjected to a second round of supervised fine-tuning to obtain a multimodal fire prevention and emergency evacuation large model.

[0034] Through the above steps, a directional indicator is set at each fork in the road at the target site, and each directional indicator is bound to a direction determination task. The task of determining the global optimal evacuation route is decomposed into multiple relatively lightweight, distributed local direction determination tasks. Each direction determination task is executed independently. During the execution of the direction determination task, an initial grid map containing at least two evacuation routes starting from the location of the directional indicator and ending at the safe evacuation exit of the target site is obtained. The grid cell corresponding to the location of the fire point in the initial grid map is marked as the key grid cell to obtain the key grid map. The current number of people in the actual area corresponding to each grid cell in each evacuation route is marked in the key grid map. The system obtains a target grid map; the current number of people in the actual area corresponding to the grid is obtained based on the image currently captured by the camera corresponding to that grid; the target grid map is input into a multimodal fire prevention and emergency evacuation model, so that the multimodal fire prevention and emergency evacuation model outputs the target direction corresponding to the current direction determination task; thus, each direction determination task can quickly determine the best passage direction to be selected at its corresponding fork in the road based on the multimodal fire prevention and emergency evacuation model and its corresponding target grid map; the complex global optimization problem is transformed into a series of independent sub-tasks that focus on local areas, simplify data, and reduce computational load, which significantly improves the response speed while ensuring the rationality of decision-making.

[0035] Specifically, the multimodal fire prevention and emergency evacuation model is equipped with a corresponding task pool; the task pool is used to receive target grid maps and, according to the priority value of the target grid map, push the target grid map with the highest priority value in the current task pool to the multimodal fire prevention and emergency evacuation model each time it is pushed to the multimodal fire prevention and emergency evacuation model.

[0036] Furthermore, the priority value corresponding to the target raster map is determined through the following steps S10-S50: S10. For each evacuation path in the target grid map, obtain the network topology map corresponding to the evacuation path; wherein, the network topology map includes several nodes and directed edges connecting the nodes; the nodes represent the branch intersections in the evacuation path, and the attributes include the identifier of the corresponding branch intersection and the position coordinates of the corresponding branch intersection in the target grid map; the directed edges represent the road segments between the branch intersections corresponding to the nodes at both ends of the edge, and the direction corresponds to the actual travel direction of the road segment, and its attributes include image feature vectors and sensor data feature vectors. The image feature vectors are obtained based on the images currently collected by the cameras corresponding to the grids covered by the corresponding road segment and corresponding to the fire point location, and the sensor data feature vectors are obtained based on the detection data of the sensors used for fire detection in the actual area corresponding to the grids covered by the corresponding road segment and corresponding to the fire point location.

[0037] Specifically, the image feature vector is obtained based on the image currently captured by the camera corresponding to the grid cell covering the corresponding road segment and containing the location of the fire point. This includes: S11. From all the grids covered by the road section, identify the grids in the actual area where the fire point exists, and designate the grids as the fire point grids.

[0038] S12. Stitch together the images currently captured by the cameras corresponding to all fire point grids into a key image.

[0039] S13. Input the key images into the image feature extraction model to obtain the image feature vector.

[0040] Specifically, the image feature extraction model is a convolutional neural network pre-trained on ImageNet, such as ResNet and MobileNet, which will not be elaborated on here.

[0041] Through the above steps, key images are generated by fusing the images currently captured by the camera corresponding to the fire point grid. These key images are then input into the image feature extraction model to obtain image feature vectors, ensuring that the image feature vectors can accurately reflect the current fire situation.

[0042] Specifically, the feature vector of the sensor data is obtained from the detection data of the sensor used for fire detection within the actual area corresponding to the grid where the corresponding road segment is covered and the location of the ignition point is located. This includes: S101. For each fire point grid, obtain the detection data sequence of each sensor used for fire detection within the actual area corresponding to the fire point grid in the current time period.

[0043] Specifically, the current time period ends at the current time point, and the time interval between the start and end times of the current time period is consistent with the preset time interval.

[0044] Specifically, the detection data in the detection data sequence are arranged in chronological order from morning to night.

[0045] S102. Extract features from each detection data sequence to obtain the corresponding temporal feature vector.

[0046] In one specific embodiment, the detection data sequence is subjected to feature extraction using pre-designed time-domain and statistical features (including mean, variance, extreme values, slope of change, etc.) to generate a fixed-dimensional time-series feature vector.

[0047] S103. According to the preset arrangement order of the sensors used for fire detection, the time-series feature vectors of all the sensors used for fire detection in the actual area corresponding to the ignition point grid are spliced ​​together to obtain the comprehensive feature vector corresponding to the ignition point grid.

[0048] S104. The mean vector of the comprehensive feature vectors corresponding to all ignition point grids is used as the feature vector of the sensor data.

[0049] Through the above steps, using the ignition point grid as the anchor point, the time-series detection data of multiple sensors within its coverage area are dynamically aggregated, and a fixed-dimensional time-series feature vector is generated using statistical and temporal features. By splicing the data in the preset sensor arrangement order and taking the average, the structural consistency of multi-source sensor information is preserved, and robust fusion and noise reduction of heterogeneous sensor data are achieved. No complex model training is required, and it has high real-time performance, low resource consumption, and strong interpretability.

[0050] S20. Based on the feature vector of the network topology map corresponding to the evacuation path, determine the evacuation path with the highest safety value and the evacuation path with the longest duration of safety status from all evacuation paths in the target grid map.

[0051] Specifically, a graph neural network (such as the GCN model or the GAT model) is used to encode the network topology graph to obtain the feature vector of the network topology graph.

[0052] Specifically, step S20 includes the following sub-steps: S21. Input the feature vector of the network topology graph corresponding to the evacuation path into the first linear regression model to obtain the safety value corresponding to the evacuation path.

[0053] Specifically, the first linear regression model is obtained by supervised training of the linear regression model using the first training sample set. The first training sample set includes several first training samples, each of which includes a feature vector of the network topology graph corresponding to a historical evacuation path and a manually labeled safety value corresponding to that feature vector.

[0054] Specifically, historical evacuation routes are fire evacuation routes obtained based on simulation or experiments.

[0055] S22. If the direction of travel from the starting point of the evacuation path corresponding to the target grid map to the first fork in the road is consistent with the direction indicated by the direction indicator corresponding to the target grid map, and the safety value corresponding to the evacuation path is greater than the preset safety value threshold, then the evacuation path is determined as the critical path corresponding to the target grid map.

[0056] S23. Input the feature vector of the network topology map corresponding to the critical path in the target grid map into the second linear regression model to obtain the duration of the safe state corresponding to the critical path.

[0057] Specifically, the second linear regression model is obtained by supervising the linear regression model using a second training sample set. The second training sample set includes several second training samples, each of which includes a feature vector of the network topology graph corresponding to a historical evacuation path and the duration of the manually labeled safe state corresponding to that feature vector.

[0058] S24. Based on the duration of the safe state corresponding to each critical path in the target grid map, determine the evacuation path with the longest duration of the safe state.

[0059] Through the above steps, the evacuation path is modeled as a directed network topology graph with attributes using graph neural networks and linear regression models. The feature vector of the network topology graph is extracted by graph neural networks, and two linear regression models are used to predict the safety value and the duration of the safety state, respectively, so as to achieve a two-dimensional quantitative assessment of the path risk.

[0060] S30. Obtain the basic priority value based on the path length and basic priority value mapping function of the evacuation path with the highest safety value; where the longer the path length, the greater the basic priority value.

[0061] In one specific embodiment, the basic priority value mapping function is a monotonically increasing function with an upper bound, such as a truncated linear function, which is preset by those skilled in the art according to actual needs and will not be described in detail here.

[0062] S40. Based on the mapping function between the duration of the safe state of the evacuation path with the longest duration of safe state and the incremental priority, obtain the incremental priority value; wherein, the longer the duration of the safe state, the smaller the incremental priority value.

[0063] In one specific embodiment, the incremental priority mapping function is a monotonically decreasing function with an upper bound, such as an exponential decay function, which is preset by those skilled in the art according to actual needs and will not be described in detail here.

[0064] Specifically, when there is no critical path corresponding to the target raster map, the incremental priority value is determined to be the preset maximum incremental priority value.

[0065] S50. The sum of the basic priority value and the incremental priority value is used as the priority value corresponding to the target raster map.

[0066] Through the above steps, the priority value corresponding to the target grid map is obtained by adding the basic priority and the incremental priority: the longer the evacuation route with the highest safety value, the higher the basic priority, indicating a larger impact range; the longer the safety duration, the lower the incremental priority, indicating a smaller risk; therefore, the higher the priority value corresponding to the target grid map, the more urgent the situation in the actual area corresponding to the target grid map; the task pool prioritizes the scheduling of map data in high-risk areas accordingly, ensuring that emergency resources such as directional indicators are given priority to serve the most dangerous scenarios, greatly improving response efficiency.

[0067] Specifically, the task pool is configured to increase the priority value of each target grid map in the current task pool by a preset value at fixed intervals.

[0068] Furthermore, when two target grid maps corresponding to a direction indicator exist simultaneously in the task pool, the target grid map with the earlier reception time is deleted, and the priority value of the target grid map with the later reception time is increased by a preset value; this avoids the target grid map... Figure 1 It is not pushed into the multimodal fire prevention and emergency evacuation model.

[0069] Specifically, the multimodal fire prevention and emergency evacuation model is configured such that the time taken to process a single direction determination task multiplied by the number of direction indicators is no greater than the time corresponding to a preset time interval, so as to ensure that a large number of target grid maps do not accumulate in the task pool.

[0070] S5. Control the direction indicator to display the target direction.

[0071] Specifically, the direction displayed by the direction indicator is the direction indicated by the direction indicator.

[0072] Specifically, the direction indicator updates the direction it displays based on the received control commands.

[0073] Specifically, after the control direction indicator displays the target direction, the direction determination task is considered complete.

[0074] Through the above steps, during the execution of the direction determination task, after obtaining the target direction, the direction indicator is controlled to display the target direction. When the fire evacuation alarm of the target site is triggered, the direction determination task is immediately executed and repeated at preset time intervals to continuously output the target direction that conforms to the current actual situation of the target site. The target direction displayed by the direction indicator will be updated as its corresponding direction determination task is executed, so that the evacuation path formed by the target directions displayed by each direction indicator can be dynamically adjusted according to the changes in the actual situation of the target site. It can reflect the optimal evacuation path that conforms to the current actual situation of the target site in a timely manner, which is conducive to improving the overall evacuation efficiency. The optimal evacuation path can be understood as the evacuation path with the lowest overall risk.

[0075] Specifically, the process after step S4 and before step S5 includes: If the direction currently displayed by the direction indicator is different from the target direction, proceed to step S5; otherwise, determine that the direction determination task has been completed.

[0076] By following the steps above, control actions are only performed when the display direction needs to be updated. If the currently displayed direction is already the target direction, no control actions are performed, reducing unnecessary operations and avoiding resource waste.

[0077] Specifically, when the direction currently displayed by the direction indicator is different from the target direction, a control command is sent to the direction indicator to make the direction indicator display the target direction.

[0078] Specifically, once it is determined that all personnel in the target area have been evacuated, all directional determination tasks will be terminated.

[0079] This invention provides a data processing system based on a multimodal fire prevention and emergency evacuation large-scale model. The system includes multiple cameras, multiple directional indicators, and the multimodal fire prevention and emergency evacuation large-scale model. Each intersection in the target site is equipped with a directional indicator; each directional indicator is bound to a direction determination task. When a fire evacuation alarm is triggered in the target site, each direction determination task is immediately executed and repeated at preset time intervals. When a direction determination task is executed, an initial grid map corresponding to the directional indicator bound to that task is acquired. This initial grid map contains at least two lines starting from the location of the directional indicator and extending to the safety boundary of the target site. Evacuation routes terminate at evacuation exits; each evacuation route consists of several continuous grids, and each grid is equipped with a camera; at least one fire ignition point is marked as a key grid in the initial grid map to obtain a key grid map; the current number of people in the actual area corresponding to each grid in each evacuation route is marked in the key grid map to obtain a target grid map; wherein, the current number of people in the actual area corresponding to the grid is obtained based on the image currently captured by the camera corresponding to that grid; the target grid map is input into the multimodal fire prevention and emergency evacuation large model so that the multimodal fire prevention and emergency evacuation large model outputs the current direction to determine the target direction corresponding to the task and controls the direction indicator to display the target direction. As can be seen, this invention decomposes the task of determining the global optimal evacuation route into multiple relatively lightweight, distributed local direction determination tasks. For each fork in the road, global path planning is no longer performed; instead, the optimal direction to be selected is determined based on local relevant data. Since the computational load of the direction determination task is small, the response delay of a single decision is significantly reduced. Furthermore, the direction determination task is repeatedly executed at preset time intervals, which can continuously output target directions that conform to the current actual situation of the target site. This allows the evacuation route, which is formed by the target directions displayed by each direction indicator, to be dynamically adjusted according to changes in the actual situation of the target site. It can promptly reflect the optimal evacuation route that conforms to the current actual situation of the target site, which is conducive to improving the overall evacuation efficiency.

[0080] The embodiments disclosed herein are preferred embodiments, but are not limited thereto. Those skilled in the art can readily grasp the spirit of the present invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of the present invention, they are all within the protection scope of the present invention.

Claims

1. A data processing system based on a multimodal fire prevention and emergency evacuation large-scale model, characterized in that, The data processing system includes: multiple cameras, multiple directional indicators, and a multimodal fire prevention and emergency evacuation model; each fork in the road in the target site is equipped with a directional indicator; each directional indicator is bound to a directional determination task; when a fire evacuation alarm in the target site is triggered, each directional determination task is immediately executed and repeated at preset time intervals; when a directional determination task is executed, the following steps are performed: S1. Obtain the initial grid map corresponding to the direction indicator bound to the direction determination task; the initial grid map contains at least two evacuation routes starting from the location of the direction indicator and ending at the safety evacuation exit of the target site; each evacuation route consists of several continuous grids and each grid corresponds to an actual area equipped with a camera. S2. Mark the grid corresponding to at least one ignition point location in the initial grid map as a key grid to obtain a key grid map; S3. Mark the current number of people in the actual area corresponding to each grid in each evacuation route in the key grid map to obtain the target grid map; the current number of people in the actual area corresponding to the grid is obtained based on the image currently captured by the camera corresponding to that grid; S4. Input the target grid map into the multimodal fire prevention and emergency evacuation model so that the multimodal fire prevention and emergency evacuation model can output the target direction corresponding to the current direction determination task. S5. Control the direction indicator to display the target direction.

2. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 1, characterized in that, The multimodal fire prevention and emergency evacuation model is equipped with a corresponding task pool. The task pool is used to receive target grid maps and, according to the priority value of the target grid map, push the target grid map with the highest priority value in the current task pool to the multimodal fire prevention and emergency evacuation model each time it is pushed to the multimodal fire prevention and emergency evacuation model. The priority value corresponding to the target raster map is determined through the following steps: S10. For each evacuation path in the target grid map, obtain the network topology map corresponding to the evacuation path; wherein, the network topology map includes several nodes and directed edges connecting the nodes; the nodes represent the branch intersections in the evacuation path, and the attributes include the identifier of the corresponding branch intersection and the position coordinates of the corresponding branch intersection in the target grid map; the directed edges represent the road segments between the branch intersections corresponding to the nodes at both ends of the edge, and the direction corresponds to the actual travel direction of the road segment, and its attributes include image feature vectors and sensor data feature vectors. The image feature vectors are obtained based on the images currently collected by the cameras corresponding to the grids covered by the corresponding road segments and corresponding to the fire point locations, and the sensor data feature vectors are obtained based on the detection data of the sensors used for fire detection in the actual area corresponding to the grids covered by the corresponding road segments and corresponding to the fire point locations. S20. Based on the feature vector of the network topology map corresponding to the evacuation path, determine the evacuation path with the highest safety value and the evacuation path with the longest duration of safety status from all evacuation paths in the target grid map. S30. Obtain the basic priority value based on the mapping function between the path length of the evacuation path with the highest safety value and the basic priority value; wherein, the longer the path length, the larger the basic priority value. S40. Based on the mapping function between the duration of the safe state of the evacuation path with the longest duration of safe state and the incremental priority, obtain the incremental priority value; where, the longer the duration of the safe state, the smaller the incremental priority value. S50. The sum of the basic priority value and the incremental priority value is used as the priority value corresponding to the target raster map.

3. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 2, characterized in that, The task pool is configured to increase the priority value of each target grid map in the current task pool by a preset value at fixed intervals.

4. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 1, characterized in that, The steps following step S4 and before step S5 include: If the direction currently displayed by the direction indicator is different from the target direction, proceed to step S5; otherwise, determine that the direction determination task has been completed.

5. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 4, characterized in that, The data processing system also includes multiple sensors for fire detection.

6. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 5, characterized in that, Sensors used for fire detection include smoke sensors, temperature sensors, and humidity sensors.

7. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 5, characterized in that, The location of the ignition point is determined based on the detection data from the plurality of sensors used for fire detection.

8. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 1, characterized in that, When a grid in the initial grid map is marked as a critical grid, it means that there is an ignition point in the actual area corresponding to the grid marked as a critical grid; the actual area corresponding to the grid is the physical area.

9. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 1, characterized in that, The direction indicator updates its displayed direction based on the received control commands.

10. The data processing system based on a multimodal fire prevention and emergency evacuation large-scale model according to claim 1, characterized in that, Once the control direction indicator displays the target direction, the direction determination task is considered complete.

Images