Multi-sensor based detection cabin fire monitoring and fire handling method and system

Abstract

This invention provides a multi-sensor-based method and system for fire monitoring and firefighting in a detection chamber, relating to the field of monitoring and processing technology. The method includes acquiring real-time monitoring data from multiple sensors, employing a deep learning model to assess the fire situation, making fire response decisions based on the assessment results using distributed edge computing, controlling a mobile fire extinguishing device to autonomously move to the optimal extinguishing position and adaptively adjusting spray parameters, and monitoring the execution effect in real time and providing feedback to optimize the deep learning model. This invention improves the accuracy of fire assessment and the efficiency of firefighting, and enhances the system's reliability and adaptability.

Description

Technical Field

[0001] This invention relates to monitoring and processing technology, and more particularly to a method and system for fire monitoring and fire handling in a detection chamber based on multiple sensors. Background Technology

[0002] With the continuous development of industrial automation and intelligent manufacturing, the safety of inspection chambers, as an important component of industrial facilities, is receiving increasing attention. Inspection chambers typically house precision instruments and flammable and explosive materials; a fire in such a chamber can cause severe property damage and trigger a chain reaction of safety incidents. Traditional fire monitoring systems for inspection chambers rely primarily on single-type sensors for fire detection, such as temperature and smoke sensors. When abnormal data is detected, an alarm is triggered, and preset fire-fighting procedures are executed. However, with the development of artificial intelligence technology, intelligent fire monitoring systems based on multi-sensor fusion and deep learning are gradually becoming a research hotspot. These systems can improve the accuracy and timeliness of fire detection by comprehensively analyzing data from multiple sensors.

[0003] However, existing fire monitoring technologies for detection chambers still have some significant shortcomings and deficiencies. First, most existing technologies employ a centralized computing architecture for fire assessment and decision-making, concentrating computing resources on a single server. This can lead to response delays under high system loads, failing to meet the demands of real-time fire handling in complex environments. Second, traditional fixed fire extinguishing devices have fixed locations and preset extinguishing agent spray parameters, lacking the ability to adaptively adjust based on dynamic changes in the fire source, resulting in low fire extinguishing efficiency, especially in complex and ever-changing fire scenarios. Finally, existing systems generally lack real-time monitoring and feedback mechanisms for firefighting effectiveness, making it impossible to dynamically adjust strategies based on actual conditions during the firefighting process or continuously optimize the system's fire assessment model, thus hindering its ability to cope with complex and ever-changing fire developments. Summary of the Invention

[0004] The embodiments of the present invention provide a method and system for fire monitoring and fire handling in a detection chamber based on multiple sensors, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for fire monitoring and fire handling in a detection chamber based on multiple sensors, comprising:

[0006] The real-time monitoring data of multiple sensors installed in the detection chamber is obtained. Based on the real-time monitoring data of the multiple sensors, a deep learning model is used to evaluate the fire status of the detection chamber in real time and obtain the fire assessment result.

[0007] Based on the fire assessment results, a distributed edge computing approach is used to make fire response decisions. The fire assessment results are distributed to multiple edge computing nodes for parallel computing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules. The fire handling strategy is voted on through a consensus protocol, and the fire handling strategy with the highest number of votes is selected as the final fire response decision.

[0008] Based on the fire response decision results, the fire protection system in the detection chamber is controlled to perform corresponding fire-fighting operations. The fire protection system includes a mobile fire extinguishing device that can autonomously move to the optimal fire extinguishing position according to the fire source location information, and adaptively adjust the spray pressure and spray angle of the extinguishing agent according to the fire development level. The system monitors the execution effect of the fire-fighting operations in real time and feeds the monitoring results back to the deep learning model to optimize the fire assessment accuracy of the deep learning model.

[0009] Real-time monitoring data from multiple sensors installed inside the detection chamber is acquired. Based on this data, a deep learning model is used to assess the fire status of the detection chamber in real time, yielding the following fire assessment results:

[0010] The system acquires real-time monitoring data from multiple sensors within the detection chamber, including a temperature sensor, a smoke sensor, and a gas concentration sensor.

[0011] A deep learning model optimized by transfer learning is used to construct a multimodal data fusion network based on the correlation analysis of temporal features of data from different types of sensors, and to perform feature extraction and fire situation analysis on the real-time monitoring data.

[0012] Based on the results of the fire situation analysis, the output includes fire occurrence probability, fire type determination, and fire development level.

[0013] Based on the fire assessment results, a distributed edge computing approach is used to make fire response decisions. The fire assessment results are distributed to multiple edge computing nodes for parallel processing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules, including:

[0014] The fire assessment results are feature-extracted based on the fire type, fire intensity, and fire source location, and the fire response task is divided into multiple priority sub-tasks based on the extracted features.

[0015] The subtasks are matched with resources based on task priority and real-time load status of edge computing nodes. A dynamic load balancing algorithm is used to allocate the subtasks to the corresponding edge computing nodes, so that multiple edge computing nodes can synchronously process subtasks of different priorities based on a distributed parallel computing architecture.

[0016] After receiving the assigned subtasks, each edge computing node generates a corresponding fire response strategy based on the fire response rules corresponding to the subtasks.

[0017] A dynamic load balancing algorithm is used to distribute subtasks to appropriate edge computing nodes, enabling multiple edge computing nodes to synchronously process subtasks of different priorities based on a distributed parallel computing architecture, including:

[0018] Real-time monitoring of resource utilization of each edge computing node; construction of node health assessment model based on resource utilization; obtaining availability score for each node; matching availability score with priority of subtask to be assigned.

[0019] A priority queue for task allocation is established based on the availability score. A task scheduling algorithm that takes load balancing into account is adopted to prioritize the allocation of priority subtasks to nodes with availability scores higher than a preset score threshold. The load level of each node is kept consistent by dynamically adjusting the task allocation ratio.

[0020] Each edge computing node starts task processing according to the assigned task priority order, and maintains computing synchronization between nodes through a timestamp-based message passing mechanism, so that subtasks of different priorities can be processed in a coordinated and consistent parallel manner under the distributed architecture.

[0021] Based on the fire response decision results, the fire protection system within the detection chamber is controlled to perform corresponding fire-fighting operations. The fire protection system includes a portable fire extinguishing device, which can autonomously move to the optimal fire-fighting position based on the fire source location information.

[0022] Based on the fire response decision results, the fire source location information and fire development parameters are analyzed, and the optimal fire extinguishing location is determined based on the fire source location information and the fire development parameters.

[0023] Based on the optimal fire extinguishing location and the real-time obstacle distribution inside the detection chamber, a movement trajectory that avoids obstacles and has the shortest path is planned for the mobile fire extinguishing device, and a motion control command sequence containing position, speed and direction is generated.

[0024] The mobile fire extinguishing device is controlled to execute the motion control command sequence, continuously update the surrounding environment information and optimize the movement trajectory in real time during autonomous movement, until it reaches the optimal fire extinguishing position and then carries out fire extinguishing operations.

[0025] The system adaptively adjusts the spray pressure and angle of the extinguishing agent according to the fire development level, monitors the effectiveness of firefighting operations in real time, and feeds the monitoring results back to the deep learning model, including:

[0026] The current fire intensity index is calculated based on the fire source temperature, flame height, and spread rate. The initial injection pressure and injection angle of the extinguishing agent are then determined based on the fire intensity index.

[0027] The initial injection pressure and the injection angle are used as control parameters to perform fire extinguishing operations. At the same time, the rate of change of fire source temperature and the coverage area of ​​fire extinguishing agent are continuously collected through a sensor network. The fire extinguishing effect evaluation index is calculated based on the rate of change of fire source temperature and the coverage area of ​​fire extinguishing agent.

[0028] The fire extinguishing effect evaluation index is input into the deep learning model. The initial spray pressure and the spray angle are fine-tuned in real time according to the fire extinguishing effect evaluation index. The fire extinguishing effect is improved through parameter iteration optimization. The optimized parameter combination is then stored in the deep learning model.

[0029] A second aspect of the present invention provides a multi-sensor-based fire monitoring and fire suppression system for a detection chamber, comprising:

[0030] The first unit is used to acquire real-time monitoring data from multiple sensors installed inside the detection chamber, and to use a deep learning model to evaluate the fire status of the detection chamber in real time based on the real-time monitoring data from the multiple sensors, thereby obtaining the fire assessment result.

[0031] The second unit is used to make fire response decisions based on the fire assessment results using a distributed edge computing approach. The fire assessment results are distributed to multiple edge computing nodes for parallel computing processing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules, and votes on the fire handling strategies through a consensus protocol, selecting the fire handling strategy with the highest number of votes as the final fire response decision result.

[0032] The third unit is used to control the fire protection system in the detection chamber to perform corresponding fire protection operations based on the fire response decision results. The fire protection system includes a mobile fire extinguishing device that can autonomously move to the optimal fire extinguishing position based on the fire source location information, and adaptively adjust the spray pressure and spray angle of the extinguishing agent according to the fire development level. It monitors the execution effect of the fire protection operations in real time and feeds the monitoring results back to the deep learning model to optimize the fire assessment accuracy of the deep learning model.

[0033] A third aspect of the present invention provides an electronic device, comprising:

[0034] processor;

[0035] Memory used to store processor-executable instructions;

[0036] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0037] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0038] The beneficial effects of this application are as follows:

[0039] By using deep learning models to perform real-time assessment of fire conditions using multi-sensor data, the accuracy and timeliness of fire detection are improved, avoiding false alarms and missed alarms caused by single sensors, and enabling the system to fully grasp the development of the fire in the detection chamber.

[0040] By adopting a distributed edge computing approach for fire response decision-making, efficient utilization of computing resources is achieved. Through multi-node parallel processing and a consensus voting mechanism, the scientific nature and reliability of fire handling strategies are ensured, significantly improving system response speed and decision quality.

[0041] The equipped mobile fire extinguishing device can autonomously move to the optimal fire extinguishing position according to the location of the fire source, and adaptively adjust the fire extinguishing parameters according to the fire development level, realizing intelligent and precise fire fighting. At the same time, through a real-time monitoring and feedback mechanism, it continuously optimizes the deep learning model, forming a self-improving closed-loop system, which greatly improves the efficiency and success rate of fire fighting. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the fire monitoring and fire handling method for a detection chamber based on multiple sensors, according to an embodiment of the present invention.

[0043] Figure 2 This is a flowchart illustrating the task allocation process for edge computing nodes in an embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0045] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0046] Figure 1 This is a flowchart illustrating the fire monitoring and fire handling method for a detection chamber based on multiple sensors according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0047] The real-time monitoring data of multiple sensors installed in the detection chamber is obtained. Based on the real-time monitoring data of the multiple sensors, a deep learning model is used to evaluate the fire status of the detection chamber in real time and obtain the fire assessment result.

[0048] Based on the fire assessment results, a distributed edge computing approach is used to make fire response decisions. The fire assessment results are distributed to multiple edge computing nodes for parallel computing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules. The fire handling strategy is voted on through a consensus protocol, and the fire handling strategy with the highest number of votes is selected as the final fire response decision.

[0049] Based on the fire response decision results, the fire protection system in the detection chamber is controlled to perform corresponding fire-fighting operations. The fire protection system includes a mobile fire extinguishing device that can autonomously move to the optimal fire extinguishing position according to the fire source location information, and adaptively adjust the spray pressure and spray angle of the extinguishing agent according to the fire development level. The system monitors the execution effect of the fire-fighting operations in real time and feeds the monitoring results back to the deep learning model to optimize the fire assessment accuracy of the deep learning model.

[0050] In one optional implementation, real-time monitoring data from multiple sensors installed inside the detection chamber is acquired. Based on the real-time monitoring data from these sensors, a deep learning model is used to perform a real-time assessment of the fire status within the detection chamber, yielding a fire assessment result including:

[0051] The system acquires real-time monitoring data from multiple sensors within the detection chamber, including a temperature sensor, a smoke sensor, and a gas concentration sensor.

[0052] A deep learning model optimized by transfer learning is used to construct a multimodal data fusion network based on the correlation analysis of temporal features of data from different types of sensors, and to perform feature extraction and fire situation analysis on the real-time monitoring data.

[0053] Based on the results of the fire situation analysis, the output includes fire occurrence probability, fire type determination, and fire development level.

[0054] Real-time monitoring data from multiple sensors within the detection chamber are acquired. These sensors include a temperature sensor, a smoke sensor, and a gas concentration sensor. The temperature sensor collects temperature change data at different locations within the chamber, with a sampling frequency of 1 Hz, a measurement range of -20°C to 150°C, and an accuracy of ±0.5°C. The smoke sensor monitors the concentration of smoke particles within the chamber, with a sampling frequency of 2 Hz, a detection range of 0-500 μg / m³, and an accuracy of ±3%. The gas concentration sensor monitors the concentrations of carbon monoxide, carbon dioxide, and combustible gases, with a sampling frequency of 1 Hz. The detection range for carbon monoxide is 0-1000 ppm, for carbon dioxide it is 0-5000 ppm, and for combustible gases it is 0-100% LEL, with accuracies of ±5 ppm, ±30 ppm, and ±3% LEL, respectively.

[0055] These sensors are connected to the central processing unit via wired or wireless networks. Real-time data is transmitted in JSON format, containing sensor ID, timestamp, numerical value, and unit information. The system uses a data preprocessing module to clean and standardize the raw data, including outlier handling, missing value imputation, and data normalization. Outlier handling employs a sliding window mean value filtering method with a window size of 5 sampling points; missing values ​​are imputed using forward imputation; and normalization uses the Min-Max method to map various sensor data to the [0,1] interval.

[0056] For temperature data, the system pays particular attention to the rate of temperature rise and temperature gradient characteristics, calculating the rate of temperature change and the temperature difference between different locations within a 10-second window. For smoke data, the system extracts the trend characteristics of smoke concentration changes, including the concentration growth rate and fluctuation frequency over a short period. For gas concentration data, the system analyzes the gas ratio relationships and concentration change patterns, especially the carbon monoxide to carbon dioxide ratio, which is of great significance for fire type determination.

[0057] A deep learning model optimized using transfer learning is employed for fire situation analysis. This model is based on a pre-trained ResNet50 architecture and adapts to the fire detection task through transfer learning. Specifically, the system pre-trains the base network on a large fire dataset, then freezes the weights of the first few layers, fine-tuning only the subsequent layers to adapt to the specific detection chamber environment. The model input is a temporal feature matrix composed of multi-sensor data, with a time window length of 30 seconds.

[0058] To achieve multimodal data fusion, this embodiment constructs a dual-stream network structure. One stream processes temperature and smoke data, employing a one-dimensional convolutional neural network to extract temporal features. The convolutional kernel size is 3, the stride is 1, and a total of 4 convolutional layers are configured, each followed by batch normalization and a ReLU activation function. The other stream processes gas concentration data, using a Long Short-Term Memory (LSTM) network structure with 128 hidden layers and 2 layers. The features from the two streams are fused through an attention mechanism to generate a comprehensive feature representation. The attention weights are dynamically adjusted based on the confidence level of different sensor data, assigning higher weights to temperature sensor data when it is reliable and increasing the weights of gas sensor data when gas concentration fluctuates significantly.

[0059] The model training in this embodiment uses the Adam optimizer with an initial learning rate of 0.001, and a cosine annealing strategy is used to dynamically adjust the learning rate. The training batch size is 64, and the total number of training epochs is 100. The training data includes 8000 samples collected in a simulated environment, covering different types of fire scenarios (electrical fires, oil fires, solid combustible material fires, etc.) and non-fire scenarios. The validation set accounts for 20% of the model and is used to monitor the training process and prevent overfitting. To enhance the robustness of the model, data augmentation techniques are also employed, including adding Gaussian noise (σ=0.02) and random time offsets (±2 seconds).

[0060] Based on the fire situation analysis results, the system outputs three parts of fire assessment results. The probability of fire occurrence is calculated using the Softmax function, with an output value between 0 and 1. A primary alarm is triggered when the probability exceeds 0.75, and a high-level alarm is triggered when the probability exceeds 0.9. Fire type determination categorizes fires into four types: electrical fires, oil fires, solid combustible material fires, and non-fires, based on the probability distribution output by the model. The fire development level is divided into four levels—latent period, initial combustion period, development period, and peak period—based on sensor data trends and historical patterns, used to predict the fire's development.

[0061] The temperature sensor inside the detection chamber detected a rapid increase in temperature from 25°C to 48°C within 10 seconds, a rise in smoke concentration from 5 μg / m³ to 120 μg / m³, a rise in carbon monoxide concentration from 3 ppm to 85 ppm, and a rise in carbon dioxide concentration from 450 ppm to 1200 ppm. After inputting this data into a deep learning model, the system calculated a fire probability of 0.94, classifying it as an electrical fire and indicating an initial combustion stage. This assessment result was calculated and output within 1.5 seconds, providing accurate and timely information for subsequent firefighting decisions.

[0062] Experimental verification shows that the method achieves an accuracy of 96.3% in early fire detection, with a false alarm rate of less than 3% and an average response time of less than 2 seconds, issuing an early warning more than 30 seconds earlier than the traditional threshold judgment method.

[0063] In one optional implementation, fire response decisions are made using distributed edge computing based on the fire assessment results. The fire assessment results are distributed to multiple edge computing nodes for parallel processing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules, including:

[0064] The fire assessment results are feature-extracted based on the fire type, fire intensity, and fire source location, and the fire response task is divided into multiple priority sub-tasks based on the extracted features.

[0065] The subtasks are matched with resources based on task priority and real-time load status of edge computing nodes. A dynamic load balancing algorithm is used to allocate the subtasks to the corresponding edge computing nodes, so that multiple edge computing nodes can synchronously process subtasks of different priorities based on a distributed parallel computing architecture.

[0066] After receiving the assigned subtasks, each edge computing node generates a corresponding fire response strategy based on the fire response rules corresponding to the subtasks.

[0067] like Figure 2 As shown, the method includes:

[0068] The raw fire data is processed through a feature extraction module. For example, for a fire occurring in a chemical industrial park, the feature extraction results include: fire type "chemical fire", fire severity level "severe", fire source location "warehouse area A", flammable and explosive materials within a 50-meter radius, fire spread rate of 5 meters per minute, northeast wind direction, and wind speed of level 4. The feature extraction uses a deep learning model, which has been trained on 10,000 fire data sets with an accuracy rate of 95.7%.

[0069] Based on the extracted features, the system divides fire response tasks into priority subtasks. In the aforementioned chemical industrial park fire scenario, the system generates the following subtasks: Subtask 1 "Personnel Evacuation" (Priority 10), Subtask 2 "Isolation of Flammable and Explosive Materials" (Priority 9), Subtask 3 "Chemical Fire Extinguishing" (Priority 8), Subtask 4 "Monitoring Fire Spread" (Priority 7), Subtask 5 "Traffic Control" (Priority 6), etc. The priority setting ranges from 1 to 10, with higher values ​​indicating higher priority.

[0070] After task partitioning, the system needs to allocate subtasks to suitable edge computing nodes. This embodiment uses a dynamic load balancing algorithm for resource matching. This algorithm continuously monitors metrics such as CPU utilization, memory usage, network bandwidth usage, and processing capacity of each edge computing node. For example, in a system with five edge computing nodes, node 1 has a CPU utilization of 30% and a memory usage of 25%; node 2 has a CPU utilization of 65% and a memory usage of 70%; node 3 has a CPU utilization of 45% and a memory usage of 50%; node 4 has a CPU utilization of 20% and a memory usage of 30%; and node 5 has a CPU utilization of 55% and a memory usage of 40%.

[0071] The dynamic load balancing algorithm comprehensively considers task priority and node load status for matching. In the above fire scenario, the system assigns subtask 1 "Personnel evacuation" (priority 10) to node 4 with a lighter load; subtask 2 "Isolation of flammable and explosive materials" (priority 9) to node 1; subtask 3 "Chemical fire extinguishing" (priority 8) to node 5; subtask 4 "Monitoring the spread of fire" (priority 7) to node 3; and subtask 5 "Traffic control" (priority 6) to node 2 with a heavier load.

[0072] To ensure effective dynamic load balancing, the system reassesses the load status of each node every 5 seconds. When the CPU utilization of a node exceeds 85% or the memory usage exceeds 90%, the system triggers a task redistribution mechanism, transferring low-priority tasks on that node to nodes with lighter loads. For example, when the CPU utilization of node 5 rises to 87%, the system transfers its subtask 3 "chemical fire extinguishing" to node 1 to continue execution.

[0073] After receiving the assigned subtasks, each edge computing node generates a corresponding fire response strategy based on preset fire response rules. These preset rules are stored in a distributed database, containing a library of response plans for different fire types, fire intensities, and environmental conditions. For example, for chemical fires, the system has 500 built-in dedicated fire extinguishing plans for different chemicals; for building fires, the system has 200 built-in fire extinguishing strategies categorized by building structure, materials, and function.

[0074] Taking subtask 1 "Personnel Evacuation" as an example, node 4 generates a detailed evacuation route map based on information such as the location of the fire source, the speed of fire spread, wind direction, and building layout. It identifies four main evacuation exits, estimates the evacuation time at 15 minutes, and generates a timetable for evacuation in batches and by area. Simultaneously, the system also identifies two emergency exits that are unusable due to the fire and plans dedicated evacuation routes and rescue plans for people with mobility impairments.

[0075] For subtask 2 "Isolate flammable and explosive materials", the processing strategy generated by node 1 includes: setting up an isolation zone 50 meters northwest of the fire source, mobilizing 3 dedicated robots to move 12 flammable gas containers in area A to a safe area, and simultaneously activating the automatic sprinkler cooling system in the storage area to control the surrounding temperature below 25°C.

[0076] To ensure consistency and coordination in distributed computing, a real-time data sharing mechanism is established among the edge computing nodes. When a processing strategy generated by one node affects the decisions of other nodes, the relevant information is immediately broadcast to all nodes. For example, when node 1 determines the route for isolating flammable and explosive materials, this information is immediately notified to node 4, which is responsible for personnel evacuation, so that it can adjust the evacuation route accordingly to avoid intersecting with the route for transporting hazardous materials.

[0077] Through the aforementioned distributed edge computing approach, the system can complete the entire process from fire assessment to generating a comprehensive processing strategy within 3 seconds. Compared with the traditional centralized computing method, the response speed is improved by 78%, and it can still maintain basic functions even in the event of partial network failure, which greatly improves the efficiency and reliability of fire emergency response.

[0078] In one optional implementation, a dynamic load balancing algorithm is used to distribute subtasks to corresponding edge computing nodes, enabling multiple edge computing nodes to synchronously process subtasks of different priorities based on a distributed parallel computing architecture, including:

[0079] Real-time monitoring of resource utilization of each edge computing node; construction of node health assessment model based on resource utilization; obtaining availability score for each node; matching availability score with priority of subtask to be assigned.

[0080] A priority queue for task allocation is established based on the availability score. A task scheduling algorithm that takes load balancing into account is adopted to prioritize the allocation of priority subtasks to nodes with availability scores higher than a preset score threshold. The load level of each node is kept consistent by dynamically adjusting the task allocation ratio.

[0081] Each edge computing node starts task processing according to the assigned task priority order, and maintains computing synchronization between nodes through a timestamp-based message passing mechanism, so that subtasks of different priorities can be processed in a coordinated and consistent parallel manner under the distributed architecture.

[0082] The system monitors the resource utilization of each edge computing node in real time. Every 5 seconds, the resource monitoring module collects data such as CPU utilization, memory usage, network bandwidth utilization, and disk I / O load. For example, the monitoring results for a certain node at a specific time might be: CPU utilization 65%, memory usage 58%, network bandwidth utilization 43%, and disk I / O load 32%. The system inputs these resource indicators into the node health assessment model for calculation. The node health assessment model uses a weighted average method, assigning weights based on the importance of different resource types; for example, CPU accounts for 40% of the weight, memory for 30%, network for 20%, and disk I / O for 10%. By calculating the weighted average of resource utilization and performing a reverse mapping, the node availability score is obtained. The availability score ranges from 0 to 100, with a higher score indicating more available resources. Under the above resource utilization conditions, the node's availability score is calculated to be 54.

[0083] The assigned subtasks are prioritized based on factors such as business importance, time urgency, and resource requirements, and are categorized into urgent tasks (Priority A, 90-100 points), high-priority tasks (Priority B, 75-89 points), medium-priority tasks (Priority C, 50-74 points), and low-priority tasks (Priority D, 0-49 points). The system maintains a task priority queue and matches task priorities with node availability scores.

[0084] During the task allocation phase, a task allocation priority queue is constructed based on node availability scores. A preset threshold of 65 points is set for node availability scores. Nodes with availability scores higher than 65 points are prioritized for allocation of Level A and Level B tasks. For example, edge node 1 has an availability score of 88 points, edge node 2 has a score of 72 points, edge node 3 has a score of 63 points, and node 4 has a score of 52 points. In this case, the system prioritizes allocating Level A tasks to node 1, Level B tasks to nodes 1 and 2, Level C tasks to nodes 2 and 3, and Level D tasks to nodes 3 and 4.

[0085] A dynamic task scheduling algorithm is used for load balancing. This algorithm sets a target load deviation threshold of 10%, meaning the load difference between any two nodes should not exceed 10%. When the system detects that the load difference between nodes exceeds the threshold, it triggers a task reallocation mechanism. For example, when the CPU utilization of node 1 reaches 85% while that of node 2 is only 45%, the system will transfer some of the B-level tasks originally allocated to node 1 to node 2 to achieve load balancing. The task allocation ratio is adjusted through a dynamic coefficient, which is determined based on the ratio of each node's current load to the system's average load. If a node's load is 20% higher than the system's average load, the task acceptance coefficient for that node will decrease by 25%; conversely, if the load is 20% lower than the system's average load, the task acceptance coefficient will increase by 25%.

[0086] After receiving the assigned tasks, each edge computing node initiates processing according to task priority. Each node maintains an internal priority queue to ensure that high-priority tasks are executed first. For level A tasks, the system allocates exclusive computing resources; for level B tasks, the system allocates 75% of available computing resources; for level C tasks, the system allocates 50% of available computing resources; and for level D tasks, the system allocates 25% of available computing resources.

[0087] To ensure computational synchronization among nodes in a distributed environment, the system implements a timestamp-based message passing mechanism. Each computational task is assigned a globally unique timestamp during execution, with timestamp precision down to the millisecond level. After a node completes its subtask, it generates a message containing a timestamp, task ID, execution result, and status information, which is then broadcast to other relevant nodes via a message queue. Receiving nodes process messages according to their timestamp order to ensure data consistency. The system sets a message synchronization timeout threshold of 200 milliseconds; if the expected message is not received within this threshold, a message retransmission mechanism is triggered.

[0088] The system also implements an exception handling mechanism. When a node's resource utilization exceeds 90% for more than 30 seconds, the system marks the node as overloaded, suspends the allocation of new tasks to it, and transfers some existing tasks to other nodes. If a node completely fails, the system triggers a task migration strategy, reassigning the node's unfinished tasks to other available nodes, and determining the migration order based on task priority.

[0089] Through the above implementation methods, the present invention achieves dynamic load balancing and task collaborative processing in an edge computing environment, which improves the overall resource utilization of the system by 32%, shortens the average response time of high-priority tasks by 45%, and increases the task processing throughput by 28%, significantly improving the performance and reliability of the edge computing system.

[0090] In one optional implementation, the fire suppression system within the detection chamber is controlled to perform corresponding fire suppression operations based on the fire response decision result. The fire suppression system includes a portable fire extinguishing device, which is capable of autonomously moving to the optimal fire suppression position based on the fire source location information.

[0091] Based on the fire response decision results, the fire source location information and fire development parameters are analyzed, and the optimal fire extinguishing location is determined based on the fire source location information and the fire development parameters.

[0092] Based on the optimal fire extinguishing location and the real-time obstacle distribution inside the detection chamber, a movement trajectory that avoids obstacles and has the shortest path is planned for the mobile fire extinguishing device, and a motion control command sequence containing position, speed and direction is generated.

[0093] The mobile fire extinguishing device is controlled to execute the motion control command sequence, continuously update the surrounding environment information and optimize the movement trajectory in real time during autonomous movement, until it reaches the optimal fire extinguishing position and then carries out fire extinguishing operations.

[0094] Based on the fire response decision, the fire suppression system within the testing chamber is controlled to perform corresponding firefighting operations. The testing chamber is equipped with a fire suppression system, which includes portable fire extinguishing devices. These portable devices can autonomously move to the optimal fire extinguishing position based on the fire source location information, thereby achieving more efficient and precise firefighting operations.

[0095] The mobile fire suppression system utilizes a wheeled mobile platform, on which fire extinguishers, cameras, lidar, processors, communication modules, and a drive system are mounted. Fire extinguishers can be carbon dioxide, dry powder, or water-based, configured according to the type of fire within the detection chamber. The camera employs a high-definition wide-angle lens, providing real-time images with a 160-degree field of view and a resolution of 1920×1080 pixels. The lidar uses a 16-line lidar with a 360-degree horizontal and 30-degree vertical scanning range, a ranging accuracy of ±2 cm, and a scanning frequency of 10 Hz. The processor features a quad-core 1.8 GHz CPU and 4 GB of RAM, enabling real-time processing of environmental perception and path planning tasks. The communication module supports Wi-Fi 802.11ac and Bluetooth 5.0 protocols, ensuring stable communication with the fire suppression system's main controller.

[0096] The system analyzes the fire source location information and fire development parameters based on the fire response decision results. These results typically include fire source coordinates, fire type, fire spread rate, and direction. For example, the decision result might indicate that the fire source is located at coordinates (x=3.5m, y=2.1m, z=0.4m) inside the detection chamber, the fire type is an electrical fire, the fire spread rate is 0.3 m² / min, and it is primarily spreading northeast. The system extracts and stores these parameters in a temporary cache to prepare for subsequent calculations of the optimal fire extinguishing location.

[0097] The optimal fire extinguishing location is determined based on the fire source location information and fire development parameters. The system comprehensively considers multiple indicators such as fire extinguishing efficiency, safety factors, and equipment limitations. For the electrical fire case mentioned above, considering the characteristics of electrical fires, the system determines that a carbon dioxide fire extinguisher should be used, and a safe distance must be maintained from the fire source to avoid the risk of electric shock. The system calculates the optimal spray distance to be 1.5 meters, and considering that the fire is mainly spreading in a northeast direction, the approach to the fire source should be from the southwest. The final optimal fire extinguishing location coordinates are determined to be (x=2.4m, y=1.0m, z=0.4m), a location that ensures the best fire extinguishing effect while guaranteeing equipment safety.

[0098] After determining the optimal fire extinguishing location, the system needs to combine this with the real-time obstacle distribution within the detection chamber to plan the movement trajectory of the mobile fire extinguishing device. The system uses lidar and cameras to acquire obstacle information within the detection chamber in real time, constructing an environmental map. In this map, the system marks the locations of fixed equipment, temporary items, and personnel within the detection chamber. For example, if a 0.5m × 0.8m equipment box is detected at (x=1.8m, y=1.5m), and a row of seats is detected between (x=3.0m, y=0.8m) and (x=4.2m, y=0.8m), the system uses the A* algorithm to calculate the path from the current location (x=0.5m, y=0.5m) to the optimal fire extinguishing location (x=2.4m, y=1.0m), planning a movement trajectory with a total length of 2.8 meters and including 3 turning points. The system generates a detailed sequence of motion control commands, including the position coordinates, speed, and direction adjustment angle for each path segment. For example: the first straight-line movement is from (0.5, 0.5) to (1.5, 0.5) at a speed of 0.5 m / s; the first turn is a 45-degree clockwise rotation; the second straight-line movement is from (1.5, 0.5) to (2.0, 0.75) at a speed of 0.3 m / s; and so on.

[0099] When controlling the mobile fire extinguishing device to execute motion control command sequences, the system adopts a closed-loop control method, comparing the deviation between the expected trajectory and the actual position in real time and making dynamic adjustments. The drive system of the mobile fire extinguishing device includes four independently controlled motors, with a maximum speed of 1.2 m / s, a minimum speed of 0.1 m / s, and a steering accuracy of ±2 degrees. During autonomous movement, the lidar scans the surrounding environment at a frequency of 10 Hz, detecting dynamic obstacles within its range. If a new obstacle is detected, or if personnel enter the movement path, the system immediately updates the environmental map and replans an avoidance path. For example, when a new obstacle is detected at the original path (x=2.0m, y=0.8m), the system calculates a new detour path, first shifting 0.3 meters to the right to bypass the obstacle and then returning to the original path.

[0100] Once the portable fire extinguishing device reaches the optimal extinguishing position, the system performs fine-tuning to ensure the nozzles are facing the fire source and adjusts to the optimal spray angle and distance. In the above case, the system adjusts the extinguishing device to face northeast and sets the spray angle to a 15-degree elevation to ensure carbon dioxide effectively covers the fire area. Then, the system activates the extinguisher and begins extinguishing operations. During the extinguishing process, the system continuously monitors the fire source temperature using a thermal imaging camera. When the temperature drops to a safe threshold (e.g., below 60°C), the system determines that the fire has been successfully extinguished, stops spraying, and reports the extinguishing result to the main fire control system controller.

[0101] The above describes in detail the specific implementation method of the mobile fire extinguishing device autonomously moving to the optimal fire extinguishing position and performing fire extinguishing operations. Through precise location perception, intelligent path planning and real-time environmental adaptation, it achieves efficient and accurate fire handling operations.

[0102] In one optional implementation, the spray pressure and spray angle of the extinguishing agent are adaptively adjusted according to the fire development level, the execution effect of the fire fighting operation is monitored in real time, and the monitoring results are fed back to the deep learning model, including:

[0103] The current fire intensity index is calculated based on the fire source temperature, flame height, and spread rate. The initial injection pressure and injection angle of the extinguishing agent are then determined based on the fire intensity index.

[0104] The initial injection pressure and the injection angle are used as control parameters to perform fire extinguishing operations. At the same time, the rate of change of fire source temperature and the coverage area of ​​fire extinguishing agent are continuously collected through a sensor network. The fire extinguishing effect evaluation index is calculated based on the rate of change of fire source temperature and the coverage area of ​​fire extinguishing agent.

[0105] The fire extinguishing effect evaluation index is input into the deep learning model. The initial spray pressure and the spray angle are fine-tuned in real time according to the fire extinguishing effect evaluation index. The fire extinguishing effect is improved through parameter iteration optimization. The optimized parameter combination is then stored in the deep learning model.

[0106] By combining fire data acquisition, fire intensity assessment, fire extinguishing parameter adjustment, and effect feedback mechanisms, the system optimizes fire extinguishing efficiency. The system first deploys a multi-sensor array, including infrared temperature sensors, optical flame detectors, and flow sensors, to collect key data such as fire source temperature, flame height, and spread rate. Temperature sensors are positioned at different heights within a 0.5-3 meter range around the fire source, with a sampling frequency of 5 times per second to ensure real-time and accurate temperature data. The optical flame detector measures flame height using image processing technology, with an accuracy of ±5 centimeters. The flame spread rate is calculated based on the time difference between the detection of flames by adjacent sensors, typically ranging from 0.1 to 2 meters per second.

[0107] Based on the acquired data on fire source temperature, flame height, and spread rate, the system calculates the current fire intensity index. The fire intensity index employs a comprehensive scoring mechanism, assigning weight coefficients to temperature, flame height, and spread rate. Temperature has a weight of 0.5, flame height a weight of 0.3, and spread rate a weight of 0.2. Specifically, when the detected fire source temperature is 400℃, the flame height is 1.2 meters, and the spread rate is 0.5 meters per second, the system calculates a fire intensity index of 7.4 (out of 10). Based on this index value, the system determines the initial spray pressure to be 2.8 MPa and the spray angle to be 35 degrees from a preset parameter mapping table. This parameter configuration is suitable for medium-intensity fires, ensuring effective fire suppression while reducing extinguishing agent waste.

[0108] After the spray system is activated, the control module transmits the initial spray pressure of 2.8 MPa and the spray angle of 35 degrees as control parameters to the actuator. The actuator includes a high-precision pressure regulating valve and a servo motor-driven nozzle angle adjustment mechanism, with a pressure regulation accuracy of ±0.05 MPa and an angle adjustment accuracy of ±1 degree. Simultaneously, the system continuously collects the fire source temperature change rate and the extinguishing agent coverage area through a sensor network. The temperature change rate sensor records temperature data every 200 milliseconds and calculates the temperature drop slope; the extinguishing agent coverage area is calculated in real time using an infrared thermal imager combined with image analysis algorithms, achieving an accuracy of ±0.1 square meters. In actual operation, when the temperature drop rate reaches -15℃ / second and the extinguishing agent coverage area is 8 square meters, the system assesses it as having a good fire extinguishing effect.

[0109] Based on the collected temperature change rate and extinguishing agent coverage area, the system calculates a fire extinguishing effectiveness evaluation index. This index comprehensively considers the temperature drop rate (weight 0.6) and extinguishing agent utilization efficiency (weight 0.4). Extinguishing agent utilization efficiency is calculated as the ratio of coverage area to consumption. For example, when the temperature drop rate is -15℃ / second, the coverage area is 8 square meters, and the extinguishing agent consumption is 20 liters / minute, the calculated fire extinguishing effectiveness evaluation index is 8.2 points (out of 10).

[0110] The fire extinguishing effectiveness evaluation index was then input into a deep learning model for analysis. This model employs a five-layer neural network structure, including an input layer (2 nodes: temperature change rate, coverage efficiency), three hidden layers (containing 64, 32, and 16 neurons respectively), and an output layer (2 nodes: pressure adjustment, angle adjustment). The model was trained using batch gradient descent with a learning rate of 0.001 and the ReLU activation function. When the input fire extinguishing effectiveness evaluation index was 8.2, the model suggested the following parameter adjustments: pressure adjustment -0.3 MPa, angle adjustment +5 degrees. Based on this, the system adjusted the spray pressure to 2.5 MPa and the spray angle to 40 degrees to improve fire extinguishing efficiency.

[0111] After parameter adjustments, the system continued to monitor the fire extinguishing effect. Ten seconds after the parameter adjustments, the temperature drop rate increased to -22℃ / second, the coverage area increased to 9.5 square meters, the extinguishing agent consumption remained unchanged, and the fire extinguishing effect evaluation index rose to 9.1 points. This indicates that the parameter adjustments effectively improved the fire extinguishing efficiency. The system stored the successful parameter combination (fire intensity index 7.4, spray pressure 2.5MPa, spray angle 40 degrees) into the experience database of the deep learning model for rapid parameter matching in similar situations in the future.

[0112] To verify the system's adaptability, tests were conducted under different fire scenarios. In a small oil fire (fire intensity index 5.2), the system automatically adjusted parameters to 1.8 MPa pressure and 25 degrees angle, reducing the extinguishing time by 32% compared to the fixed parameter scheme. In a large solid material fire (fire intensity index 8.7), the system adjusted to 3.4 MPa pressure and 45 degrees angle, reducing the amount of extinguishing agent used by 25%. This demonstrates that the system can intelligently adjust its extinguishing strategy according to different fire characteristics.

[0113] It also features self-learning optimization capabilities. After each firefighting operation, the actual effect data is fed back to the deep learning model, and the model weights are updated through error backpropagation. After accumulating data from 50 real-world tests, the model's prediction accuracy improved from the initial 82% to 94%, the average time for parameter adjustment was reduced from 1.2 seconds to 0.4 seconds, and the firefighting efficiency was improved by an average of 38% compared to traditional fixed-parameter solutions.

[0114] This implementation method achieves high efficiency in the use of extinguishing agents and optimal fire extinguishing effect through a closed-loop system that adaptively adjusts the fire development level, monitors in real time, and provides a feasible technical path for intelligent fire protection technology.

[0115] A second aspect of the present invention provides a multi-sensor-based fire monitoring and fire suppression system for a detection chamber, comprising:

[0116] The first unit is used to acquire real-time monitoring data from multiple sensors installed inside the detection chamber, and to use a deep learning model to evaluate the fire status of the detection chamber in real time based on the real-time monitoring data from the multiple sensors, thereby obtaining the fire assessment result.

[0117] The second unit is used to make fire response decisions based on the fire assessment results using a distributed edge computing approach. The fire assessment results are distributed to multiple edge computing nodes for parallel computing processing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules, and votes on the fire handling strategies through a consensus protocol, selecting the fire handling strategy with the highest number of votes as the final fire response decision result.

[0118] The third unit is used to control the fire protection system in the detection chamber to perform corresponding fire protection operations based on the fire response decision results. The fire protection system includes a mobile fire extinguishing device that can autonomously move to the optimal fire extinguishing position based on the fire source location information, and adaptively adjust the spray pressure and spray angle of the extinguishing agent according to the fire development level. It monitors the execution effect of the fire protection operations in real time and feeds the monitoring results back to the deep learning model to optimize the fire assessment accuracy of the deep learning model.

[0119] A third aspect of the present invention provides an electronic device, comprising:

[0120] processor;

[0121] Memory used to store processor-executable instructions;

[0122] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0123] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0124] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for fire monitoring and fire handling in a detection chamber based on multiple sensors, characterized in that, include: Real-time monitoring data from multiple sensors installed inside the detection chamber is acquired. Based on this data, a deep learning model is used to assess the fire status of the detection chamber in real time, yielding a fire assessment result, including: The system acquires real-time monitoring data from multiple sensors within the detection chamber, including a temperature sensor, a smoke sensor, and a gas concentration sensor. A deep learning model optimized by transfer learning is used to construct a multimodal data fusion network based on the correlation analysis of temporal features of data from different types of sensors, and to perform feature extraction and fire situation analysis on the real-time monitoring data. Based on the results of the fire situation analysis, the output includes fire occurrence probability, fire type determination, and fire development level; To achieve multimodal data fusion, a dual-stream network structure was constructed. One stream processes temperature and smoke data, using a one-dimensional convolutional neural network to extract temporal features, with batch normalization and ReLU activation functions applied to each layer. The other stream processes gas concentration data, employing a long short-term memory network structure. The features from the two streams are fused through an attention mechanism to generate a comprehensive feature representation. The attention weights are dynamically adjusted based on the confidence level of different sensor data. Temperature sensor data is assigned a weight higher than a preset threshold when it is reliable, and the weight of gas sensor data is increased when gas concentration fluctuates significantly. Based on the fire assessment results, a distributed edge computing approach is used to make fire response decisions. The fire assessment results are distributed to multiple edge computing nodes for parallel computing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules. The fire handling strategy is voted on through a consensus protocol, and the fire handling strategy with the highest number of votes is selected as the final fire response decision. Based on the fire response decision results, the fire protection system in the detection chamber is controlled to perform corresponding fire-fighting operations. The fire protection system includes a mobile fire extinguishing device that can autonomously move to the optimal fire extinguishing position according to the fire source location information, and adaptively adjust the spray pressure and spray angle of the extinguishing agent according to the fire development level. The system monitors the execution effect of the fire-fighting operations in real time and feeds the monitoring results back to the deep learning model to optimize the fire assessment accuracy of the deep learning model.

2. The method according to claim 1, characterized in that, Based on the fire assessment results, a distributed edge computing approach is used to make fire response decisions. The fire assessment results are distributed to multiple edge computing nodes for parallel processing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules, including: The fire assessment results are feature-extracted based on the fire type, fire intensity, and fire source location, and the fire response task is divided into multiple priority sub-tasks based on the extracted features. The subtasks are matched with resources based on task priority and real-time load status of edge computing nodes. A dynamic load balancing algorithm is used to allocate the subtasks to the corresponding edge computing nodes, so that multiple edge computing nodes can synchronously process subtasks of different priorities based on a distributed parallel computing architecture. After receiving the assigned subtasks, each edge computing node generates a corresponding fire response strategy based on the fire response rules corresponding to the subtasks.

3. The method according to claim 2, characterized in that, A dynamic load balancing algorithm is used to distribute subtasks to appropriate edge computing nodes, enabling multiple edge computing nodes to synchronously process subtasks of different priorities based on a distributed parallel computing architecture, including: Real-time monitoring of resource utilization of each edge computing node; construction of node health assessment model based on resource utilization; obtaining availability score for each node; matching availability score with priority of subtask to be assigned. A priority queue for task allocation is established based on the availability score. A task scheduling algorithm that takes load balancing into account is adopted to prioritize the allocation of priority subtasks to nodes with availability scores higher than a preset score threshold. The load level of each node is kept consistent by dynamically adjusting the task allocation ratio. Each edge computing node starts task processing according to the assigned task priority order, and maintains computing synchronization between nodes through a timestamp-based message passing mechanism, so that subtasks of different priorities can be processed in a coordinated and consistent parallel manner under the distributed architecture.

4. The method according to claim 1, characterized in that, Based on the fire response decision results, the fire protection system within the detection chamber is controlled to perform corresponding fire-fighting operations. The fire protection system includes a portable fire extinguishing device, which can autonomously move to the optimal fire-fighting position based on the fire source location information. Based on the fire response decision results, the fire source location information and fire development parameters are analyzed, and the optimal fire extinguishing location is determined based on the fire source location information and the fire development parameters. Based on the optimal fire extinguishing location and the real-time obstacle distribution inside the detection chamber, a movement trajectory that avoids obstacles and has the shortest path is planned for the mobile fire extinguishing device, and a motion control command sequence containing position, speed and direction is generated. The mobile fire extinguishing device is controlled to execute the motion control command sequence, continuously update the surrounding environment information and optimize the movement trajectory in real time during autonomous movement, until it reaches the optimal fire extinguishing position and then carries out fire extinguishing operations.

5. The method according to claim 1, characterized in that, The system adaptively adjusts the spray pressure and angle of the extinguishing agent according to the fire development level, monitors the effectiveness of firefighting operations in real time, and feeds the monitoring results back to the deep learning model, including: The current fire intensity index is calculated based on the fire source temperature, flame height, and spread rate. The initial injection pressure and injection angle of the extinguishing agent are then determined based on the fire intensity index. The initial injection pressure and the injection angle are used as control parameters to perform fire extinguishing operations. At the same time, the rate of change of fire source temperature and the coverage area of ​​fire extinguishing agent are continuously collected through a sensor network. The fire extinguishing effect evaluation index is calculated based on the rate of change of fire source temperature and the coverage area of ​​fire extinguishing agent. The fire extinguishing effect evaluation index is input into the deep learning model. The initial spray pressure and the spray angle are fine-tuned in real time according to the fire extinguishing effect evaluation index. The fire extinguishing effect is improved through parameter iteration optimization. The optimized parameter combination is then stored in the deep learning model.

6. A multi-sensor-based fire monitoring and fire suppression system for a detection chamber, used to implement the method described in any one of claims 1-5, characterized in that, include: The first unit is used to acquire real-time monitoring data from multiple sensors installed inside the detection chamber, and to use a deep learning model to evaluate the fire status of the detection chamber in real time based on the real-time monitoring data from the multiple sensors, thereby obtaining the fire assessment result. The second unit is used to make fire response decisions based on the fire assessment results using a distributed edge computing approach. The fire assessment results are distributed to multiple edge computing nodes for parallel computing processing. Each edge computing node generates a corresponding fire handling strategy based on preset fire response rules, and votes on the fire handling strategies through a consensus protocol, selecting the fire handling strategy with the highest number of votes as the final fire response decision result. The third unit is used to control the fire protection system in the detection chamber to perform corresponding fire protection operations based on the fire response decision results. The fire protection system includes a mobile fire extinguishing device that can autonomously move to the optimal fire extinguishing position based on the fire source location information, and adaptively adjust the spray pressure and spray angle of the extinguishing agent according to the fire development level. It monitors the execution effect of the fire protection operations in real time and feeds the monitoring results back to the deep learning model to optimize the fire assessment accuracy of the deep learning model.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images