Intelligent fire safety pre-judgment management linkage system based on Internet

Abstract

The invention, which relates to the technical field of fire safety management, discloses an Internet-based intelligent fire safety pre-judgment management linkage system comprising a real-time monitoring module, an intelligent pre-judgment module, an automatic linkage module, an edge-cloud cooperation module and a model loading module. The real-time monitoring module is responsible for all-weather monitoring of various parameters in a fire-fighting environment, including fire-fighting equipment states, environment parameters and possible fire signs. According to the intelligent fire safety pre-judgment management linkage system based on the Internet, close cooperation between the edge nodes and the cloud is achieved by optimizing the edge computing architecture, the edge nodes can rapidly obtain more computing resources or data support from the cloud, the multiple edge nodes work cooperatively in combination with load balancing, and the safety of fire safety pre-judgment management is improved. Complex data analysis tasks are jointly processed, and the fire response speed is remarkably increased. And in cooperation with standardized protocol layering, efficient cooperative work among the modules is realized, and fire response delay is further reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fire safety management, and in particular to an Internet-based intelligent fire safety prejudgment management linkage system. Background Art

[0002] Fire safety management is the process of preventing, controlling, and eliminating fire risks through systematic measures to protect lives and property. Its core objectives are fire prevention, emergency response, and personnel safety. Traditional management methods rely on manual inspections, resulting in an average two-hour delay in identifying potential hazards, a high rate of missed detections, and emergency measures typically initiated only after a fire has occurred. With the development of automated management and the continuous maturity of computer technology, intelligent fire safety prediction and management linkage systems have emerged, gradually replacing traditional manual inspections and becoming the mainstream method of fire safety management.

[0003] Specifically, the Smart Fire Safety Predictive Management Linkage System is an intelligent platform based on the Internet of Things (IoT), big data, and AI. It enables closed-loop management of fire risk from a complete process: real-time monitoring, intelligent prediction, and automatic linkage. Compared to manual inspections, this system not only provides around-the-clock inspections but also enables predictive maintenance (based on equipment lifecycle models), providing enhanced fire prevention and control capabilities.

[0004] However, in actual application, the existing prediction and management linkage system is inseparable from edge computing. Currently, 80% of AI analysis still relies on the cloud (AWS / Azure). Edge nodes only perform simple filtering, resulting in an average delay of 4.2 seconds in fire identification. In addition, the existing system is essentially an embedded system and cannot dynamically load new algorithm models. Model updates require on-site firmware burning, which takes more than 2 hours per device. Summary of the Invention

[0005] The purpose of the present invention is to provide an Internet-based intelligent fire safety prediction management linkage system to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solutions: an Internet-based intelligent fire safety prediction management linkage system, comprising:

[0007] Real-time monitoring module: responsible for monitoring various parameters in the fire environment around the clock, including the status of fire equipment (such as fire extinguisher pressure, whether fire hydrants are available, etc.), environmental parameters (such as temperature, humidity, smoke concentration, flames, etc.) and possible signs of fire;

[0008] Intelligent prediction module: uses AI algorithms to analyze data collected by the real-time monitoring module to predict the possibility of fire;

[0009] Automatic linkage module: Based on the results of the intelligent prediction module, it automatically triggers corresponding emergency response measures, such as starting fire sprinklers and alarms;

[0010] Edge-cloud collaboration module: responsible for data interaction and computing resource scheduling between edge nodes and the cloud, enabling edge nodes to quickly obtain more computing resources or data support from the cloud when needed;

[0011] Model loading module: Responsible for dynamically loading and updating AI algorithm models, eliminating the need for on-site firmware burning. It decouples the algorithm model from the main program, and the model is loaded into memory as an independent file for execution at runtime.

[0012] Furthermore, the real-time monitoring module collects data in real time through IoT sensors deployed in various places and uploads it to edge nodes or the cloud for processing. Specifically, the IoT sensors perceive and measure environmental parameters or device status in real time, connect to the network wirelessly or wiredly, and transmit and share data in real time. For example, temperature sensors are used to detect changes in ambient temperature, smoke sensors are used to detect smoke concentration, and flame sensors are used to directly detect the presence of flames.

[0013] Furthermore, the deployment locations of the IoT sensors include but are not limited to near fire-fighting equipment (deployed near fire-fighting equipment such as fire extinguishers and fire hydrants to monitor the status of the equipment, such as whether the pressure of the fire extinguisher is sufficient and whether the fire hydrant is available), areas sensitive to environmental parameters (in high-risk areas prone to fire, such as kitchens, computer rooms, warehouses, etc., temperature sensors, humidity sensors, smoke concentration sensors, etc. are deployed to monitor changes in environmental parameters in real time. Changes in these parameters indicate the early signs of fire), key passages and exits (sensors are deployed on key passages, exits and evacuation routes of buildings to monitor direct signs of fire such as flames and smoke to ensure that they can respond quickly and guide people to evacuate when a fire occurs), near electrical equipment and lines (electrical equipment and lines are one of the common sources of fire. Deploying sensors at these locations can monitor the temperature, current and other parameters of electrical equipment in real time to promptly detect potential fire risks), storage areas for flammable and explosive items (for areas where flammable and explosive items are stored, corresponding sensors are deployed to monitor the status of these items and changes in the surrounding environment in real time to prevent the occurrence of fire and explosion accidents).

[0014] Furthermore, the intelligent prediction module runs an AI algorithm model on the edge node or in the cloud to analyze real-time data to generate prediction results. The AI ​​algorithm architecture is as follows:

[0015] 1) Input layer

[0016] Multimodal data fusion: temperature curve (1Hz sampling) + smoke particle concentration (0.1% obs / m accuracy) + CO / VOC gas (ppm level) + video flame characteristics (30fps RTSP stream);

[0017] Time series data processing: LSTM network processes sensor historical sequences with a time window of ≥15 minutes;

[0018] 2) Core Model

[0019] Lightweight model: deployed on the edge, decision tree ensemble based on the TinyML architecture, with an F1-score ≥ 0.92;

[0020] High-precision model: Deployed in the cloud, 3D CNN + Transformer hybrid model, with a false alarm rate of <0.5 times / 1,000 days;

[0021] 3) Output layer

[0022] Risk level quantification: 0-100 points, with 50 points triggering an early warning and 80 points initiating linkage.

[0023] Furthermore, the edge-cloud division of labor of the intelligent prediction module is as follows:

[0024] Edge nodes:

[0025] Mission: Anomaly detection, using locally deployed AI algorithm models to quickly analyze data collected by the real-time monitoring module to detect anomalies or potential signs of fire;

[0026] Delay requirement: <500ms,

[0027] Hardware configuration: NVIDIA Jetson Orin (20TOPS);

[0028] Cloud:

[0029] Mission: Multi-building situation prediction, using complex AI algorithm models such as graph neural networks (GNNs) to globally predict and analyze the fire safety situation of multiple buildings or areas to provide a more comprehensive and accurate fire risk assessment;

[0030] Delay requirement: <5s;

[0031] Hardware configuration: AWS EC2 P4d instance (8×A100, equipped with 8 NVIDIA A100 GPUs).

[0032] Furthermore, the automatic linkage module connects the prediction results with the emergency response equipment through preset logical rules and communication protocols to achieve automatic linkage, and the execution logic of the automatic linkage module includes a hierarchical response mechanism, which is as follows:

[0033] Level 1. Warning level:

[0034] Action: Sound and light alarm + LED marking of hidden danger location;

[0035] Trigger condition: Two or more sensors exceed thresholds, such as temperature > 60°C and smoke > 2% obs / m;

[0036] Level 2. Disposal level:

[0037] Action: Close the fire damper and start the smoke exhaust fan (Modbus TCP command);

[0038] Conditions: AI confidence > 70% and manual confirmation delay > 30 seconds;

[0039] Level 3. Emergency:

[0040] Action: sprinkler system activation + elevator forced landing + emergency broadcast;

[0041] Conditions: Flame video verification, ResNet50 classification probability > 95%.

[0042] Furthermore, the automatic linkage module achieves full-link collaboration through a communication protocol stack (a set of layered protocols used to achieve communication between devices in a computer network, each layer is responsible for a specific function and interacts with adjacent layers through standardized interfaces), that is, data transmission and instruction execution between IoT sensors, edge nodes, the cloud, and firefighting equipment;

[0043] 1)

[0044] Protocol stack: LoRa physical layer + LoRaWAN MAC layer + MQTT-SN application layer;

[0045] Data flow: IoT sensors send data packets via LoRa radio frequency, periodically reporting environmental data (such as temperature and smoke concentration) to edge nodes. The LoRaWAN gateway aggregates the data, and the edge node parses the data and issues configuration instructions (such as sampling frequency adjustment) via MQTT-SN.

[0046] 2)

[0047] Protocol stack: Ethernet physical layer + TCP / IP + TLS + HTTP / RESTful API;

[0048] Data flow: Edge nodes aggregate data and upload it to the cloud, which then sends down analysis results or control strategies (such as optimizing fire warning models).

[0049] 3)

[0050] Protocol stack: HTTP / WebSocket (real-time commands) application layer + cloud-based rules engine (such as AWS IoT Rules, triggering device actions);

[0051] Command flow: The cloud sends a fire extinguishing command (activating the sprinkler system), and the device returns a status confirmation;

[0052] 4) (Local Emergency Coordination)

[0053] Protocol stack: DDS application layer + Wi-Fi physical layer or direct hardware trigger (such as GPIO);

[0054] Process: After the edge node detects a dangerous situation, it directly controls the device (such as triggering an audible or visual alarm) by bypassing the cloud;

[0055] 5) (Direct connection mode)

[0056] Protocol stack: CoAP over UDP+DTLS (secure), LoRaWAN (via gateway);

[0057] Usage: Sensors are directly connected to the cloud for remote monitoring (e.g., in remote areas) and rely on edge nodes for preliminary processing to reduce cloud load.

[0058] 6) (Device Collaboration)

[0059] Protocol stack: Zigbee, WirelessHART, or dedicated fire protection protocols (such as NFPA standards);

[0060] Process: After the sprinkler system is triggered, the smoke exhaust equipment is activated.

[0061] Furthermore, the edge-cloud collaboration module establishes a tight edge-cloud collaboration mechanism, including data flow optimization, as follows:

[0062] Edge Cache: locally retains the original data for the last 72 hours;

[0063] Cloud synchronization: Metadata is compressed and transmitted using the Apache Parquet format, with a compression ratio of ≥10:1, and resumable transmission is performed based on MQTT persistent sessions.

[0064] Furthermore, the edge-cloud collaboration module includes the following submodules:

[0065] Data interaction submodule: responsible for data transmission and synchronization between edge nodes and the cloud, ensuring the real-time and consistency of data;

[0066] Resource scheduling submodule: Dynamically schedules computing resources based on the computing needs of edge nodes and the resource status of the cloud to meet real-time requirements;

[0067] Load balancing submodule: distributes computing tasks among multiple edge nodes to achieve load balancing and improve the efficiency of the overall system.

[0068] Furthermore, the resource scheduling submodule monitors the computing load and performance indicators of the edge nodes, evaluates their computing needs in real time, and communicates with the cloud to obtain the resource status of the cloud, including available computing resources, network bandwidth, etc., and combines the needs of the edge nodes with the resource status of the cloud to derive a computing resource scheduling strategy, migrate the computing tasks of some edge nodes to the cloud for execution, or obtain additional computing resources from the cloud to enhance the computing power of the edge nodes.

[0069] Furthermore, the execution process of the load balancing submodule is as follows:

[0070] Monitor the computing load of each edge node and evaluate its processing capacity;

[0071] Allocate computing tasks based on their size and complexity, as well as the processing capabilities of edge nodes;

[0072] Dynamically adjust the allocation strategy of computing tasks to adapt to real-time changing computing needs.

[0073] Furthermore, the model loading module dynamically loads the AI ​​algorithm model including the following steps:

[0074] Step 1. Model preparation: Convert the trained model into a format compatible with edge devices (such as TensorFlowLite.tflite and ONNX.onnx), assign a unique identifier (model ID) to each model file, and record the model to be used based on specific conditions (such as time, location, device type, etc.) in the system configuration file.

[0075] Step 2: Model upload: upload the model file to the specified location of the system, such as cloud storage or local storage of the edge node;

[0076] Step 3: Model loading: Based on the preset trigger conditions, read the configuration information, determine the model to be loaded, and trigger the model loading operation to load the model file from the storage location into the memory, ready for execution;

[0077] Step 4: Associating the model with the main program: After the model is loaded, associate the model with the main program, that is, introduce the model as an independent module or object into the main program so that the main program can call the model for data analysis or prediction;

[0078] Step 5: Model execution: The main program calls the loaded model, which is executed in memory, analyzes or predicts the input data, and outputs the results.

[0079] The model loading module checks the update status of the model file regularly or as needed. When a new model file is available, the system will repeat steps 2 to 4 above to achieve dynamic update of the model. When the model is updated or no longer needed, the old model resources will be cleaned up to release the occupied memory space.

[0080] The present invention provides an Internet-based intelligent fire safety prediction and management linkage system, which has the following beneficial effects:

[0081] This system monitors the fire environment in real time and quickly runs AI algorithm models on edge computing nodes or the cloud for analysis. Once signs of fire are detected, emergency response measures can be triggered immediately, greatly shortening the time interval from fire detection to response.

[0082] 1. By optimizing the edge computing architecture, we achieve close collaboration between edge nodes and the cloud, enabling them to quickly access more computing resources or data support from the cloud. Combined with load balancing, multiple edge nodes work together to handle complex data analysis tasks, significantly improving fire response speed. Combined with standardized protocol layering, we seamlessly connect IoT sensor data, edge computing decisions, cloud analysis, and terminal execution devices, achieving efficient collaboration between modules and further reducing fire response delays.

[0083] 2. Decoupling the algorithm model from the main program allows the model to be dynamically loaded into memory and executed as an independent file or module at runtime, rather than statically bound to the firmware during compilation. This allows the system to easily load and update new AI algorithm models without the need to burn firmware on site, greatly reducing the time and cost of model updates. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 This is a logic block diagram of an Internet-based intelligent fire safety prediction management linkage system of the present invention;

[0085] Figure 2 This is a schematic diagram of the AI ​​algorithm architecture of an Internet-based intelligent fire safety prediction and management linkage system of the present invention;

[0086] Figure 3This is a schematic diagram of the dynamic loading steps of the AI ​​algorithm model of an Internet-based intelligent fire safety prediction management linkage system of the present invention. DETAILED DESCRIPTION

[0087] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0088] like Figure 1-Figure 3 As shown, an Internet-based intelligent fire safety prediction management linkage system includes:

[0089] Real-time monitoring module: monitors various parameters in the fire environment around the clock, including the status of fire equipment (such as fire extinguisher pressure, whether fire hydrants are available, etc.), environmental parameters (such as temperature, humidity, smoke concentration, flames, etc.) and possible signs of fire.

[0090] In this embodiment, the real-time monitoring module collects data in real time through IoT sensors deployed in various places and uploads it to edge nodes or the cloud for processing. The specific operations are as follows: IoT sensors perceive and measure environmental parameters or device status in real time, connect to the network wirelessly or wiredly, and transmit and share data in real time. For example, temperature sensors are used to detect changes in ambient temperature, smoke sensors are used to detect smoke concentration, and flame sensors are used to directly detect the presence of flames. Among them, the deployment locations of IoT sensors include but are not limited to near fire-fighting equipment (deployed near fire-fighting equipment such as fire extinguishers and fire hydrants to monitor the status of the equipment, such as whether the pressure of the fire extinguisher is sufficient and whether the fire hydrant is available), areas sensitive to environmental parameters (in high-risk areas prone to fire, such as kitchens, computer rooms, warehouses, etc., temperature sensors, humidity sensors, smoke concentration sensors, etc. are deployed to monitor changes in environmental parameters in real time. Changes in these parameters indicate the initial signs of fire), critical passages and exits (sensors are deployed on key passages, exits and evacuation routes of buildings to monitor direct signs of fire such as flames and smoke to ensure a rapid response and guide personnel evacuation when a fire occurs), near electrical equipment and lines (electrical equipment and lines are one of the common sources of fire. Deploying sensors at these locations can monitor the temperature, current and other parameters of electrical equipment in real time to promptly detect potential fire risks), and storage areas for flammable and explosive items (for areas where flammable and explosive items are stored, corresponding sensors are deployed to monitor the status of these items and changes in the surrounding environment in real time to prevent the occurrence of fire and explosion accidents).

[0091] Intelligent prediction module: Uses AI algorithms to analyze data collected by the real-time monitoring module to predict the possibility of fire.

[0092] In this embodiment, the intelligent prediction module runs an AI algorithm model on the edge node or in the cloud to analyze real-time data to generate prediction results. The AI ​​algorithm architecture is as follows:

[0093] 1) Input layer

[0094] Multimodal data fusion: temperature curve (1Hz sampling) + smoke particle concentration (0.1% obs / m accuracy) + CO / VOC gas (ppm level) + video flame characteristics (30fps RTSP stream);

[0095] Time series data processing: LSTM network processes sensor historical sequences with a time window of ≥15 minutes;

[0096] 2) Core Model

[0097] Lightweight model: deployed on the edge, decision tree ensemble based on the TinyML architecture, with an F1-score ≥ 0.92;

[0098] High-precision model: Deployed in the cloud, 3D CNN + Transformer hybrid model, with a false alarm rate of <0.5 times / 1,000 days;

[0099] 3) Output layer

[0100] Risk level quantification: 0-100 points, with 50 points triggering an early warning and 80 points initiating linkage.

[0101] In this embodiment, the edge-cloud division of labor of the intelligent prediction module is as follows:

[0102] Edge nodes:

[0103] Mission: Anomaly detection, using locally deployed AI algorithm models to quickly analyze data collected by the real-time monitoring module to detect anomalies or potential signs of fire;

[0104] Delay requirement: <500ms,

[0105] Hardware configuration: NVIDIA Jetson Orin (20TOPS).

[0106] Cloud:

[0107] Mission: Multi-building situation prediction, using complex AI algorithm models such as graph neural networks (GNNs) to globally predict and analyze the fire safety situation of multiple buildings or areas to provide a more comprehensive and accurate fire risk assessment;

[0108] Delay requirement: <5s;

[0109] Hardware configuration: AWS EC2 P4d instance (8×A100, equipped with 8 NVIDIA A100 GPUs).

[0110] Automatic linkage module: According to the results of the intelligent prediction module, it automatically triggers corresponding emergency response measures, such as starting fire sprinklers and alarms.

[0111] In this embodiment, the automatic linkage module connects the prediction results with the emergency response equipment through preset logical rules and communication protocols to achieve automatic linkage, and the execution logic of the automatic linkage module includes a hierarchical response mechanism, which is as follows:

[0112] Level 1. Warning level:

[0113] Action: Sound and light alarm + LED marking of hidden danger location;

[0114] Trigger condition: Two or more sensors exceed thresholds, such as temperature > 60°C and smoke > 2% obs / m;

[0115] Level 2. Disposal level:

[0116] Action: Close the fire damper and start the smoke exhaust fan (Modbus TCP command);

[0117] Conditions: AI confidence > 70% and manual confirmation delay > 30 seconds;

[0118] Level 3. Emergency:

[0119] Action: sprinkler system activation + elevator forced landing + emergency broadcast;

[0120] Conditions: Flame video verification, ResNet50 classification probability > 95%;

[0121] In this embodiment, the automatic linkage module realizes full-link collaboration through the communication protocol stack (Protocol Stack is a set of layered protocols used to realize communication between devices in computer networks. Each layer has specific functions and interacts with adjacent layers through standardized interfaces), that is, data transmission and instruction execution between IoT sensors, edge nodes, cloud and fire-fighting equipment.

[0122] 1)

[0123] Protocol stack: LoRa physical layer + LoRaWAN MAC layer + MQTT-SN application layer;

[0124] Data flow: IoT sensors send data packets via LoRa radio frequency, periodically reporting environmental data (such as temperature and smoke concentration) to edge nodes. The LoRaWAN gateway aggregates the data, and the edge node parses the data and issues configuration instructions (such as sampling frequency adjustment) via MQTT-SN.

[0125] 2)

[0126] Protocol stack: Ethernet physical layer + TCP / IP + TLS + HTTP / RESTful API;

[0127] Data flow: Edge nodes aggregate data and upload it to the cloud, which then sends down analysis results or control strategies (such as optimizing fire warning models).

[0128] 3)

[0129] Protocol stack: HTTP / WebSocket (real-time commands) application layer + cloud-based rules engine (such as AWS IoT Rules, triggering device actions);

[0130] Command flow: The cloud sends a fire extinguishing command (activating the sprinkler system), and the device returns a status confirmation;

[0131] 4) (Local Emergency Coordination)

[0132] Protocol stack: DDS application layer + Wi-Fi physical layer or direct hardware trigger (such as GPIO);

[0133] Process: After the edge node detects a dangerous situation, it directly controls the device (such as triggering an audible or visual alarm) by bypassing the cloud;

[0134] 5) (Direct connection mode)

[0135] Protocol stack: CoAP over UDP+DTLS (secure), LoRaWAN (via gateway);

[0136] Usage: Sensors are directly connected to the cloud for remote monitoring (e.g., in remote areas) and rely on edge nodes for preliminary processing to reduce cloud load.

[0137] 6) (Device Collaboration)

[0138] Protocol stack: Zigbee, WirelessHART, or dedicated fire protection protocols (such as NFPA standards);

[0139] Process: After the sprinkler system is triggered, the smoke exhaust equipment is activated.

[0140] Edge-cloud collaboration module: Data interaction and computing resource scheduling between edge nodes and the cloud enable edge nodes to quickly obtain more computing resources or data support from the cloud when needed.

[0141] Data interaction submodule: Data transmission and synchronization between edge nodes and the cloud to ensure the real-time and consistency of data.

[0142] Resource scheduling submodule: Dynamically schedules computing resources based on the computing needs of edge nodes and the resource status of the cloud to meet real-time requirements; the resource scheduling submodule monitors the computing load and performance indicators of edge nodes, evaluates their computing needs in real time, and communicates with the cloud to obtain the resource status of the cloud, including available computing resources, network bandwidth, etc., and combines the needs of edge nodes with the resource status of the cloud to derive computing resource scheduling strategies, migrate some edge node computing tasks to the cloud for execution, or obtain additional computing resources from the cloud to enhance the computing power of edge nodes.

[0143] Load balancing submodule: distributes computing tasks among multiple edge nodes to achieve load balancing and improve the efficiency of the overall system; the execution process of the load balancing submodule is as follows:

[0144] Monitor the computing load of each edge node and evaluate its processing capacity;

[0145] Allocate computing tasks based on their size and complexity, as well as the processing capabilities of edge nodes;

[0146] Dynamically adjust the allocation strategy of computing tasks to adapt to real-time changing computing needs.

[0147] In this embodiment, the edge-cloud collaboration module establishes a close edge-cloud collaboration mechanism, including data flow optimization, as follows:

[0148] Edge Cache: locally retains the original data for the last 72 hours;

[0149] Cloud synchronization: Metadata is compressed and transmitted using the Apache Parquet format, with a compression ratio of ≥10:1, and resumable transmission is performed based on MQTT persistent sessions.

[0150] Model loading module: Dynamically loads and updates AI algorithm models without the need for on-site firmware burning. It decouples the algorithm model from the main program and loads the model into memory as an independent file for execution at runtime.

[0151] In this embodiment, the model loading module dynamically loads the AI ​​algorithm model including the following steps:

[0152] Step 1. Model preparation: Convert the trained model into a format compatible with edge devices (such as TensorFlowLite.tflite and ONNX.onnx), assign a unique identifier (model ID) to each model file, and record the model to be used based on specific conditions (such as time, location, device type, etc.) in the system configuration file.

[0153] Step 2: Model upload: upload the model file to the specified location of the system, such as cloud storage or local storage of the edge node;

[0154] Step 3: Model loading: Based on the preset trigger conditions, read the configuration information, determine the model to be loaded, and trigger the model loading operation to load the model file from the storage location into the memory, ready for execution;

[0155] Step 4: Associating the model with the main program: After the model is loaded, associate the model with the main program, that is, introduce the model as an independent module or object into the main program so that the main program can call the model for data analysis or prediction;

[0156] Step 5: Model execution: The main program calls the loaded model, which is executed in memory, analyzes or predicts the input data, and outputs the results.

[0157] The model loading module checks the update status of the model file regularly or as needed. When a new model file is available, the system will repeat steps 2 to 4 above to achieve dynamic update of the model. When the model is updated or no longer needed, the old model resources will be cleaned up to release the occupied memory space.

[0158] The embodiments of the present invention are presented for purposes of illustration and description and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those skilled in the art. The embodiments are chosen and described in order to better illustrate the principles of the invention and its practical application and to enable those skilled in the art to understand the invention and design various embodiments with various modifications as suited for specific applications.

Claims

1. An Internet-based intelligent fire safety prediction management linkage system, characterized by: Including real-time monitoring module, intelligent prediction module, automatic linkage module, edge-cloud collaboration module, and model loading module; The real-time monitoring module is responsible for monitoring various parameters in the fire environment around the clock, including the status of fire equipment, environmental parameters, and possible signs of fire; The intelligent prediction module uses AI algorithms to analyze the data collected by the real-time monitoring module to predict the possibility of fire; The automatic linkage module automatically triggers corresponding emergency response measures based on the results of the intelligent prediction module; The edge-cloud collaboration module is responsible for data interaction and computing resource scheduling between edge nodes and the cloud, so that edge nodes can quickly obtain more computing resources or data support from the cloud. The edge-cloud collaboration module includes the following submodules: Data interaction submodule: responsible for data transmission and synchronization between edge nodes and the cloud; Resource scheduling submodule: dynamically schedules computing resources based on the computing needs of edge nodes and the resource status of the cloud; Load balancing submodule: distributes computing tasks among multiple edge nodes to achieve load balancing; The model loading module is responsible for dynamically loading and updating the AI ​​algorithm model, decoupling the algorithm model from the main program, and the model is loaded into the memory as an independent file for execution at runtime.

2. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The real-time monitoring module collects data in real time through IoT sensors deployed in various locations and uploads it to edge nodes or the cloud for processing. Specifically, the IoT sensors sense and measure environmental parameters or device status in real time, connect to the network wirelessly or wiredly, and transmit and share the data in real time; The deployment locations of the IoT sensors include near fire-fighting equipment, areas sensitive to environmental parameters, key passages and exits, near electrical equipment and lines, and storage areas for flammable and explosive items.

3. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The intelligent prediction module runs an AI algorithm model on the edge node or in the cloud to analyze real-time data to generate prediction results. The AI ​​algorithm architecture is as follows: 1) Input layer Multimodal data fusion: temperature curve + smoke particle concentration + CO / VOC gas + video flame characteristics; Time series data processing: LSTM network processes sensor historical sequences with a time window of ≥15 minutes; 2) Core Model Lightweight model: deployed on the edge, decision tree ensemble based on the TinyML architecture, with an F1-score ≥ 0.92; High-precision model: Deployed in the cloud, 3D CNN + Transformer hybrid model, with a false alarm rate of <0.5 times / 1,000 days; 3) Output layer Risk level quantification: 0-100 points, with 50 points triggering an early warning and 80 points initiating linkage.

4. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The edge-cloud division of labor of the intelligent prediction module is as follows: Edge nodes: Mission: Anomaly detection, using locally deployed AI algorithm models to quickly analyze data collected by the real-time monitoring module to detect anomalies or potential signs of fire; Delay requirement: <500ms, Hardware configuration: NVIDIA Jetson Orin; Cloud: Task: Multi-building situation prediction, using graph neural network models to perform global prediction and analysis of the fire safety situation of multiple buildings or areas to provide fire risk assessment; Delay requirement: <5s; Hardware configuration: AWS EC2 P4d instance.

5. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The execution logic of the automatic linkage module includes a hierarchical response mechanism, as follows: Level 1. Warning level: Action: Sound and light alarm + LED marking of hidden danger location; Trigger condition: more than two sensors exceed the threshold; Level 2. Disposal level: Action: Close the fire damper and start the smoke exhaust fan; Conditions: AI confidence > 70% and manual confirmation delay > 30 seconds; Level 3. Emergency: Action: sprinkler system activation + elevator forced landing + emergency broadcast; Conditions: Flame video verification, ResNet50 classification probability > 95%.

6. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The automatic linkage module realizes full-link collaboration through the communication protocol stack, that is, data transmission and command execution between IoT sensors, edge nodes, cloud and firefighting equipment; 1) IoT sensors edge nodes Protocol stack: LoRa physical layer + LoRaWAN MAC layer + MQTT-SN application layer; Data flow: IoT sensors periodically report environmental data to edge nodes, which then issue configuration instructions. 2) Edge Node Cloud Protocol stack: Ethernet physical layer + TCP / IP + TLS + HTTP / RESTful API; Data flow: Edge nodes aggregate data and upload it to the cloud, which then sends down analysis results or control strategies. 3) Cloud Firefighting equipment Protocol stack: HTTP / WebSocket application layer + cloud-based rule engine; Command flow: The cloud sends a fire extinguishing command, and the device returns a status confirmation; 4) Edge Nodes Firefighting equipment Protocol stack: DDS application layer + Wi-Fi physical layer or direct hardware trigger; Process: After the edge node detects a dangerous situation, it directly controls the device, bypassing the cloud; 5) IoT Sensors Cloud Protocol stack: CoAP over UDP+DTLS, LoRaWAN; Usage: Sensors are directly connected to the cloud for remote monitoring and rely on edge nodes for preliminary processing to reduce cloud load; 6) Firefighting equipment Firefighting equipment Protocol stack: Zigbee, WirelessHART or dedicated fire protection protocol; Process: After the sprinkler system is triggered, the smoke exhaust equipment is activated.

7. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The edge-cloud collaboration module establishes an edge-cloud collaboration mechanism, including data flow optimization, as follows: Edge Cache: locally retains the original data for the last 72 hours; Cloud synchronization: Metadata is compressed and transmitted using the Apache Parquet format, with a compression ratio of ≥10:1, and resumable transmission is performed based on MQTT persistent sessions.

8. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The resource scheduling submodule monitors the computing load and performance indicators of the edge nodes, evaluates their computing needs in real time, and communicates with the cloud to obtain the resource status of the cloud. It combines the needs of the edge nodes with the resource status of the cloud to derive a computing resource scheduling strategy.

9. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The execution process of the load balancing submodule: Monitor the computing load of each edge node and evaluate its processing capacity; Allocate computing tasks based on their size and complexity, as well as the processing capabilities of edge nodes; Dynamically adjust the allocation strategy of computing tasks to adapt to real-time changing computing needs.

10. The Internet-based intelligent fire safety prediction management linkage system according to claim 1 is characterized in that: The model loading module dynamically loads the AI ​​algorithm model including the following steps: Step 1: Model preparation: Convert the trained model into a format compatible with edge devices and assign a unique identifier, namely, a model ID, to each model file. At the same time, record the model to be used based on specific conditions in the system configuration file. Step 2: Model upload: upload the model file to the specified location of the system; Step 3: Model loading: Based on the preset trigger conditions, read the configuration information, determine the model to be loaded, and trigger the model loading operation to load the model file from the storage location into the memory, ready for execution; Step 4: Associating the model with the main program: After the model is loaded, associate the model with the main program, that is, introduce the model into the main program as an independent module or object; Step 5: Model execution: The main program calls the loaded model, which is executed in memory, analyzes or predicts the input data, and outputs the results. The model loading module checks the update status of the model file regularly or as needed. When a new model file is available, the system will repeat steps 2 to 4 above to achieve dynamic update of the model. When the model is updated or no longer needed, the old model resources will be cleaned up to release the occupied memory space.

Images