A fire-fighting Internet of Things safety transmission method, device, equipment and medium

Abstract

The application discloses a fire-fighting Internet of Things safety transmission method, device, equipment and medium; the method comprises the following steps: an edge node acquires local fire-fighting data of multiple modes and performs data preprocessing on the local fire-fighting data to obtain target fire-fighting data, wherein the local fire-fighting data is fire-fighting data collected by a local sensing device; the edge node performs feature extraction on the target fire-fighting data in a preset feature dimension based on a pre-trained machine learning model to obtain a data feature vector; the edge node performs encryption processing on the data feature vector, and sends the encrypted data feature vector obtained after the encryption processing to a cloud server, so that the cloud server performs decryption processing on the encrypted data feature vectors sent by each edge node, and monitors a fire disaster based on the decrypted data feature vectors obtained after the decryption and the local pre-trained machine learning model. The application embodiment can greatly reduce network bandwidth occupation and end-to-end response delay, and improve the real-time performance and accuracy of fire disaster early warning.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of fire-fighting Internet of Things, and in particular to a fire-fighting Internet of Things security transmission method, device, equipment and medium. BACKGROUND

[0002] With the wide application of Internet of Things technology in the field of fire-fighting, a fire-fighting Internet of Things system can collect a large amount of fire-fighting data such as temperature, smoke concentration, and flame image information in real time by deploying a wide range of front-end sensing devices such as temperature sensors, smoke concentration sensors, and high-definition cameras.

[0003] However, the raw data collected by the front-end sensing devices in the above prior art has a large amount of data, which is directly uploaded to the cloud for processing, not only occupying a large amount of network bandwidth resources, but also causing high data transmission delay, affecting the timeliness of fire warning. In addition, there is a risk of sensitive data leakage in the process of uploading data to the cloud, threatening user privacy and fire-fighting information security.

[0004] Therefore, there is an urgent need for a fire-fighting Internet of Things security transmission method with low network bandwidth occupation, low delay, and high security. SUMMARY

[0005] The present application provides a fire-fighting Internet of Things security transmission method, device, equipment and medium, which can realize low network bandwidth occupation, low delay, and high security for fire-fighting data transmission.

[0006] In a first aspect, the embodiments of the present application provide a fire-fighting Internet of Things security transmission method applied to an edge node in a fire-fighting data management system, the fire-fighting data management system comprising a cloud server and a plurality of edge nodes, the edge node being used to perform a data transmission task of transmitting data to the cloud server, the plurality of edge nodes comprising at least one core edge node, the core edge node comprising at least three identical data function modules and a majority voting module, the data function module comprising the functions of data reception, processing and sending; the method comprises:

[0007] obtaining multi-modal local fire-fighting data and performing data preprocessing on the local fire-fighting data to obtain target fire-fighting data, the local fire-fighting data being fire-fighting data collected by local sensing devices, the multi-modal local fire-fighting data comprising at least one of the following: time series signal type, image / video type, and event trigger type;

[0008] perform feature extraction on the target fire-fighting data in a preset feature dimension based on a pre-trained machine learning model, to obtain a data feature vector; the preset feature dimension includes at least one of a fire source area, a high-temperature area, and a fire spread area, and the pre-trained machine learning model is obtained by the cloud server and the plurality of edge nodes through multi-modal federated learning based on parameter aggregation rules for phased collaborative training;

[0009] perform encryption processing on the data feature vector, and send the encrypted data feature vector obtained after the encryption processing to the cloud server, so that the cloud server performs decryption processing on the encrypted data feature vectors sent by each edge node, and monitors a fire situation based on the decrypted data feature vectors obtained after the decryption and a local pre-trained machine learning model;

[0010] For each core edge node, send the data feature vectors output by each data function module to the plurality of majority voting modules, and compare the data feature vectors extracted by each data function module for the target fire-fighting data through the plurality of majority voting modules;

[0011] When there are identical data feature vectors in the plurality of data feature vectors, the majority voting module determines the data feature vectors with the largest number of identical data feature vectors as the target data feature vectors to be finally encrypted and sent to the cloud server by the core edge node.

[0012] In a second aspect, the embodiments of the present application provide a fire-fighting Internet of Things safety transmission method, applied to a cloud server in a fire-fighting data management system, the fire-fighting data management system including the cloud server and a plurality of edge nodes, and the method including:

[0013] Receiving encrypted data feature vectors sent by each edge node, the encrypted data feature vectors being data obtained by the edge nodes after acquiring local fire-fighting data and sequentially performing data preprocessing, feature extraction, and encryption processing on the local fire-fighting data;

[0014] Decrypting each encrypted data feature vector, and determining a fire situation according to each decrypted data feature vector obtained after the decryption and a local pre-trained machine learning model, the encrypted data feature vector and the decrypted data feature vector corresponding to each other.

[0015] In a third aspect, the embodiments of the present application also provide a fire-fighting Internet of Things secure transmission device, applied to an edge node in a fire-fighting data management system, the fire-fighting data management system comprising a cloud server and a plurality of edge nodes, the edge node being configured to perform a data transmission task of transmitting data to the cloud server, the plurality of edge nodes comprising at least one core edge node, the core edge node comprising at least three same data function modules and a majority voting module, the data function module comprising the functions of data receiving, processing and sending; the device comprising:

[0016] an acquisition module configured to acquire local fire-fighting data in multiple modalities and perform data preprocessing on the local fire-fighting data to obtain target fire-fighting data, the local fire-fighting data being fire-fighting data collected by a local sensing device, the local fire-fighting data in multiple modalities comprising at least one of the following: time series signals, images / videos and event triggers;

[0017] a feature extraction module configured to perform feature extraction on the target fire-fighting data in a preset feature dimension based on a pre-trained machine learning model to obtain a data feature vector; the preset feature dimension comprising at least one of the following: a fire source area, a high-temperature area and a fire spread area, the pre-trained machine learning model being obtained by the cloud server and the plurality of edge nodes through collaborative training in stages based on multi-modal federated learning according to a parameter aggregation rule;

[0018] an encryption module configured to perform encryption processing on the data feature vector and send the encrypted data feature vector obtained after the encryption processing to the cloud server, so that the cloud server performs decryption processing on the encrypted data feature vectors sent by each edge node, and monitors a fire situation based on the decrypted data feature vectors obtained after the decryption and a local pre-trained machine learning model; for each core edge node, the data feature vectors output by each data function module are sent to the majority voting module, and the data feature vectors extracted by each data function module for the target fire-fighting data are compared by the majority voting module; when there are same data feature vectors in a plurality of the data feature vectors, the data feature vectors with the largest number of same data feature vectors are determined as the target data feature vector of the core edge node to be finally encrypted and sent to the cloud server by the majority voting module.

[0019] In a fourth aspect, the embodiments of the present application also provide a fire-fighting Internet of Things secure transmission device, applied to a cloud server in a fire-fighting data management system, the fire-fighting data management system comprising the cloud server and a plurality of edge nodes, the device comprising:

[0020] The receiving module is configured to receive encrypted data feature vectors sent by the edge nodes, wherein the encrypted data feature vectors are obtained by the edge nodes through acquiring local fire-fighting data, and sequentially performing data preprocessing, feature extraction and encryption processing on the local fire-fighting data.

[0021] The determining module is configured to decrypt the encrypted data feature vectors, and determine a fire disaster according to the decrypted data feature vectors and a locally pre-trained machine learning model.

[0022] In a fifth aspect, an electronic device is provided, including:

[0023] one or more processors;

[0024] a memory configured to store one or more programs,

[0025] When the one or more programs are executed by the one or more processors, the one or more processors implement the fire-fighting Internet of Things secure transmission method according to any one of the first aspect, or implement the fire-fighting Internet of Things secure transmission method according to any one of the second aspect.

[0026] In a sixth aspect, a storage medium is provided, which stores a computer program, and when the program is executed by a processor, the fire-fighting Internet of Things secure transmission method according to any one of the first aspect, or the fire-fighting Internet of Things secure transmission method according to any one of the second aspect is implemented.

[0027] The embodiment of the application provides a fire-fighting Internet of Things safety transmission method, device, equipment and medium. The method is applied to an edge node in a fire-fighting data management system, the fire-fighting data management system comprises a cloud server and a plurality of edge nodes, the edge node is used for executing a data transmission task of transmitting data to the cloud server, and the method comprises the following steps: acquiring local fire-fighting data in multiple modes and performing data preprocessing on the local fire-fighting data to obtain target fire-fighting data. The local fire-fighting data is fire-fighting data collected by a local sensing device, and the local fire-fighting data in multiple modes comprises at least one of the following: a time sequence signal type, an image / video type and an event trigger type; a pre-trained machine learning model is used to extract features of the target fire-fighting data in a preset feature dimension to obtain a data feature vector; the preset feature dimension comprises at least one of the following: a fire source area, a high-temperature area and a fire spread area, and the pre-trained machine learning model is obtained through stage-by-stage collaborative training of the cloud server and the plurality of edge nodes based on multi-modal federated learning according to a parameter aggregation rule; the data feature vector is encrypted, and the encrypted data feature vector obtained after the encryption is sent to the cloud server, so that the cloud server decrypts the encrypted data feature vector sent by each edge node, and monitors a fire situation based on the decrypted data feature vector and the local pre-trained machine learning model.

[0028] In other words, the technical solution of this application, by performing localized preprocessing on the raw fire data collected by local sensing devices at the edge node, effectively filters out redundancy, noise, and invalid information, generating concise target fire data with fire discrimination value, thereby significantly reducing the amount of data transmitted subsequently. Furthermore, the edge node utilizes a pre-trained machine learning model, collaboratively trained by the cloud server and multiple edge nodes, to extract features from the target fire data, obtaining highly representative data feature vectors. This collaborative training mechanism ensures the model's generalization ability to fire features across multiple scenarios and regions, while avoiding cross-domain transmission of raw sensitive data, thus improving model adaptability and privacy security. Based on this, the edge node encrypts the data feature vectors and uploads them to the cloud server, ensuring the confidentiality and integrity of the feature data during transmission. After receiving and decrypting each encrypted data feature vector, the cloud server performs a fusion analysis based on the one-to-one correspondence of the decrypted data feature vectors and its locally deployed pre-trained machine learning model to accurately determine the fire situation. Therefore, compared with existing technologies, the fire protection IoT secure transmission method proposed in this application not only significantly reduces network bandwidth consumption and end-to-end response latency, improving the real-time performance and accuracy of fire early warning, but also effectively avoids user privacy leaks and data security risks through encryption mechanisms and collaborative model training. Thus, while ensuring the security and reliability of the fire protection data management system, it constructs a highly efficient, low-latency, and privacy-protected intelligent fire protection data processing and fire assessment system. Furthermore, the technical solution of this application is simple and convenient to implement, easy to popularize, and has a wider range of applications. Attached Figure Description

[0029] Figure 1 This is a schematic diagram illustrating an application scenario of the fire protection IoT security transmission method provided in an embodiment of this application.

[0030] Figure 2 A schematic flowchart illustrating a fire protection IoT security transmission method provided in an embodiment of this application;

[0031] Figure 3 A schematic flowchart of a model pre-training method provided in an embodiment of this application;

[0032] Figure 4 A flowchart illustrating a fire protection IoT security transmission method provided in another embodiment of this application;

[0033] Figure 5 A flowchart illustrating a fire protection IoT security transmission method provided in another embodiment of this application;

[0034] Figure 6 This is a schematic diagram of the structure of a fire protection IoT safety transmission device provided in an embodiment of this application;

[0035] Figure 7 A structural schematic diagram of a fire-fighting Internet of Things safety transmission device provided for another embodiment of the present application is shown in the figure.

[0036] Figure 8 A structural schematic diagram of an electronic device provided for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0037] In order to make the purposes, advantages and features of the present application clearer, the present application will be described clearly and completely below in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only parts related to the present application are shown in the drawings, but not all structures.

[0038] In order to facilitate understanding of the technical solutions of the present application, first, the related concepts involved in the present application are introduced.

[0039] Server Cluster refers to a group of servers that are loosely or tightly connected together to work, and is usually used to perform large-scale jobs. Clustered servers usually have higher computing efficiency than single servers with comparable speed or availability. By using multiple servers for parallel computation and multiple computing resources to solve problems, the computing speed and processing speed of the cluster system can be improved, and the overall performance of the cluster system can be improved. Each server in the cluster is connected to each other through a network, and each server runs its own operating system.

[0040] Containers can make different application programs run in a relatively isolated and secure environment to achieve isolation between application programs and isolation between application programs and the outside environment. Containers are a kind of lightweight virtualization technology, which has the advantages of fast startup, easy deployment and migration, good security and scalability, and is widely used in AI and cloud computing fields.

[0041] Distributed tasks refer to a large computing task that is decomposed into multiple subtasks, which are distributed to each computing node (e.g., an edge node that can perform computing tasks) in a server cluster. Each computing node independently executes its assigned subtask, and each computing node can exchange data and work together through a network. Typical distributed tasks include distributed training of AI models. The training data set can be divided into multiple batches, and each batch of data set is trained in parallel by each computing node, thereby improving the training efficiency.

[0042] Next, the prior art and problems existing in the prior art related to the present application are described:

[0043] With the wide application of Internet of Things technology in the field of fire protection, the fire protection Internet of Things system can collect a large amount of fire protection data such as temperature, smoke concentration, flame image and the like in real time by deploying a wide range of front-end sensing devices such as temperature sensors, smoke concentration sensors and high-definition cameras.

[0044] However, in the process of implementing the present application, the applicant found that at least the following technical problems exist in the existing fire protection Internet of Things data transmission process:

[0045] Problem one, the raw data collected by the front-end sensing device is large in data volume, and directly uploading to the cloud for processing will not only occupy a large amount of network bandwidth resources, but also cause high data transmission delay, affecting the timeliness of fire warning. Problem two, there is a risk of sensitive data leakage in the process of uploading data to the cloud, threatening user privacy and fire safety information security. Problem three, the reliability of the traditional transmission node architecture is low, and once the key transmission node fails, it may cause the data transmission of the entire fire protection Internet of Things system to be interrupted, and the fire information cannot be transmitted in time, delaying the opportunity for fire rescue.

[0046] In order to solve at least part of the above technical problems, the present application provides a fire protection Internet of Things safe transmission method, a fire protection Internet of Things safe transmission device corresponding to the method, an electronic device capable of implementing the fire protection Internet of Things safe transmission method, a computer readable storage medium and a computer program product. The following embodiments are provided to explain the above method, device, electronic device and computer readable storage medium in detail.

[0047] The application scenario of the fire protection Internet of Things safe transmission method provided by an embodiment of the present application will be introduced below in combination with Figure 1 , the application scenario of the fire protection Internet of Things safe transmission method provided by an embodiment of the present application will be introduced below in combination with Figure 1 , the application scenario of the fire protection Internet of Things safe transmission method provided by an embodiment of the present application will be introduced below in combination with

[0048] As shown in Figure 1 , the application scenario at least includes a fire protection data management system, which includes a cloud server 110 and edge nodes 120 (for example: edge node 1, edge node 2, …, edge node n, n is a natural number greater than or equal to 1), the cloud server 110 and the edge nodes 120 are connected, and an AI development platform is deployed on the cloud server 110. The AI development platform refers to a comprehensive platform that provides a series of tools, libraries, frameworks and services, which can help developers design, train, deploy and manage AI models more quickly and efficiently.

[0049] The edge node 120 refers to a node responsible for performing sub-tasks assigned by the cloud server 110 in a distributed system. These sub-tasks are executed in parallel on different edge nodes 120, and the entire distributed task (such as the task of training the distributed collaborative model in the embodiments of the present application) is completed through mutual cooperation.

[0050] The edge node 120 can be different types of servers, such as CPU servers, GPU servers, neural network processing unit (NPU) servers, service clusters, etc., based on the different functions implemented by different types of AI models. Among them, the CPU server has strong versatility and is suitable for deep learning fields such as pattern recognition, target detection, and image processing; the GPU server can be used to specifically process image computing tasks and can accelerate graphics rendering; the NPU server can be used to accelerate the training and inference calculation of artificial neural networks and can efficiently perform large-scale neural network calculations.

[0051] In one embodiment, the edge node 120 can be an NPU server, which can include multiple NPU acceleration cards, such as NPU1 to NPU8. The cloud server 110 refers to a node responsible for coordinating and managing the execution of the entire distributed task in a distributed system, and can schedule the task to an edge node that meets the resource configuration information when the user 100 (such as an algorithm engineer) issues a model training task according to the resource configuration information.

[0052] Figure 2 The flowchart of the fire-fighting Internet of Things safety transmission method provided by an embodiment of the present application is shown in the figure. The method is applied to a fire-fighting data management system, which includes a cloud server and multiple edge nodes. The method can be executed by a fire-fighting Internet of Things safety transmission device or an electronic device, which can be realized by software and / or hardware. The device or electronic device can be integrated into any intelligent device with network communication function. As shown in the figure, the fire-fighting Internet of Things safety transmission method can include the following steps: Figure 2

[0053] S201, the edge node acquires multi-modal local fire-fighting data and performs data preprocessing on the local fire-fighting data to obtain target fire-fighting data. The local fire-fighting data is fire-fighting data collected by a local sensing device.

[0054] Among them, the multi-modal local fire-fighting data includes at least one of the following: time series signal type, image / video type, and event trigger type.

[0055] ​In this application, the edge node in the fire data management system refers to an intelligent processing unit with local computing, storage and communication capabilities deployed near the fire sensing device or the field terminal side, which serves as an intermediate layer between the front-end perception layer and the cloud server, and undertakes key functions such as data collection, local processing, preliminary analysis and safe transmission. Among them, the local sensing device (such as temperature sensor, smoke concentration sensor, high-definition camera) can collect a large amount of fire data (such as temperature, smoke concentration, flame image information) in real time. The edge node also stores certain sensitive data, including: user privacy data: building personnel distribution, household information, enterprise production raw material storage location (chemical industry park scene); fire safety sensitive data: fire safety equipment operating status (such as fire hydrant water pressure, fire extinguisher expiration date), sensor real-time monitoring data (temperature, smoke concentration, gas composition), historical fire hazard records.

[0056] The edge node side is deployed with local sensing devices of different data types, quantities and sampling frequencies, and the collected multi-modal local fire data includes time series signal data (such as temperature, humidity, smoke concentration, gas concentration, etc.), image / video data (such as visible light camera, infrared thermal imaging, flame recognition image, etc.) and event triggered signal (such as manual alarm button state, water pressure switch signal, electrical fire monitoring signal, etc.), which has significant modal heterogeneity.

[0057] In this step, the edge node is used to obtain original fire sensing data from the local sensing device in communication connection therewith as local fire data. Subsequently, the edge node performs data preprocessing on the local fire data. The data preprocessing operation includes but is not limited to one or more of data cleaning, noise filtering, format standardization, feature extraction or data compression, to eliminate redundant or invalid information and retain key features related to fire risk identification, thereby generating optimized target fire data.

[0058] Optionally, when the local fire data includes image data, the preprocessing method for the image data can be: identifying target image data containing target image in the local fire data based on an image recognition model, and performing image preprocessing on the target image data to obtain data preprocessed target image data, the image preprocessing including at least one of the following: reducing noise, enhancing image information.

[0059] ​In this step, when the local fire data contains image data, the edge node is configured with a pre-installed image recognition model to intelligently analyze the image data to identify target objects related to fire risk (such as flames, smoke, abnormally high temperature areas, or specific fire facility states) therein, thereby extracting image regions containing the target objects as target image data. The image recognition model can include but is not limited to a convolutional neural network (CNN), a lightweight target detection model (such as YOLO-Lite, MobileNet-SSD), or a rule-based image feature matching algorithm, which is deployed locally in the edge node to achieve low-latency and efficient target recognition. Subsequently, the edge node performs image preprocessing operations on the extracted target image data, which at least includes any of the following: noise reduction processing on the image to suppress unstructured interference information introduced by changes in illumination, sensor noise, or transmission interference. And, enhancement processing is performed on the image to improve the recognizability of key visual features, including but not limited to contrast stretching, histogram equalization, edge sharpening, or local brightness adjustment. Through the above image preprocessing, the quality and discriminability of the target image data can be effectively improved, while reducing data redundancy in subsequent transmission and analysis.

[0060] S202, the edge node performs feature extraction on the target fire data in a preset feature dimension based on a pre-trained machine learning model, to obtain a data feature vector, the preset feature dimension including at least one of a fire source area, a high temperature area, and a fire spread area, and the pre-trained machine learning model being obtained by the cloud server and the plurality of edge nodes through multi-modal federated learning based on parameter aggregation rules for phased collaborative training.

[0061] In the embodiments of the present application, the pre-trained machine learning model used by the edge node not only has general fire discrimination ability, but is also specially optimized for fine perception of fire situation. Specifically, the model is configured to perform structured feature extraction on the target fire data in a preset feature dimension, which includes but is not limited to: a fire source area (i.e. the most likely initial position of the fire), a high temperature area (representing the current heat release intensity and the dangerous core area), and a fire spread area (reflecting the direction of fire expansion and the potential risk diffusion path). These dimensions correspond to key decision elements in fire emergency response, making the generated data feature vector have both semantic interpretability and task orientation.

[0062] To achieve the above-mentioned accurate extraction of preset feature dimensions, the pre-trained machine learning model adopts a multi-modal input fusion architecture. For example, when the target fire data contains infrared images, the model locates high-temperature connected domains through a lightweight convolution branch and outputs their spatial coordinates and areas. When the input is time series data of a distributed temperature sensor array, the model uses a time series modeling module (such as LSTM or Temporal ConvNet) to identify temperature surge points and gradient propagation directions, and infer the fire source and spread trend. If there is a visible light video at the same time, the visual flame features and thermal distribution are fused through an attention mechanism to improve the robustness of fire source positioning. Finally, the outputs of each modal branch are mapped to a unified feature space to form a data feature vector containing structured information such as fire source (x, y), high-temperature region mask, and spread speed vector.

[0063] It is particularly critical that the pre-training process of the model uses a cloud-edge collaborative mechanism based on multi-modal federated learning to address the challenge of highly heterogeneous edge node sensor configurations in actual deployment. Different edge nodes may only have partial modal data (such as only temperature sensing without images, or only smoke sensing without gas sensors), and traditional centralized training cannot cover such scenarios. In this application, each edge node independently performs forward inference and backward update based on local real fire data (including historical fire alarms, drills, or normal operating data), and only uploads encrypted model parameter increments or semantic embeddings to the cloud server. The cloud server aggregates multi-source heterogeneous knowledge according to the parameter aggregation rule, such as generating global optimization model parameters through distillation or weighted average strategy, and then distributes them to each edge node to complete model parameter synchronization.

[0064] As mentioned above, the phased collaborative training refers to dividing the pre-training process into multiple successive training phases, each phase setting specific training objectives, input modal ranges, and model optimization strategies: focusing on single-modal basic fire situation discrimination ability construction in the early stage, introducing multi-modal fusion to enhance the robustness of complex scene recognition in the middle stage, and fine-tuning high-order fire situation semantics (such as fire source positioning and spread prediction) in the later stage. In each phase, the cloud server fuses the encrypted model updates uploaded by the edge nodes according to the preset parameter aggregation rule (such as modal weighted average, task-oriented aggregation), and dynamically adjusts the communication frequency, security strategy, and model structure, thereby gradually evolving a fire-specific machine learning model with generalization, precision, and explainability under the premise of ensuring privacy and resource efficiency.

[0065] In the present application, the cooperative training mechanism not only ensures that sensitive data such as original images, positions, and sensor sequences are always kept local, meeting compliance requirements for public safety information, but also significantly improves the model's generalization recognition ability for complex fire pattern (such as lithium battery smoldering accompanied by local high temperature, underground space smoke backflow leading to nonlinear spread) through cross-regional and multi-scenario knowledge fusion. Especially in high-risk tasks such as fire source positioning and spread prediction, the model absorbs cooperative experience from diverse environments such as shopping malls, factories, and tunnels, showing much better robustness and accuracy than single-point training models.

[0066] The present application explicitly embeds fire physical semantic dimensions (fire source, high temperature area, spread area) into feature extraction targets, and combines multi-modal federated learning driven cooperative pre-training architecture, so that edge nodes can still output high-value, structured fire feature vectors under resource limited conditions, providing accurate, real-time, and safe data support for subsequent cloud fire grade determination, emergency resource scheduling, and evacuation path planning, realizing the intelligent transition from "with or without fire" to "where the fire is, how to develop, and how to respond".

[0067] S203, the edge node encrypts the data feature vector, and sends the encrypted data feature vector obtained after encryption to the cloud server.

[0068] In this step, after generating the data feature vector, the edge node calls the built-in security encryption module, such as symmetric encryption algorithm (such as AES, SM4) or asymmetric encryption algorithm (such as RSA, SM2), to encrypt the extracted data feature vector and generate an encrypted data feature vector. Subsequently, the edge node transmits the encrypted data feature vector to the cloud server through a secure communication protocol (such as TLS, DTLS or IPSec), which ensures the confidentiality, integrity and anti-eavesdropping ability of the encrypted data feature vector during transmission, prevents sensitive fire information from being leaked or maliciously tampered with, thereby significantly reducing the transmission bandwidth occupation of original image data under the premise of ensuring the safety of the fire data management system, and improving the response speed and security of the entire fire data management system, thereby supporting the centralized analysis and cooperative decision-making of the cloud server on fire risk.

[0069] S204, the cloud server receives the encrypted data feature vectors sent by each edge node.

[0070] In this step, the cloud server can receive the encrypted data feature vectors uploaded by multiple edge nodes through a secure communication interface.

[0071] S205, the cloud server decrypts each encrypted data feature vector, and determines the fire situation of the fire according to the pre-trained machine learning model and each decrypted data feature vector obtained after decryption. The encrypted data feature vector and the decrypted data feature vector correspond one-to-one.

[0072] In this step, after receiving the encrypted data feature vectors uploaded by each edge node, the cloud server performs corresponding decryption processing on the encrypted data feature vectors based on the key information previously negotiated or configured with each edge node to restore the original data feature vectors. Each encrypted data feature vector and the decrypted data feature vector obtained after decryption form a one-to-one correspondence, ensuring that the data source is traceable and the semantic consistency is maintained. Subsequently, the cloud server inputs multiple decrypted data feature vectors into the pre-trained machine learning model deployed locally, and determines the fire situation of the current fire scene based on the classification or regression results output by the model. The pre-trained machine learning model is trained on a large amount of historical fire data and has the ability to fuse and analyze multi-source features and determine fire.

[0073] The fire safety transmission method proposed in the embodiments of the present application includes the following steps: an edge node obtains multi-modal local fire data and performs data preprocessing on the local fire data to obtain target fire data. The local fire data is fire data collected by a local sensing device. The multi-modal local fire data includes at least one of the following: time series signal, image / video, and event trigger. A pre-trained machine learning model is used to extract features of the target fire data in a preset feature dimension to obtain a data feature vector. The preset feature dimension includes at least one of the following: fire source area, high temperature area, and fire spread area. The pre-trained machine learning model is obtained by collaborative training of a cloud server and multiple edge nodes based on multi-modal federated learning according to a parameter aggregation rule. The edge node encrypts the data feature vector and sends the encrypted data feature vector to the cloud server. The cloud server receives the encrypted data feature vectors sent by each edge node. The cloud server decrypts each encrypted data feature vector, and determines the fire situation of the fire according to the pre-trained machine learning model and each decrypted data feature vector obtained after decryption. The encrypted data feature vector and the decrypted data feature vector correspond one-to-one.

[0074] That is, in the technical solution of the present application, the original fire-fighting data collected by the local sensing device is pre-processed locally at the edge node side to effectively filter out redundant, noisy and invalid information, and generate target fire-fighting data that is concise and has fire discrimination value, thereby significantly reducing the amount of data for subsequent transmission. Further, the edge node uses a pre-trained machine learning model trained by the cloud server and multiple edge nodes to extract features from the target fire-fighting data, obtaining a data feature vector with high representation ability. This collaborative training mechanism not only guarantees the generalization ability of the model to fire-fighting features in multiple scenarios and regions, but also avoids cross-domain transmission of original sensitive data, improving the adaptability and privacy security of the model. On this basis, the edge node encrypts the data feature vector and uploads it to the cloud server to ensure the confidentiality and integrity of the feature data during transmission. After receiving and decrypting each encrypted data feature vector, the cloud server performs fusion analysis based on the one-to-one decrypted data feature vectors and its locally deployed pre-trained machine learning model to accurately determine the fire situation. Therefore, compared with the prior art, the fire safety transmission method proposed in the embodiments of the present application not only significantly reduces network bandwidth occupation and end-to-end response delay, improves the real-time and accuracy of fire warning, but also effectively avoids user privacy leakage and data security risks through encryption mechanism and collaborative model training, thereby constructing an efficient, low-latency and strong privacy protection intelligent fire data processing and fire situation evaluation system while ensuring the safety and reliability of the fire data management system. Moreover, the technical solution of the embodiments of the present application is simple and convenient to implement, easy to popularize, and has a wider application range.

[0075] Another embodiment of the present application also provides a model training method. Based on the above technical solution, further optimization and expansion can be combined with the above various optional embodiments, Figure 3 The flowchart of the model pre-training method provided by an embodiment of the present application is shown in FIG. 1. Figure 3 As shown in the figure, the model pre-training method includes the following steps:

[0076] S311, the edge node acquires local fire data, and determines training sample data based on the local fire data.

[0077] S312, the edge node performs model training on the initial model based on the training sample data to obtain a machine learning model with updated model parameters.

[0078] S313, the edge node encrypts the model update parameters, and sends the encrypted model update parameters to the cloud server.

[0079] S314, the cloud server receives the model update encryption parameters sent by each edge node, and sequentially decrypts, aggregates and encrypts the model update encryption parameters sent by each edge node to obtain encrypted global model update parameters. The model update parameter refers to the model parameter updated by this training.

[0080] S315, the cloud server sends the global model update parameters to each edge node.

[0081] S316, the edge node receives the global model update parameters sent by the cloud server, decrypts the global model update parameters, and updates the model parameters of the initial model based on the decrypted global model update parameters.

[0082] S317, the edge node judges whether the machine learning model reaches the preset convergence condition, if yes, S318 is executed; if not, S311-S317 are repeatedly executed.

[0083] S318, the pre-training is ended, and the pre-trained machine learning model is obtained.

[0084] Next, S311-S318 are described in detail.

[0085] In this step, the initial model refers to the machine learning model before the pre-trained machine learning model is pre-trained. Optionally, the untrained initial model in the edge node is the machine learning model received by the edge node from the cloud server. The untrained initial machine learning model is uniformly sent by the cloud server to each edge node, which ensures the consistency of the model structure and initial parameters of all edge nodes in the fire data management system, and lays a standardized foundation for subsequent collaborative training. At the same time, this method avoids the convergence deviation or unstable training problem caused by independent initialization of each edge node, improves the efficiency of the federated learning process and the generalization ability of the global model, and further enhances the collaborative intelligence level and response consistency of the entire fire data management system.

[0086] In the present embodiment, the machine learning model pre-training method adopts an iterative federated learning mechanism based on an edge-cloud collaborative architecture. By repeatedly executing the training loop comprising S311 to S316, the initial model is optimized for multiple rounds until it meets the preset convergence condition, such as the loss function threshold, accuracy indicator, or maximum number of iterations, thereby obtaining the pre-trained machine learning model. Specifically, in each round of training, each edge node first obtains the original local fire data from the locally deployed fire sensing device and constructs the local training sample data suitable for model training based on the data. Subsequently, the edge node performs local forward propagation and back propagation calculations on the current version of the initial model using the training sample data to generate model parameter updates and obtain the locally updated machine learning model. To protect data privacy and communication security during model training, the edge node does not upload the original data but only encrypts the model update parameters, forms model update encrypted parameters, and sends them to the cloud server. After receiving the model update encrypted parameters from multiple edge nodes, the cloud server performs decryption operations in sequence to restore the model update parameters of each node, then uses an aggregation algorithm (such as weighted average, FedAvg, etc.) to fuse all decrypted parameters to generate global model update parameters representing global knowledge, and encrypts the global parameters again to form encrypted global model update parameters. Subsequently, the cloud server broadcasts the encrypted global model update parameters to each participating edge node. Each edge node receives and decrypts the global model update parameters and synchronously updates the parameters of the local initial model based on the decrypted global model update parameters, completing the current round of collaborative training. The above process is repeated until the model performance meets the preset convergence condition, and finally outputs a pre-trained machine learning model with high generalization ability.

[0087] Among them, the core idea of federated learning (Federated Learning, FL for short) is that the calculation process of model training is completed on the edge node where the data is located, and only the model parameter updates (such as gradients, weight increments) after training are exchanged between nodes or with the cloud, and finally aggregated to form a globally optimal model.

[0088] The model training method provided by the embodiments of the present application adopts a federated learning mode of "local fire data not out of domain, only transmission of model update encryption parameters" in the model pre-training stage, effectively avoids the centralized upload of original fire sensitive data, and significantly improves the user privacy and system information security. At the same time, through the decryption-aggregation-re-encryption processing of the cloud server on the multi-edge node encryption parameters and the synchronous distribution of the global model parameters, the knowledge fusion of cross-regional and multi-scene fire data is realized, so that the obtained pre-training model has both local adaptability and global generalization ability. Further, this collaborative training mode is naturally adapted to the fire deployment environment with limited edge computing resources, and can continuously optimize the model performance without relying on large-scale centralized labeled data sets. Therefore, when the obtained pre-training machine learning model is used for feature extraction on the edge node side and fire disaster judgment on the cloud server side in the subsequent stage, it not only has higher accuracy and robustness, but also ensures the end-to-end credibility of the entire fire data management system due to the embedded security encryption mechanism in the training process, thereby significantly improving the intelligent level, response timeliness and system scalability of the fire warning under the premise of ensuring data security.

[0089] Further, to overcome the dependence of federated learning on input isomorphism, the present application also provides another possible implementation mode for collaborative pre-training of a machine learning model, including the following steps:

[0090] S321, each edge node collects original local fire data based on a locally deployed heterogeneous fire sensing device, and constructs an adaptive local student model according to the sensing configuration thereof; the local fire data includes at least one of a time series signal type, an image / video type and an event trigger type, and the structure of the local student model of each edge node is adaptively determined according to available modalities.

[0091] S322, each edge node performs forward inference on the local student model using the local fire data, generates soft logits or normalized embedding vectors representing the semantics of the current fire as knowledge carriers, and performs encryption processing on the knowledge carriers to obtain encrypted knowledge data.

[0092] S323, each edge node uploads the encrypted knowledge data to the cloud server; the cloud server receives and decrypts the encrypted knowledge data to obtain the soft labels or embedding vectors of each edge node, and compares them with the output of the global teacher model on the multi-modal synthetic input to construct a distillation loss function.

[0093] S324, the cloud server updates the parameters of the global teacher model based on the distillation loss function, and generates corresponding distillation guidance signals according to the differences in modalities of each edge node, and the distillation guidance signals are encrypted and then distributed to the corresponding edge nodes.

[0094] S325: Each edge node receives and decrypts the distillation guidance signal from the cloud server, uses the signal as a supervision target, and fine-tunes the parameters of the local student model in a self-supervised distillation manner, so that the output of the local student model is aligned with the global teacher model in the unified fire semantic space.

[0095] S326: Repeat S322 to S325 until the global teacher model meets the preset convergence condition, to obtain a pre-trained machine learning model, wherein the preset convergence condition includes at least one of a fire identification accuracy threshold, a validation loss stability, and / or a maximum iteration round.

[0096] Next, S321-S326 are described in detail.

[0097] In the cooperative pre-training method provided by the present application, in the embodiments of the present application, each edge node first collects corresponding original local fire data according to the type of the sensing device actually deployed by the edge node (for example, a low-cost node containing only temperature sensing / smoke sensing, or a high-performance node equipped with an infrared camera and a gas sensor). The data naturally has characteristics such as multi-modal, non-aligned, and non-independent and identically distributed (Non-IID). To adapt to such heterogeneous input, each edge node dynamically configures a local student model that matches its sensing capability, for example, a node with only time series input uses LSTM or 1D-CNN structure, and a node with image input integrates a lightweight visual backbone network (such as MobileNet or EfficientNet-Lite). This avoids “one-size-fits-all” model deployment and ensures efficient use of local computing resources.

[0098] In the present application, each edge node no longer uploads original data or complete model parameters, but performs a forward inference to generate a high semantic abstract knowledge carrier. The carrier can be soft logits after temperature scaling (i.e., classification probability distribution without argmax), or a normalized embedding vector output by an intermediate layer of the model, which has a fixed dimension (such as 128 dimensions) and can effectively represent the semantic features of the current fire situation (such as “incipient smoldering” and “flaming spread”). Subsequently, the knowledge carrier is encrypted (such as AES-256 or SM4) by a preset key to form encrypted knowledge data, significantly reducing the risk of data leakage, and compressing the communication load to the order of KB, which is much lower than the traditional parameter upload method.

[0099] Subsequently, the cloud server as a collaborative training hub receives and decrypts the encrypted knowledge data from all participating edge nodes. Since the nodes upload semantic-aligned soft labels or embeddings, the cloud server does not need to care about the original modality differences. The cloud server compares the local knowledge with the output of the global teacher model (e.g., a Transformer-based multi-modal fusion large model) maintained by itself under the synthesized multi-modal input, and constructs a distillation loss function centered on KL divergence. This loss function not only measures the prediction consistency, but also can introduce dynamic weights (such as based on node historical accuracy or regional risk level) to preferentially learn the knowledge of high-credibility nodes.

[0100] After the cloud server updates the global teacher model based on the distillation loss, instead of issuing the complete model, it generates lightweight distillation guidance signals, such as target soft labels, intermediate layer feature maps, or gradient direction prompts, and differentiates encryption according to node capabilities. After receiving and decrypting, each edge node takes the difference between the current output of its local student model and the guidance signal as a supervision signal to perform local fine-tuning (such as single-round SGD update). This process realizes the dual optimization of "global semantic guidance and local personalized adaptation", which not only guarantees the overall cognitive consistency of the system, but also retains the flexibility of the edge side.

[0101] Finally, the above collaborative training cycle continues to iterate until the model meets the preset convergence condition. The condition can be a composite criterion: for example, the fire identification accuracy on the global validation set is greater than or equal to 96%, the loss reduction amplitude is less than 0.5% for 3 consecutive rounds, and the local accuracy of all high-risk area edge nodes meets the standard. Once converged, the resulting model can be used for subsequent real-time fire monitoring.

[0102] Thus, under the premise of ensuring that the original fire data does not leave the domain, the application effectively solves the problems of data misalignment and model heterogeneity caused by differences in sensor configuration, significantly reduces communication overhead, improves the recognition robustness and generalization ability of the model for complex fire conditions (such as smoldering accompanied by gas leakage or local high temperature), and supports personalized deployment and efficient inference of edge nodes, providing a model training technical solution with privacy security, communication efficiency, and intelligent collaboration capability for high-reliability, large-scale intelligent fire data management systems.

[0103] An optional implementation, the plurality of edge nodes includes at least one core edge node, the core edge node includes at least three same data function modules and a majority voting module, and the data function module includes the functions of data receiving, processing and sending. Figure 4 The flowchart of the fire safety transmission method provided by another embodiment of the application is shown in Figure 2. As shown in Figure 2, the fire safety transmission method provided by the embodiment of the application further includes the following steps: Figure 4

[0104] ​S401. For each core edge node, send the data feature vectors output by each data function module to a majority voting module, which compares the data feature vectors extracted by each data function module for the target fire-fighting data.

[0105] S402. Determine whether there are identical data feature vectors in the plurality of data feature vectors. If yes, perform S403. If no, perform S404-S405.

[0106] S403. When there are identical data feature vectors in the plurality of data feature vectors, determine the data feature vector with the largest number of identical data feature vectors as the target data feature vector that the core edge node finally encrypts and sends to the cloud server.

[0107] S404. When there are no identical data feature vectors in the plurality of data feature vectors, trigger monitoring of the running state of at least three identical data function modules.

[0108] The running state includes at least one of the following: CPU usage, memory occupancy ratio, data processing delay.

[0109] S405. When a fault is detected in the target data function module, isolate the target data function module to enable data transmission tasks to be completed by data function modules other than the target data function module.

[0110] The above S401-S405 will be described in detail below.

[0111] In the embodiments of the present application, at least one core edge node is provided in the plurality of edge nodes, and the core edge node internally integrates at least three data function modules with the same structure and function and a majority voting module. Each data function module has complete data receiving, local processing (including preprocessing and feature extraction) and encryption sending functions, and can independently perform feature extraction processing on the same target fire-fighting data and output the corresponding data feature vector. On this basis, the embodiments of the present application further introduce a redundancy fault-tolerant mechanism, which specifically includes the following steps:

[0112] In S401, each core edge node inputs the data feature vectors output by each data function module in the core edge node into a majority voting module. The majority voting module compares the consistency of the data feature vectors generated in parallel. If it is determined in S402 that there are two or more identical data feature vectors (i.e., a majority consensus is reached), S403 is performed, and the data feature vector with the highest occurrence frequency is determined as the target data feature vector of the core edge node for encryption and uploading to the cloud server, thereby improving the reliability and anti-interference ability of the result by using redundant calculation. On the contrary, if it is determined in S402 that all data feature vectors are different from each other (i.e., a majority consensus is not reached), the fault diagnosis process is entered: in S404, real-time monitoring of the running state of each data function module is triggered, including but not limited to CPU usage, memory occupancy ratio, data processing delay, and other health indicators. Then in S405, if it is identified by monitoring that a data function module (referred to as a "target data function module") has performance abnormalities or suspected faults, the module is immediately isolated logically or physically, so that it exits the current data processing flow, and the remaining normally operating data function modules continue to complete the subsequent data feature extraction and data transmission tasks of transmitting data to the cloud server, ensuring continuous availability.

[0113] The fire-fighting Internet of Things safety transmission method provided by the embodiment of the application improves the accuracy and robustness of feature extraction by deploying at least three isomorphic data function modules in the core edge node and introducing a majority voting mechanism. Further, when all module outputs are inconsistent, the fire-fighting data management system automatically activates the running state monitoring mechanism and executes the fault module isolation strategy, realizing the closed-loop fault tolerance capability from "error tolerance" to "fault self-healing". Thus, not only is the single-point failure risk effectively resisted, ensuring the high reliability and continuity of the fire-fighting key data in the edge processing and uploading process, but also the problem of service interruption of the entire edge node caused by local module failure is avoided. In combination with the aforementioned encryption transmission and collaborative training mechanism, the embodiment of the application further strengthens the edge resilience and industrial-level availability of the fire-fighting data management system while ensuring data security and model intelligence, and is particularly suitable for high-risk fire-fighting scenarios such as high-rise buildings, underground spaces, and chemical parks, which have high reliability requirements.

[0114] In an optional implementation, the fire-fighting fire situation can include multiple fire-fighting fire situation levels, such as general, larger, major, and particularly major fire-fighting fire situation levels, for representing the development trend, influence range, and emergency response level of the fire event. Figure 5 The flowchart of the fire-fighting Internet of Things safety transmission method provided by another embodiment of the application is shown in FIG. 6. As shown in FIG. 6, the method includes the following steps. Figure 5As shown, in order to dynamically adapt to the data sensing and transmission requirements under different fire levels, the fire-fighting Internet of Things safety transmission method provided by the embodiment further includes the following steps:

[0115] S501, when detecting that the fire level is upgraded, the cloud server controls the first number of edge nodes to start performing the data transmission task according to the current fire level.

[0116] In this step, when the cloud server detects that the current fire level is upgraded, for example, from "general" to "larger", or from "major" to "particularly major", the cloud server will automatically trigger the resource scheduling strategy according to the current fire level after the upgrade, and control and activate the first number of additional edge nodes to join the fire data collection and transmission task. The value of the first number can be a preset value or dynamically calculated based on the fire level, for example, the higher the fire level, the more new edge nodes activated to enhance the sensing density, data coverage and information update frequency of the surrounding area of the fire scene. The activated edge nodes can include nodes in a dormant, low-power or standby state, which immediately start the data collection, preprocessing, feature extraction and encrypted upload process of the local sensing device after receiving the scheduling instruction, thereby forming an elastic data transmission network matched with the severity of the fire.

[0117] S502, when detecting that the fire level is downgraded, the cloud server controls the second number of edge nodes to stop performing the data transmission task according to the current fire level, and performs data archiving and / or model parameter updating on the second number of edge nodes.

[0118] In this step, when the cloud server detects that the current fire level is downgraded, for example, from "major" to "larger", or from "larger" to "general", in order to optimize resource utilization efficiency and reduce energy consumption, the embodiment can dynamically control the second number of edge nodes to stop performing the real-time collection and transmission of fire data according to the current fire level after the downgrade. The value of the second number can be determined dynamically according to the preset strategy or based on the current fire level and the load state of the edge node, to ensure that the scale of edge nodes participating in data transmission is reasonably reduced on the premise of ensuring necessary monitoring capability.

[0119] Optionally, the plurality of edge nodes includes at least one core edge node, the core edge node includes at least three same data function modules and a majority voting module, and the data function module includes the functions of data receiving, processing and sending. Then, one optional implementation of "controlling the second number of edge nodes to stop performing the data transmission task according to the current fire level" can be: controlling the second number of edge nodes other than the core edge node to stop performing the data transmission task according to the current fire level.

[0120] In this step, the continuous operation state of the core edge node is preferentially retained, and only the non-core edge node is task decommissioned. Specifically, when the fire-fighting fire level is detected to be degraded and the data transmission resource needs to be reduced, the cloud server determines the number of edge nodes (i.e., the second number) to be disabled according to the current fire-fighting fire level, and selects a corresponding number of nodes from the ordinary edge nodes other than the core edge node to control the nodes to stop performing the collection, processing and uploading tasks of the fire-fighting data. This can ensure that the high-reliability and high-confidence data channel provided by the core edge node is still maintained in the fire relief stage, and the continuous monitoring capability of the key areas or critical facilities is guaranteed, while the overall energy consumption and network load are effectively reduced by closing part of the non-core nodes, realizing the fine and safety consideration of resource scheduling.

[0121] Figure 6 The structural diagram of the fire-fighting Internet of Things safety transmission device provided by an embodiment of the present application is applied to an edge node in a fire-fighting data management system, the fire-fighting data management system includes a cloud server and a plurality of edge nodes, and the edge node is used to perform a data transmission task of transmitting data to the cloud server. As shown in Figure 6 The fire-fighting Internet of Things safety transmission device includes an acquisition module 601, a feature extraction module 602 and an encryption module 603; wherein,

[0122] The acquisition module 601 is used to acquire local fire-fighting data of multiple modalities and perform data preprocessing on the local fire-fighting data to obtain target fire-fighting data, the local fire-fighting data is fire-fighting data collected by a local sensing device, and the local fire-fighting data of multiple modalities includes at least one of the following: time series signal type, image / video type and event trigger type;

[0123] The feature extraction module 602 is used to perform feature extraction on the target fire-fighting data in a preset feature dimension based on a pre-trained machine learning model to obtain a data feature vector; the preset feature dimension includes at least one of the following: a fire source area, a high-temperature area and a fire spread area, and the pre-trained machine learning model is obtained by the cloud server and a plurality of edge nodes through collaborative training in stages based on multi-modal federated learning according to a parameter aggregation rule;

[0124] The encryption module 603 is used to perform encryption processing on the data feature vector and send the encrypted data feature vector obtained after the encryption processing to the cloud server, so that the cloud server performs decryption processing on the encrypted data feature vector sent by each edge node, and monitors the fire-fighting fire based on the decrypted data feature vector obtained after the decryption and the local pre-trained machine learning model.

[0125] In an optional implementation, the initial model is a machine learning model before the pre-trained machine learning model is pre-trained. The apparatus further includes a model training module, which is specifically configured to:

[0126] Obtain local fire data, and determine training sample data based on the local fire data;

[0127] Perform model training on the initial model based on the training sample data to obtain a machine learning model with updated model parameters;

[0128] Encrypt the model update parameters, and send the encrypted model update parameters to the cloud server, so that the cloud server sequentially performs decryption, aggregation and encryption on the model update parameters sent by each edge node to obtain encrypted global model update parameters, wherein the model update parameters refer to the updated model parameters obtained in this training;

[0129] Receive the global model update parameters sent by the cloud server, decrypt the global model update parameters, and update the model parameters of the initial model based on the decrypted global model update parameters;

[0130] Determine whether the machine learning model meets a preset convergence condition, if yes, pre-training is ended to obtain the pre-trained machine learning model, and if not, repeat the above steps of obtaining local fire data, determining training sample data, performing model training, obtaining model update parameters and sending them to the cloud server, receiving the global model update parameters sent by the cloud server and updating the model parameters based on the global model update parameters.

[0131] In an optional implementation, the plurality of edge nodes includes at least one core edge node, the core edge node includes at least three same data function modules and a majority voting module, the data function module includes the functions of data receiving, processing and sending; the apparatus further includes a verification module, which is specifically configured to:

[0132] For each core edge node, send the data feature vectors output by each data function module to the majority voting module, and compare the data feature vectors extracted by each data function module for the target fire data through the majority voting module;

[0133] When there are same data feature vectors in the plurality of data feature vectors, the majority voting module determines the data feature vectors with the largest number of same data feature vectors as the target data feature vectors to be finally encrypted and sent to the cloud server by the core edge node.

[0134] An optional implementation, the check module is further configured to:

[0135] When there is no same data feature vector in the plurality of data feature vectors, triggering monitoring of the running state of the at least three same data function modules, the running state comprising at least one of the following: CPU usage, memory occupation ratio, data processing delay;

[0136] If it is monitored that the target data function module has a fault, isolating the target data function module, so as to complete the data transmission task according to the data function module other than the target data function module.

[0137] The fire-fighting Internet of Things safety transmission device can execute the edge node side technical solution provided by any one of the method embodiments of the first aspect of the application, and has the corresponding function modules and beneficial effects of the execution method. Technical details not described in detail in this embodiment can be referred to the fire-fighting Internet of Things safety transmission method provided by any one of the method embodiments of the first aspect of the application.

[0138] Figure 7 The structure diagram of the fire-fighting Internet of Things safety transmission device provided by another embodiment of the application is applied to a cloud server in a fire-fighting data management system, and the fire-fighting data management system comprises the cloud server and a plurality of edge nodes. As shown in Figure 7 The fire-fighting Internet of Things safety transmission device comprises a receiving module 701 and a determining module 702; wherein,

[0139] The receiving module 701 is configured to receive encrypted data feature vectors sent by each edge node, wherein the encrypted data feature vectors are data obtained by an edge node acquiring local fire-fighting data and sequentially performing data preprocessing, feature extraction and encryption processing on the local fire-fighting data;

[0140] The determining module 702 is configured to perform decryption processing on each encrypted data feature vector, and determine a fire disaster according to each decrypted data feature vector obtained after decryption and a locally pre-trained machine learning model, wherein the encrypted data feature vector and the decrypted data feature vector correspond one by one.

[0141] An optional implementation, the fire disaster comprises a plurality of fire disaster grades; the device further comprises a control module, and the control module is specifically configured to:

[0142] When detecting that the fire disaster grade is upgraded, controlling the first number of edge nodes to start performing a data transmission task according to the current fire disaster grade; or,

[0143] When detecting the downgrade of the fire disaster level, the second quantity of edge nodes are controlled to stop performing the data transmission task according to the current fire disaster level, and the data archiving and / or model parameter updating processing is performed on the second quantity of edge nodes.

[0144] An optional implementation, the plurality of edge nodes includes at least one core edge node, the core edge node includes at least three identical data function modules and majority voting module, the data function module includes the function of data receiving, processing and sending; the control module is further used for:

[0145] According to the current fire disaster level, a second quantity of edge nodes other than the core edge node are controlled to stop performing the data transmission task.

[0146] The above-mentioned fire Internet of Things safety transmission device can execute the cloud server side technical solution provided by any one of the method embodiments of the second aspect of the application, and has the corresponding function modules and beneficial effects of the execution method. Technical details not described in detail in the embodiment can be referred to the fire Internet of Things safety transmission method provided by any one of the method embodiments of the second aspect of the application.

[0147] The embodiment of the application further provides a computer program product.

[0148] The various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuitry, integrated circuitry, field programmable gate arrays, application specific integrated circuits, application specific standard products, systems on chips, load programmable logic devices, computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer program products that can include one or more computer programs tangibly embodied in a non-transitory machine readable storage medium that can be executed by, and / or loadable on, a programmable system including at least one programmable processor that can be special or general purpose programmable processor, and that can receive data and instructions as input from storage, at least one input device, and / or at least one output device.

[0149] Figure 8 A structural schematic diagram of an electronic device is provided for an embodiment of the application. The electronic device in the embodiment of the application can be a cloud server or an edge node. Referring to Figure 8 , Figure 8 The electronic device 12 shown is merely an example and should not impose any limitation on the functions and use range of the embodiments of the application. As Figure 8As shown, the electronic device 12 is represented in the form of a general-purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0150] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. Examples of these architectures include, but are not limited to, industry-standard architecture buses, microchannel architecture buses, enhanced industry-standard architecture buses, Video Electronics Standards Association (VESA) local buses, and peripheral component interconnect buses.

[0151] Electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0152] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory 30 and / or cache memory 32. Electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 8 Not shown; usually referred to as a "hard drive"). Although Figure 8 As not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. System memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this application.

[0153] A program / utility 40 having a set (at least one) of program modules 46 may be stored, for example, in system memory 28. Such program modules 46 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 46 typically perform the functions and / or methods described in the embodiments of this application.

[0154] The electronic device 12 can also communicate with one or more external devices 14 such as a keyboard, a pointing device, a display 24, etc.; other devices that enable a user to interact with the electronic device 12; and / or any devices (e.g., a network card, a modem, a Figure 8 Other hardware and / or software modules that can be used in conjunction with the electronic device 12 but are not shown in FIG. 1 include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0155] The processing unit 16 performs various function applications and data processing by running programs stored in the system memory 28, such as implementing the fire-fighting Internet of Things safety transmission method provided by any of the method embodiments of the first aspect of the present application: acquiring multi-modal local fire-fighting data and pre-processing the local fire-fighting data to obtain target fire-fighting data, the local fire-fighting data being fire-fighting data collected by a local sensing device, the multi-modal local fire-fighting data including at least one of the following: time series signal type, image / video type, and event trigger type; performing feature extraction on the target fire-fighting data in a preset feature dimension based on a pre-trained machine learning model to obtain a data feature vector, the preset feature dimension including at least one of the following: a fire source area, a high-temperature area, and a fire spread area, and the pre-trained machine learning model being obtained by the cloud server and multiple edge nodes through collaborative training in stages based on multi-modal federated learning according to a parameter aggregation rule; performing encryption processing on the data feature vector, and sending the encrypted data feature vector obtained after the encryption processing to the cloud server, so that the cloud server performs decryption processing on the encrypted data feature vectors sent by each edge node, and monitors a fire situation based on the decrypted data feature vectors obtained after the decryption and the local pre-trained machine learning model.

[0156] For example, the fire-fighting Internet of Things safety transmission method provided by any one of the embodiments of the second aspect of the application is implemented as follows: encrypted data feature vectors sent by each edge node are received, the encrypted data feature vectors being data obtained after the edge node acquires local fire-fighting data and sequentially performs data preprocessing, feature extraction, and encryption processing on the local fire-fighting data; each encrypted data feature vector is decrypted, and a fire disaster is determined according to each decrypted data feature vector obtained after decryption and a locally pre-trained machine learning model, the encrypted data feature vector and the decrypted data feature vector corresponding to each other.

[0157] The embodiments of the application provide a computer-readable storage medium, which stores a computer program, and the program is executed by a processor to implement the fire-fighting Internet of Things safety transmission method provided by any one of the embodiments of the first aspect of the application: local fire-fighting data of multiple modalities is acquired and data preprocessing is performed on the local fire-fighting data to obtain target fire-fighting data, the local fire-fighting data being fire-fighting data collected by a local sensing device, and the local fire-fighting data of multiple modalities including at least one of the following: a time series signal type, an image / video type, and an event trigger type; a data feature vector is obtained by performing feature extraction on the target fire-fighting data in a preset feature dimension based on a pre-trained machine learning model, the preset feature dimension including at least one of the following: a fire source area, a high-temperature area, and a fire disaster spreading area, and the pre-trained machine learning model being obtained by the cloud server and multiple edge nodes through collaborative training in stages based on multi-modal federated learning according to a parameter aggregation rule; an encrypted data feature vector is obtained by performing encryption processing on the data feature vector, and the encrypted data feature vector is sent to the cloud server, so that the cloud server decrypts encrypted data feature vectors sent by each edge node, and monitors a fire disaster based on decrypted data feature vectors obtained after decryption and a locally pre-trained machine learning model.

[0158] Alternatively, the fire-fighting Internet of Things safety transmission method provided by any of the method embodiments of the second aspect of the application is implemented: encrypted data feature vectors sent by each edge node are received, the encrypted data feature vectors being data obtained after the edge node acquires local fire-fighting data and sequentially performs data preprocessing, feature extraction, and encryption processing on the local fire-fighting data; each encrypted data feature vector is decrypted, and a fire disaster is determined according to each decrypted data feature vector obtained after decryption and a locally pre-trained machine learning model, the encrypted data feature vector and the decrypted data feature vector corresponding one-to-one. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor electronic device, device, or component, or any combination thereof. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more conductive wires, a portable computer diskette, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory (erasable programmable read-only memory or flash memory), an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution electronic device, device, or component.

[0159] The computer-readable signal medium can include a data signal propagated in a baseband or as a part of a carrier wave, in which a computer-readable program code is carried. Such a propagated data signal can take multiple forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution electronic device, device, or component.

[0160] The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, radio frequency, etc., or any suitable combination thereof.

[0161] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network ("LAN") or a wide area network ("WAN"), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0162] It is to be understood that the above description is merely a preferred embodiment of the application and the applied technical principles. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, modifications and substitutions can be made without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A fire protection IoT security transmission method, characterized in that, An edge node is applied in a fire data management system, which includes a cloud server and multiple edge nodes. The edge nodes perform data transmission tasks to transmit data to the cloud server. The multiple edge nodes include at least one core edge node, which includes at least three identical data function modules and a majority voting module. The data function modules include data receiving, processing, and sending functions. The method includes: Acquire multimodal local fire protection data and perform data preprocessing on the local fire protection data to obtain target fire protection data. The local fire protection data is fire protection data collected by local sensing devices. The multimodal local fire protection data includes at least one of the following: time-series signal type, image / video type, and event-triggered type. Based on a pre-trained machine learning model, features are extracted from the target fire data in a preset feature dimension to obtain a data feature vector. The preset feature dimension includes at least one of the following: fire source area, high temperature area, and fire spread area. The pre-trained machine learning model is obtained by the cloud server and multiple edge nodes through phased collaborative training based on multimodal federated learning and parameter aggregation rules. The data feature vector is encrypted, and the encrypted data feature vector is sent to the cloud server so that the cloud server can decrypt the encrypted data feature vector sent by each edge node and monitor the fire situation based on the decrypted data feature vector and the local pre-trained machine learning model. For each of the core edge nodes, the data feature vectors output by each of the data function modules are sent to the majority voting module, and the majority voting module compares the data feature vectors extracted by each of the data function modules for the target fire data. When multiple data feature vectors contain the same data feature vector, the majority voting module determines the data feature vector with the largest number of identical data feature vectors as the target data feature vector that the core edge node will ultimately encrypt and send to the cloud server.

2. The method according to claim 1, characterized in that, The initial model is the machine learning model before the pre-trained machine learning model was pre-trained. The steps for pre-training the initial model include: Acquire local fire protection data and determine training sample data based on the local fire protection data; The initial model is trained based on the training sample data to obtain a machine learning model with updated model parameters. The model update parameters are encrypted, and the encrypted model update parameters are sent to the cloud server. The cloud server then decrypts, aggregates, and encrypts the encrypted model update parameters sent by each edge node in sequence to obtain the encrypted global model update parameters. The model update parameters refer to the updated model parameters obtained in this training. Receive the global model update parameters sent from the cloud server, decrypt the global model update parameters, and update the model parameters of the initial model based on the decrypted global model update parameters; Determine whether the machine learning model has reached the preset convergence condition. If yes, the pre-training ends and the pre-trained machine learning model is obtained. If no, repeat the above steps of obtaining local fire data, determining training sample data, training the model, obtaining model update parameters and sending them to the cloud server, receiving global model update parameters sent by the cloud server, and updating the model parameters based on the global model update parameters.

3. The method according to claim 1, characterized in that, The method further includes: When there are no identical data feature vectors among the multiple data feature vectors, the monitoring of the running status of the at least three identical data function modules is triggered. The running status includes at least one of the following: CPU utilization, memory usage ratio, and data processing latency. If a fault is detected in the target data function module, the target data function module is isolated so that the data transmission task can be completed by the data function modules other than the target data function module.

4. A fire protection IoT security transmission method, characterized in that, A cloud server is used in a fire data management system, the fire data management system including the cloud server and multiple edge nodes, and the method includes: The edge node receives encrypted data feature vectors sent by each edge node. The encrypted data feature vectors are data obtained by the edge node after acquiring local fire data and performing data preprocessing, feature extraction and encryption processing on the local fire data in sequence. The encrypted data feature vectors are decrypted, and the fire situation is determined based on the decrypted data feature vectors and the local pre-trained machine learning model. The encrypted data feature vectors correspond one-to-one with the decrypted data feature vectors.

5. The method according to claim 4, characterized in that, The fire situation includes multiple fire severity levels; the method further includes: When an escalation of the fire situation is detected, a first number of edge nodes are added to the fire suppression system to begin data transmission tasks, based on the current fire situation level; or... When the fire situation level is downgraded, the second number of edge nodes are controlled to stop performing data transmission tasks according to the current fire situation level, and data archiving and / or model parameter update processing is performed on the second number of edge nodes.

6. The method according to claim 5, characterized in that, The plurality of edge nodes includes at least one core edge node, and the core edge node includes at least three identical data function modules and a majority voting module. The data function modules include data receiving, processing, and sending functions. The step of controlling a second number of edge nodes to stop performing data transmission tasks according to the current fire situation level also includes: Based on the current fire situation level, control a second number of edge nodes (excluding the core edge node) to stop performing data transmission tasks.

7. A fire protection IoT safety transmission device, characterized in that, An edge node is used in a fire data management system, which includes a cloud server and multiple edge nodes. The edge nodes perform data transmission tasks to transmit data to the cloud server. The multiple edge nodes include at least one core edge node, which includes at least three identical data function modules and a majority voting module. The data function modules include data receiving, processing, and sending functions. The device includes: The acquisition module is used to acquire multimodal local fire protection data and perform data preprocessing on the local fire protection data to obtain target fire protection data. The local fire protection data is fire protection data collected by local sensing devices. The multimodal local fire protection data includes at least one of the following: time-series signal type, image / video type, and event-triggered type. The feature extraction module is used to extract features from the target fire data on a preset feature dimension based on a pre-trained machine learning model to obtain a data feature vector. The preset feature dimension includes at least one of the following: fire source area, high temperature area and fire spread area. The pre-trained machine learning model is obtained by the cloud server and multiple edge nodes through phased collaborative training based on multimodal federated learning and parameter aggregation rules. An encryption module is used to encrypt the data feature vectors and send the encrypted data feature vectors to the cloud server. The cloud server then decrypts the encrypted data feature vectors sent by each edge node and monitors fire conditions based on the decrypted data feature vectors and a locally pre-trained machine learning model. For each core edge node, the data feature vectors output by each data function module are sent to the majority voting module. The majority voting module compares the data feature vectors extracted by each data function module for the target fire data. When multiple data feature vectors contain the same data feature vector, the majority voting module determines the data feature vector with the most identical data feature vectors as the target data feature vector that the core edge node will ultimately encrypt and send to the cloud server.

8. A fire protection IoT safety transmission device, characterized in that, A cloud server used in a fire data management system, the fire data management system including the cloud server and multiple edge nodes, the device comprising: The receiving module is used to receive encrypted data feature vectors sent by each of the edge nodes. The encrypted data feature vectors are data obtained by the edge nodes after acquiring local fire protection data and performing data preprocessing, feature extraction and encryption processing on the local fire protection data in sequence. The determination module is used to decrypt each of the encrypted data feature vectors and determine the fire situation based on the decrypted data feature vectors and the local pre-trained machine learning model. The encrypted data feature vectors correspond one-to-one with the decrypted data feature vectors.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the fire protection IoT security transmission method as described in any one of claims 1 to 3, or implement the fire protection IoT security transmission method as described in any one of claims 4 to 6.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the fire protection IoT security transmission method as described in any one of claims 1 to 3, or implements the fire protection IoT security transmission method as described in any one of claims 4 to 6.

Images