An intelligent fire positioning method and system based on a large model

Abstract

The embodiment of the present disclosure relates to a kind of intelligent fire locating method and system based on large model, it is related to intelligent positioning field, the present disclosure compared with previous fire monitoring system, it solves the problem that the false alarm rate and the missing report rate of existing fire monitoring system are higher, leading to the fire risk in gymnasium and other venues;The present disclosure integrates multiple sensor devices, forms all-round, multi-level monitoring network;Using advanced data fusion algorithm to integrate heterogeneous data;Based on the improved target detection network, using distributed sensor network and Internet of Things technology, real-time collection of multi-dimensional information of target venue, combined with advanced positioning algorithm, realize the rapid three-dimensional positioning of fire source;The present disclosure greatly improves the accuracy and response speed of fire positioning. At the same time, through adaptive learning ability and modular design ensure long-term stable operation and future upgrade convenience, provide comprehensive, efficient and intelligent solution for the fire safety management of target venue.

Description

Technical Field

[0001] The present disclosure relates to the field of intelligent positioning technology, and in particular to an intelligent fire positioning method and system based on a large model. Background Art

[0002] With the rapid development of urbanization and the construction of large-scale public facilities, fire safety management in gymnasiums, large public venues that integrate sports competitions, theatrical performances, and mass gatherings, has become particularly important. However, due to the unique architectural characteristics of gymnasiums—extensive space, dense crowds, and complex layouts—fires can easily spread rapidly once they occur, posing significant challenges to evacuation and fire rescue.

[0003] Existing fire detection systems typically rely on traditional smoke detectors and temperature sensors. These systems may not be sensitive or responsive enough in large spaces such as stadiums and often lack intelligent analysis capabilities, making it difficult to accurately identify and pinpoint the fire source in the first place, delaying the optimal rescue opportunity. Summary of the Invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a large model-based intelligent fire location method and system.

[0005] An embodiment of the present disclosure provides an intelligent fire location method based on a large model. Environmental monitoring sensors are deployed at various key locations in a target venue to form a distributed sensor network. The environmental monitoring sensors include multiple types of sensors. The method includes: a data acquisition module collects sensor data from each of the environmental monitoring sensors according to preset acquisition rules and sends it to an Internet of Things gateway. The Internet of Things gateway cleans and formats all the sensor data to obtain sensor data to be processed and sends it to a central processing unit; the central processing unit sends the sensor data to be processed to a data processing and analysis server, inputs the sensor data to be processed into a pre-built target detection network to obtain an output result, constructs a state equation at a target time based on the output result, and determines an observation equation based on the state equation and an extended Kalman filter algorithm; inputs the sensor data to be processed into a pre-built artificial neural network to obtain a position estimate and a velocity estimate, fuses the position estimate and the velocity estimate with the output of the observation equation to obtain an observation vector, and performs a prediction based on the observation vector and the extended Kalman filter algorithm to obtain a fire source location; obtains fire confirmation information based on the fire source location, and controls an alarm system to activate the speaker and lighting system of the target venue to issue an alarm.

[0006] Optionally, the sensor data is sent to the Internet of Things gateway via a low-power communication module equipped on the sensor node using a wireless signal.

[0007] Optionally, the method further includes: dividing the distributed sensor network into multiple functional modules; and adjusting configuration information of the multiple functional modules according to scale information of the target venue.

[0008] Optionally, the method also includes: the Internet of Things gateway obtains the transmission information of the sensor node and adjusts the data transmission path based on the transmission information; or, the Internet of Things gateway and the sensor node are authenticated based on a preset encryption authentication method, and data transmission is performed after the Internet of Things gateway and the sensor node are authenticated.

[0009] Optionally, the pre-constructed target detection network includes a neck module, a feature conversion module and a multi-scale feature fusion module; the neck module includes a convolution module of a mixed convolution layer and a mixed convolution layer, and an output convolution layer; wherein the mixed convolution layer includes depth-separable convolution and standard convolution; the feature conversion module includes a splitting layer, a feature extraction and enhancement module, a splicing layer and a channel compression convolution layer; wherein, the input to the feature conversion module is split into two features through the splitting layer; wherein one feature is input to n series-connected feature extraction and enhancement modules, and the output of the last feature extraction and enhancement module is spliced ​​with the other feature of the split layer through the splicing layer, and the splicing result is output after channel compression processing by the channel compression convolution layer; wherein n is a positive integer greater than 0; the multi-scale feature fusion module determines the weight information corresponding to different scale features according to the target task, and performs feature fusion on the different scale features based on the weight information corresponding to each scale feature.

[0010] Optionally, the step of inputting the sensor data to be processed into a pre-built artificial neural network to obtain a position estimate and a speed estimate includes: performing feature extraction on the sensor data to be processed based on a convolutional neural network to obtain a first feature vector; performing feature extraction on the sensor data to be processed input in a time series using a recurrent neural network to obtain a second feature vector; fusing the first feature vector and the second feature vector using feature splicing to obtain a fused feature vector, and converting the fused feature vector into the position estimate and the speed estimate through a fully connected layer.

[0011] Optionally, the state equation has a corresponding process noise covariance matrix and state transfer matrix, and the observation equation has a corresponding observation noise covariance matrix and observation matrix. The method also includes: adjusting the process noise covariance matrix and / or the observation noise covariance matrix based on the environmental information and / or sensor information of the target venue; adjusting the state transfer matrix and / or the observation matrix based on the difference information between the fire source location and the actual fire source information.

[0012] The disclosed embodiment also provides an intelligent fire location system based on a large model, wherein environmental monitoring sensors are deployed at various key locations in the target venue to form a distributed sensor network, wherein the environmental monitoring sensors include multiple types of sensors, and the system includes: a data acquisition module, for collecting sensor data of each sensor in the environmental monitoring sensors according to preset acquisition rules and sending the data to an Internet of Things gateway; the Internet of Things gateway, for performing data cleaning and formatting on all the sensor data, obtaining sensor data to be processed and sending it to a central processing unit; the central processing unit, for sending the sensor data to be processed to a data processing and analysis server; the data processing and analysis server, for sending the sensor data to be processed to a central processing unit; The sensor data is input into a pre-built target detection network to obtain an output result, and a state equation at the target moment is constructed based on the output result, and an observation equation is determined according to the state equation and the extended Kalman filter algorithm; the data processing and analysis server is also used to input the sensor data to be processed into a pre-built artificial neural network to obtain a position estimate and a speed estimate, and fuse the position estimate and the speed estimate with the output of the observation equation to obtain an observation vector, and make a prediction based on the observation vector based on the extended Kalman filter algorithm to obtain the fire source location; the alarm module is used to obtain fire confirmation information based on the fire source location, and control the alarm system to activate the loudspeaker and lighting system of the target venue to issue an alarm.

[0013] An embodiment of the present disclosure also provides an electronic device, comprising: a processor; a memory for storing instructions executable by the processor; the processor for reading the executable instructions from the memory and executing the instructions to implement the large model-based intelligent fire location method provided in the embodiment of the present disclosure.

[0014] The embodiment of the present disclosure further provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the large model-based intelligent fire location method provided in the embodiment of the present disclosure.

[0015] The embodiments of the present disclosure also provide a computer program product, including a computer program, wherein the computer program is executed by a processor as the large model-based intelligent fire location method provided in the embodiments of the present application.

[0016] The technical solution provided by the embodiments of the present disclosure has the following advantages over the prior art: the intelligent fire location solution based on a large model provided by the embodiments of the present disclosure deploys environmental monitoring sensors at various key locations in a target venue to form a distributed sensor network, wherein the environmental monitoring sensors include multiple types of sensors. The method includes: a data acquisition module collects sensor data from each of the environmental monitoring sensors according to preset acquisition rules and sends it to an Internet of Things gateway; the Internet of Things gateway cleans and formats all sensor data to obtain sensor data to be processed and sends it to a central processing unit; the central processing unit sends the sensor data to be processed to a data processing and analysis server, inputs the sensor data to be processed into a pre-built target detection network to obtain an output result, constructs a state equation at the target time based on the output result, and determines an observation equation based on the state equation and an extended Kalman filter algorithm; inputs the sensor data to be processed into a pre-built artificial neural network to obtain a position estimate and a velocity estimate, fuses the position estimate and the velocity estimate with the output of the observation equation to obtain an observation vector, and performs a prediction based on the observation vector based on the extended Kalman filter algorithm to obtain the fire source location; obtains fire confirmation information based on the fire source location, and controls the alarm system to activate the loudspeaker and lighting system of the target venue to sound an alarm. In this way, the problem that the existing fire monitoring system has high false alarm and missed alarm rates, which leads to fire hazards in the target venue, is solved. By integrating multiple sensor devices, an all-round and multi-level monitoring network is formed, and advanced data fusion algorithms are used to integrate these heterogeneous data into a unified platform; based on the improved target detection network, it is specifically aimed at the identification of fire and smoke; using distributed sensor networks and Internet of Things technology, multi-dimensional information of the fire scene is collected in real time, and combined with advanced positioning algorithms, rapid three-dimensional positioning of the fire source is achieved; the present invention greatly improves the accuracy and response speed of fire positioning in the target venue; at the same time, the adaptive learning ability and modular design ensure long-term stable operation and convenience of future upgrades, providing a comprehensive, efficient and intelligent solution for fire safety management of the target venue. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the originals and elements are not necessarily drawn to scale.

[0018] Figure 1A schematic diagram of a flow chart of a large model-based intelligent fire location method provided in an embodiment of the present disclosure;

[0019] Figure 2 A schematic diagram of the neck module structure provided in an embodiment of the present disclosure;

[0020] Figure 3 A schematic diagram of the structure of a feature conversion module provided in an embodiment of the present disclosure;

[0021] Figure 4 A schematic diagram of the structure of a multi-scale feature fusion module provided in an embodiment of the present disclosure;

[0022] Figure 5 A schematic structural diagram of another large-model-based intelligent fire location system provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0023] The following describes embodiments of the present disclosure in more detail with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0024] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0025] As used herein, the term "including" and its variations are open-ended, i.e., "including but not limited to." The term "based on" means "based, at least in part, on." The term "one embodiment" means "at least one embodiment," the term "another embodiment" means "at least one additional embodiment," and the term "some embodiments" means "at least some embodiments." Other terms are defined in the following description.

[0026] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0027] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0028] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0029] While various fire detectors and automatic alarm systems are widely used in existing technologies, they still have significant limitations, particularly in complex environments like large stadiums. Existing fire monitoring systems often suffer from high false alarm and missed alarm rates. In dynamic environments like stadiums, due to dramatic temperature and humidity fluctuations, the presence of non-fire-related dust and smoke (such as stage effects, heat generated by dense crowds, and increased carbon dioxide concentrations), traditional point sensors are prone to false alarms, leading to unnecessary emergency evacuations or wasted firefighting resources. Furthermore, complex structures and obstructions, such as seats, pillars, and billboards, can prevent sensors from effectively covering the entire area, increasing the risk of missed alarms.

[0030] With the continuous advancement of science and technology, cutting-edge technologies such as big data, the Internet of Things (IoT), and artificial intelligence (AI) have brought innovative solutions to stadium fire management. Among them, intelligent fire location technology based on large models is particularly eye-catching. Large models specifically refer to machine learning models with large parameter counts and complex structures, such as deep neural networks (DNNs). These models can demonstrate extraordinary pattern recognition and prediction capabilities after being trained on large amounts of data. The application of this technology enables stadiums to instantly identify and accurately locate the source of a fire in its initial stages, buying valuable time for timely response measures.

[0031] The purpose of the present disclosure is to provide an intelligent fire location method and system based on a large model, which can monitor the environment in the target venue in real time, quickly and accurately locate the fire source, and issue an alarm in time. Specifically, the present disclosure integrates a variety of sensor devices to form an all-round, multi-level monitoring network; uses advanced data fusion algorithms to integrate these heterogeneous data into a unified platform; based on an improved target detection network, it is specifically aimed at the identification of fire and smoke; uses distributed sensor networks and Internet of Things technology to collect multi-dimensional information of the fire scene in real time, combined with advanced positioning algorithms, to achieve rapid three-dimensional positioning of the fire source; the present disclosure greatly improves the accuracy and response speed of fire location in the target venue; at the same time, it also includes adaptive learning capabilities and modular design to ensure long-term stable operation and convenience of future upgrades, providing a comprehensive, efficient and intelligent solution for fire safety management of the target venue. The following is combined with Figure 1 Provide a detailed description.

[0032] Specifically, Figure 1 This is a flow chart of a large model-based intelligent fire location method provided by an embodiment of the present disclosure, wherein the method can be implemented using software and / or hardware and can generally be integrated into electronic devices. Figure 1 As shown, the method includes:

[0033] Step 101: The data acquisition module collects sensor data from each sensor in the environmental monitoring sensor according to the preset acquisition rules and sends it to the Internet of Things gateway. The Internet of Things gateway cleans and formats all sensor data, obtains sensor data to be processed, and sends it to the central processing unit.

[0034] In the disclosed embodiments, a distributed sensor network is constructed by deploying environmental monitoring sensors at key locations within a target venue. The environmental monitoring sensors include a variety of sensor types. Specifically, the target venue typically refers to a large public venue such as a gymnasium. Key locations may include entrances, exits, auditoriums, and competition areas. Environmental monitoring sensors can integrate multiple sensor types, including smoke detectors, temperature and humidity sensors, gas sensors, millimeter-wave radar, and video surveillance, to form a comprehensive, multi-layered monitoring network capable of detecting subtle changes in the early stages of a fire, such as increases in smoke concentration or abnormally high temperatures.

[0035] That is to say, multiple types of sensor equipment are integrated at each key location to achieve all-round monitoring. For example, at the entrance, smoke detectors may be deployed to monitor smoke concentration, temperature and humidity sensors may be deployed to monitor ambient temperature and humidity, and millimeter-wave radars may be deployed to detect human bodies when smoke obscures the area. Multiple types of sensors work together to further improve monitoring accuracy.

[0036] In the embodiment of the present disclosure, the data acquisition module is used to interface with the sensor, collect raw data and perform preliminary processing (such as format conversion) to make it data that can be recognized by the system.

[0037] Specifically, the data acquisition module collects sensor data from each sensor in the environmental monitoring sensor according to the preset acquisition rules and sends it to the Internet of Things gateway; among them, the preset acquisition rules can drive the sensor to collect data by timed acquisition or waking up according to preset rules, which can ensure the real-time and accuracy of the data.

[0038] In some embodiments, the sensor data is sent to the IoT gateway via wireless signals through a low-power communication module equipped with the sensor node; wherein, the IoT gateway is used to receive data from the data acquisition module, perform protocol conversion (unifying different sensor protocols), data cleaning (noise removal), formatting and other operations, and then transmit it to the central processing unit, while maintaining network stability and security.

[0039] Specifically, the IoT gateway is responsible for receiving data from all sensors, performing necessary data cleaning and formatting, such as removing noise data, erroneous data, and duplicate data, and unifying data in different formats collected by different sensors into a standardized format, and then transmitting the data to the central processing unit via a wired or wireless network.

[0040] In some embodiments, the IoT gateway obtains transmission information from the sensor node and adjusts the data transmission path based on the transmission information; or, the IoT gateway and the sensor node are authenticated based on a preset encryption authentication method, and data transmission is performed after the IoT gateway and the sensor node are authenticated.

[0041] Specifically, the IoT gateway is responsible for maintaining the stability and security of the sensor network; for example, by dynamically optimizing the network topology and automatically adjusting the data transmission path; when a sensor node or link fails, the IoT gateway can quickly re-plan the route to ensure that data transmission is not blocked and network connectivity is guaranteed; or adopt encryption authentication methods (such as key-based two-way authentication). When a sensor node connects to the IoT gateway, both parties need to verify the legitimacy of the identity to prevent illegal devices from forging nodes to access the network and ensure that only authorized sensors can communicate with the gateway.

[0042] In some embodiments, the distributed sensor network is divided into multiple functional modules, and configuration information of the multiple functional modules is adjusted according to scale information of the target venue.

[0043] In the disclosed embodiments, the target venue's scale information refers to the size of the target venue, which can be determined by parameters such as the target venue's area. Configuration information refers to the number of functional modules configured and the number of each functional module configured. Specifically, the sensor network is designed to be modular and scalable, adapting to target venues of varying sizes. The distributed sensor network is divided into modules with different functional modules (e.g., temperature and humidity detection modules, gas detection modules, etc.), each of which can be independently designed, manufactured, maintained, or replaced, like building blocks, facilitating flexible assembly and management. Based on the target venue's scale (e.g., small target venue versus large, comprehensive target venue), the number of functional modules can be easily increased or decreased, and their configuration adjusted. For example, a small target venue can have a streamlined number of functional modules, while a large target venue can have expanded modules to cover a wider area. This ensures that the sensor network can adapt and operate effectively regardless of the target venue's size, enhancing the system's versatility and practicality.

[0044] It should be noted that the data acquisition module, IoT gateway and central processing unit can be integrated into one device (such as an industrial gateway or edge computing device) to simplify the architecture, reduce transmission delay, and be suitable for small scenarios or scenarios with high integration requirements; they can also be deployed separately, which is more flexible and allows for individual component upgrades as needed (such as replacing a high-performance gateway) and disperses the risk of failure; for example, in large stadiums, where sensors are widely distributed, gateways are deployed close to the sensors to reduce transmission losses, and the central processing unit and data processing and analysis servers are centrally managed.

[0045] Step 102: The central processing unit sends the sensor data to be processed to the data processing and analysis server, so as to input the sensor data to be processed into the pre-built target detection network to obtain the output result, and construct the state equation at the target moment based on the output result, and determine the observation equation according to the state equation and the extended Kalman filter algorithm.

[0046] Step 103: Input the sensor data to be processed into a pre-built artificial neural network to obtain position estimation and velocity estimation, and fuse the position estimation and the velocity estimation with the output of the observation equation to obtain an observation vector, and perform prediction based on the observation vector based on the extended Kalman filter algorithm to obtain the fire source location.

[0047] In the disclosed embodiment, the data processing and analysis server is used to receive data from the central processing unit, run large model analysis and complex algorithms (such as fire sign identification and diffusion path prediction), and realize deep data mining and application.

[0048] In some embodiments, the pre-constructed target detection network includes a neck module, a feature conversion module and a multi-scale feature fusion module; the neck module includes a convolution module of a mixed convolution layer and a mixed convolution layer, and an output convolution layer; wherein the mixed convolution layer includes a depth-separable convolution and a standard convolution; the feature conversion module includes a splitting layer, a feature extraction and enhancement module, a splicing layer and a channel compression convolution layer; wherein, the input to the feature conversion module is split into two features through the splitting layer; wherein, one feature is input to n series-connected feature extraction and enhancement modules, and the output of the last feature extraction and enhancement module is spliced ​​with the other feature of the split layer through the splicing layer, and the splicing result is output after channel compression processing through the channel compression convolution layer; wherein n is a positive integer greater than 0; the multi-scale feature fusion module determines the weight information corresponding to different scale features according to the target task, and performs feature fusion on different scale features based on the weight information corresponding to each scale feature.

[0049] In the embodiment of the present disclosure, the pre-built target detection network is an improved YOLOv8 (You Only Look Once version 8, a deep learning framework for achieving real-time object detection) network, such as Figure 2 As shown in the figure, an improved neck module is introduced to replace the original neck module of the YOLOv8 network. The neck module of the present invention is composed of a convolutional module composed of a convolutional layer and a hybrid convolutional layer, and an output convolutional layer. The hybrid convolutional layer is composed of a depth-wise separable convolution and a standard convolution. Using a hybrid convolutional layer composed of depth-wise separable convolutions for feature information extraction can increase the speed of feature extraction. Further combination with the convolutional layer can improve the ability of feature nonlinear expression and information reuse. Figure 2 As shown in the figure, the input feature (In) passes through two convolutional layers (Conv), the output of one convolutional layer is processed by the mixed convolutional layer and then concatenated with the output of the other convolutional layer through the concatenation layer (Concat), and then processed by the convolutional layer (Conv) to output the feature (Out); Among them, the mixed convolutional layer processes the input feature (In) through two depth-wise separable convolutions and fuses the input feature (In) through the convolutional layer to output the feature (Out).

[0050] Specifically, such as Figure 3 As shown in the figure, an improved feature conversion module (i.e., C2f module) is used to replace the original C2f module. In the improved C2f module, the input to the C2f module through the convolution layer (Conv) is divided into two features through the Split layer. One feature is input to n series-connected Bottleneck (feature extraction and enhancement) modules, and the output of the last Bottleneck module is concatenated with the other feature of the Split layer through the concatenation layer (Concat), and then output after channel compression processing through the convolution layer (Conv). The improved C2f module can enhance the feature extraction of fire targets while reducing the number of parameters and computational complexity, alleviating the problem of decreased accuracy due to channel information diversion. Figure 3 The composition of the Bottleneck module is also shown in the figure. The input features (In) pass through the convolution layer (Conv3*3) and pass through ‌BN (BatchNormalization, batch normalization) and ‌ReLU (Rectified Linear Unit, activation function), and then pass through the convolution layer (Conv3*3) again and output the features (Out) through ‌BN and ‌ReLU.

[0051] Specifically, such as Figure 4As shown in the figure, multi-scale feature fusion is performed on the input features input 1, input 2, input 3, input 4, and input 5. This fusion is performed using a multi-scale feature fusion module. Traditional fusion methods simply add all input features together, while the multi-scale feature fusion module assigns weights based on the importance of different input features. Small-scale feature maps contain more detailed information, facilitating the recognition of small objects; large-scale feature maps have stronger semantic information, benefiting the recognition of large objects and overall scene understanding. The multi-scale feature fusion module assigns weights based on the contribution of each scale feature to the target task (e.g., fire sign recognition). For example, when identifying weak, early signs of a fire, small-scale features may be more critical, so they are assigned a higher weight. However, when determining large-scale scene information, such as the overall spread of a fire, large-scale features are given a higher weight.

[0052] Therefore, the improved YOLOv8 network quickly extracts features of sensor data through hybrid convolutional layers, which further enhance the nonlinear expression and information reuse of features. These features are input into a classifier (such as a fully connected layer) and compared with a predefined fire feature model to determine whether there are signs of fire (such as excessive smoke concentration, abnormally high temperature, and other feature combinations).

[0053] Therefore, the large model analysis unit combines time series sensor data with an improved YOLOv8 network to model fire-related features at different times. By analyzing the changing trends of features over time and combining the predictive capabilities of neural networks, the distribution of fire-related features at the next moment is predicted, and then the direction and range of fire spread are inferred to form a diffusion path prediction. For example, by continuously monitoring changes in smoke concentration distribution, after the network learns the smoke diffusion pattern, it predicts the areas that may be covered by smoke in the future and determines the fire diffusion path.

[0054] As a result, the large-scale model analysis unit, based on a modified YOLOv8 network distributed sensor network and IoT technology, is specifically designed for fire and smoke recognition. It can not only identify early signs of fire but also accurately distinguish between real fires and false positives in complex environments, significantly reducing false alarm rates. Leveraging distributed sensor networks and IoT technology, it can collect multi-dimensional information about the fire scene in real time. Furthermore, combined with advanced positioning algorithms, it can achieve rapid three-dimensional localization of the fire source.

[0055] Specifically, determining the precise location of the fire source based on a positioning algorithm includes the following steps.

[0056] Get the output of the improved YOLOv8 network and construct the system state vector at the target time, such as time k: , so the state equation of the system is: ;in, is a nonlinear function, which represents the state transfer law; is the control input vector; is the process noise, which satisfies the Gaussian distribution, specifically: ,in, is the process noise covariance matrix.

[0057] The observation equation is ;in, is the observation vector; is a nonlinear observation function; is the observation noise, which satisfies the Gaussian distribution, specifically ,in, is the observation noise covariance matrix.

[0058] Further nonlinear function Performs a Taylor expansion linearization on the current estimate.

[0059] For the state equation, Expand to ;in, is the Jacobian matrix of the state transition matrix.

[0060] For the observation equation, Taylor expansion linearization is performed at: ;in, is the original observation function; For The observation function expanded at ; is the Jacobian matrix of the state transition matrix.

[0061] In some embodiments, the sensor data to be processed is input into a pre-built artificial neural network to obtain a position estimate and a speed estimate, including: performing feature extraction on the sensor data to be processed based on a convolutional neural network to obtain a first feature vector; using a recurrent neural network to extract features on the sensor data to be processed input in a time series to obtain a second feature vector; fusing the first feature vector and the second feature vector using feature splicing to obtain a fused feature vector, and converting the fused feature vector into a position estimate and a speed estimate through a fully connected layer.

[0062] Specifically, the sensor data is preprocessed and then spliced ​​to obtain the processed sensor data as the input of the artificial neural network, the convolutional neural network is used to extract features from the processed sensor data, and the recurrent neural network is used to extract features from the processed sensor data input in time series. The feature vectors extracted by the convolutional neural network and the recurrent neural network are fused by feature splicing to obtain a fused feature vector, and the fused feature vector is converted into a position estimate related to positioning through a fully connected layer. and speed estimation .

[0063] The above position estimate and speed estimation and the observation equation Fusion, observation vector The improved version is updated as follows: ;in, is the updated sensor observation value.

[0064] Thus we get the new observation function , ;in, is the transpose of the vector; and is the part of the state vector that is related to position and velocity.

[0065] The observation noise covariance matrix Make the following updates .

[0066] in, is the updated observation noise covariance matrix; and For position estimation and speed estimation The covariance of .

[0067] Further fire path prediction and state prediction are performed based on the fused observation matrix and observation noise covariance matrix ; Covariance prediction ; Calculate Kalman gain Status update ; Covariance update .

[0068] Therefore, the predicted fire source location is obtained based on the positioning algorithm, and the predicted state is corrected according to the current observation value. After continuous iterative calculation, the optimal estimated state is finally obtained, which is the fire source location coordinate.

[0069] In some embodiments, the state equation has a corresponding process noise covariance matrix and state transfer matrix, and the observation equation has a corresponding observation noise covariance matrix and observation matrix. The method also includes: adjusting the process noise covariance matrix and / or observation noise covariance matrix based on the environmental information and / or sensor information of the target venue; adjusting the state transfer matrix and / or observation matrix based on the difference information between the fire source location and the actual fire source information.

[0070] Specifically, the process noise covariance matrix can be adjusted based on the adaptive mechanism. , observation noise covariance matrix , state transfer matrix and the observation matrix Dynamic adjustments can also be made based on environmental changes and the learning of deep learning models. Online learning algorithms can be used to enable the model to better adapt to changes in target motion patterns or changes in sensor characteristics.

[0071] Specifically, in real-world fire scenarios, the environment is dynamically changing (e.g., airflow fluctuations and smoke spread can affect sensor accuracy), and sensor characteristics may also change (e.g., accuracy drift after prolonged use). The adaptive mechanism enables the Kalman filter model to adapt to these changes. By dynamically adjusting the process noise covariance matrix, observation noise covariance matrix, state transition matrix, and observation matrix, the model can more accurately track the fire source status. For example, if a sudden increase in airflow is detected, smoke spread may accelerate, making the fire source location more difficult to predict. In this case, an increase in the process noise covariance matrix indicates increased internal uncertainty in the system. At the same time, smoke interference may further inaccurate sensor observations, increasing the observation noise covariance matrix. Using an online learning algorithm, new sensor data and actual fire source status information (if additional precise monitoring methods are available) are continuously collected. By comparing the differences between the model's predictions and the actual values, matrix parameters are adjusted inversely. For example, if the predicted fire source location deviates increasingly from the actual location, the state transition matrix and observation matrix are adjusted accordingly to ensure that the model predictions are more consistent with the actual situation.

[0072] It is important to note that the fire tracking and location process is also visualized through a high-precision digital twin model of the target venue, providing the emergency command center with an intuitive, real-time map of the fire situation, facilitating rapid decision-making. The system's built-in adaptive learning mechanism continuously optimizes fire identification, location, and evacuation strategies based on historical data and real-time feedback. Using online learning algorithms, the system automatically adjusts model parameters to adapt to changes in the stadium environment across seasons and event sizes, ensuring the long-term effectiveness of fire monitoring and emergency response.

[0073] Step 104: Obtain fire confirmation information based on the location of the fire source, and control the alarm system to activate the loudspeaker and lighting system of the target venue to sound an alarm.

[0074] Specifically, when the precise location of the fire source is analyzed, that is, the location of the fire source meets the preset fire judgment criteria (such as multiple sensor data simultaneously exceed the normal threshold and the model identifies it as a fire), the fire is automatically confirmed. After the fire is confirmed, the alarm system will be activated immediately and an alarm will be issued through the speakers and lighting system in the target venue to notify the personnel in the target venue and the emergency response team.

[0075] As an example, several types of environmental monitoring sensor devices are configured in the target venue, and a distributed sensor network is built based on them; combined with Internet of Things technology, the sensor data at the fire scene is collected in real time through the data acquisition module and transmitted to the central processing unit; based on the large model analysis unit, an improved YOLOv8 network is built to analyze the collected sensor data, identify signs of fire, and predict the spread path of the fire; the positioning module determines the precise location of the fire source (fire source location) based on the results of the large model analysis and the positioning algorithm; when the alarm system detects that the precise location of the fire source meets the preset fire judgment criteria, it automatically confirms that a fire has occurred, and the alarm system immediately activates the speakers and lighting systems in the gymnasium to sound an alarm.

[0076] Among them, the distributed sensor network design can be a modular and scalable network, in which each sensor node is equipped with a low-power communication module for sending the data collected by the data acquisition module to the Internet of Things gateway through wireless signals; the Internet of Things gateway is used to receive data from the sensors collected by the data acquisition module, clean and format the data, and then transmit the data to the central processing unit through a wired or wireless network; during the fire positioning process, the fire process is also visualized through the digital twin model of the target venue; during the fire positioning process, an online learning algorithm is also used to realize dynamic adjustment of model parameters during the positioning process; the large model analysis unit and the positioning module have built-in adaptive learning mechanisms for continuously optimizing fire identification, positioning and evacuation strategies based on historical data and real-time feedback.

[0077] Therefore, by introducing an intelligent analysis system based on a large model, the accuracy and robustness of fire identification can be significantly improved; the large model uses its powerful learning ability to learn normal and abnormal patterns in different situations from historical data, thereby effectively distinguishing real fire signals from other interference factors; combined with multimodal perception technology and the use of advanced deep learning models, the ability to extract fire characteristics can be enhanced, and accurate monitoring can be achieved even in complex environments, reducing the occurrence of false alarms and missed alarms; by integrating data from multiple sensors and monitoring equipment already in the target venue, the intelligent analysis system can analyze the fire situation more comprehensively and achieve more accurate and comprehensive fire location and early warning.

[0078] In summary, applying the present disclosure to fire monitoring in target venues can quickly and accurately determine the location of a fire. In the early stages of a fire, every second counts, and timely positioning helps rationally allocate rescue resources. Based on the fire location information, the appropriate number and type of firefighting equipment and rescue information can be pre-arranged. For example, in crowded venues like gymnasiums, this system can also provide precise guidance for evacuation. By integrating with the target venue's broadcasting system and signage system, spectators and staff can be informed of the fire's location and safe evacuation routes in real time. This helps reduce panic and confusion during the evacuation process, preventing people from becoming trapped due to disorientation.

[0079] In addition, by quickly locating the fire location, measures can be taken in advance to isolate the dangerous area and protect the safety of personnel; accurate fire positioning allows firefighters to directly target the fire source to extinguish the fire, avoiding large-scale water spraying or the use of fire extinguishing agents to damage unaffected areas; for expensive sports facilities, stage equipment, electronic equipment, etc. in the gymnasium, this precise fire extinguishing method can minimize property losses; the intelligent fire positioning system can also be connected to the venue's facility management system to cut off the power supply of related equipment in time when a fire occurs, protecting the electrical system and other key facilities; having an advanced intelligent fire positioning system reflects the target venue's safety management It will enhance the high-tech level and foresight in management; help venue managers better fulfill their safety responsibilities and carry out refined management of venue facilities, event organization, etc.; for example, when hosting large-scale sports events or cultural performances, it will be able to demonstrate the venue's safety assurance capabilities to organizers and audiences, thereby enhancing the venue's competitiveness; by providing a safer environment, the target venue can enhance the public's trust and satisfaction in it; during daily operations and activities, audiences and participants will feel more at ease knowing that the venue has advanced fire warning and positioning capabilities; it will have a positive impact on the long-term development and brand building of the venue, attracting more events and audiences, and increasing economic benefits.

[0080] Figure 5 This is a schematic diagram of the structure of an intelligent fire location system based on a large model provided by an embodiment of the present disclosure. The system can be implemented by software and / or hardware and can generally be integrated into electronic devices. Figure 5 As shown, environmental monitoring sensors are deployed at key locations in the target venue to form a distributed sensor network. The environmental monitoring sensors include multiple types of sensors. The system includes:

[0081] The data acquisition module 501 is used to collect sensor data of each sensor in the environmental monitoring sensor according to a preset acquisition rule and send it to the Internet of Things gateway 502.

[0082] The IoT gateway 502 is configured to perform data cleaning and formatting processing on all the sensor data, obtain sensor data to be processed, and send the processed sensor data to the central processing unit 503 .

[0083] The central processing unit 503 is configured to send the sensor data to be processed to a data processing and analysis server.

[0084] The data processing and analysis server 504 is used to input the sensor data to be processed into a pre-built target detection network to obtain an output result, and to construct a state equation at the target moment based on the output result, and to determine an observation equation based on the state equation and the extended Kalman filter algorithm.

[0085] The data processing and analysis server 504 is also used to input the sensor data to be processed into a pre-built artificial neural network to obtain position estimation and speed estimation, and to fuse the position estimation and speed estimation with the output of the observation equation to obtain an observation vector, and to make a prediction based on the observation vector based on the extended Kalman filter algorithm to obtain the fire source location.

[0086] The alarm module 505 is configured to obtain fire confirmation information based on the fire source location, and control the alarm system to activate the loudspeaker and lighting system of the target venue to sound an alarm.

[0087] Optionally, the sensor data is sent to the Internet of Things gateway via a wireless signal through a low-power communication module equipped in the sensor node.

[0088] Optionally, the system further includes: a processing module, configured to divide the distributed sensor network into a plurality of functional modules; and adjust configuration information of the plurality of functional modules according to scale information of the target venue.

[0089] Optionally, the system also includes: the Internet of Things gateway 502 obtains the transmission information of the sensor node and adjusts the data transmission path based on the transmission information; or, the Internet of Things gateway 502 and the sensor node are authenticated based on a preset encryption authentication method, and data transmission is performed after the Internet of Things gateway 502 and the sensor node are authenticated.

[0090] Optionally, the pre-constructed target detection network includes a neck module, a feature conversion module and a multi-scale feature fusion module; the neck module includes a convolution module of a mixed convolution layer and a mixed convolution layer, and an output convolution layer; wherein the mixed convolution layer includes depth-separable convolution and standard convolution; the feature conversion module includes a splitting layer, a feature extraction and enhancement module, a splicing layer and a channel compression convolution layer; wherein, the input to the feature conversion module is split into two features through the splitting layer; wherein one feature is input to n series-connected feature extraction and enhancement modules, and the output of the last feature extraction and enhancement module is spliced ​​with the other feature of the split layer through the splicing layer, and the splicing result is output after channel compression processing by the channel compression convolution layer; wherein n is a positive integer greater than 0; the multi-scale feature fusion module determines the weight information corresponding to different scale features according to the target task, and performs feature fusion on the different scale features based on the weight information corresponding to each scale feature.

[0091] Optionally, the data processing and analysis server 504 is further used to: perform feature extraction on the sensor data to be processed based on a convolutional neural network to obtain a first feature vector; use a recurrent neural network to perform feature extraction on the sensor data to be processed input in a time series to obtain a second feature vector; fuse the first feature vector and the second feature vector by feature splicing to obtain a fused feature vector, and convert the fused feature vector into the position estimate and the speed estimate through a fully connected layer.

[0092] Optionally, the state equation has a corresponding process noise covariance matrix and state transfer matrix, and the observation equation has a corresponding observation noise covariance matrix and observation matrix. The data processing and analysis server 504 is specifically used to adjust the process noise covariance matrix and / or the observation noise covariance matrix based on the environmental information and / or sensor information of the target venue; and adjust the state transfer matrix and / or the observation matrix based on the difference information between the fire source location and the actual fire source information.

[0093] The large model-based intelligent fire location system provided by the embodiments of the present disclosure can execute the large model-based intelligent fire location method provided by any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0094] The embodiments of the present disclosure further provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the large model-based intelligent fire location method provided by any embodiment of the present disclosure.

[0095] According to one or more embodiments of the present disclosure, the present disclosure provides an electronic device, including:

[0096] processor;

[0097] a memory for storing instructions executable by the processor;

[0098] The processor is configured to read the executable instructions from the memory and execute the instructions to implement any large model-based intelligent fire location method provided in the present disclosure.

[0099] According to one or more embodiments of the present disclosure, the present disclosure provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute any large model-based intelligent fire location method provided by the present disclosure.

[0100] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the present disclosure is not limited to technical solutions formed by specific combinations of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the scope of the above disclosure. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

[0101] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details have been included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0102] Although the subject matter has been described in language specific to structural features and / or methodological logical acts, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely example forms of implementing the claims.

Claims

1. An intelligent fire location method based on a large model, characterized in that: Environmental monitoring sensors are deployed at key locations in the target venue to form a distributed sensor network. The environmental monitoring sensors include multiple types of sensors. The method includes: The data acquisition module collects sensor data from each sensor in the environmental monitoring sensor according to the preset acquisition rules and sends it to the Internet of Things gateway. The Internet of Things gateway cleans and formats all the sensor data, obtains the sensor data to be processed and sends it to the central processing unit; The central processing unit sends the sensor data to be processed to the data processing and analysis server, so as to input the sensor data to be processed into the pre-built target detection network to obtain an output result, and construct the state equation at the target moment based on the output result, and determine the observation equation according to the state equation and the extended Kalman filter algorithm; wherein, the pre-built target detection network includes a neck module, a feature conversion module and a multi-scale feature fusion module; the neck module includes a convolution module of a convolution layer and a hybrid convolution layer, and an output convolution layer; wherein, the hybrid convolution layer includes depth-separable convolution and standard convolution; the feature conversion module includes a split layer, a feature extraction and enhancement module, splicing layer and channel compression convolution layer; wherein, the input to the feature conversion module is split into two features through the splitting layer; wherein, one feature is input to n series-connected feature extraction and enhancement modules, the output of the last feature extraction and enhancement module is spliced ​​with the other feature of the splitting layer through the splicing layer, and the splicing result is output after channel compression processing through the channel compression convolution layer; wherein n is a positive integer greater than 0; the multi-scale feature fusion module determines the weight information corresponding to the different scale features according to the target task, and performs feature fusion on the different scale features based on the weight information corresponding to each scale feature; Inputting the sensor data to be processed into a pre-built artificial neural network to obtain a position estimate and a velocity estimate, fusing the position estimate and the velocity estimate with the output of the observation equation to obtain an observation vector, and performing a prediction based on the observation vector using the extended Kalman filter algorithm to obtain a fire source location; Fire confirmation information is obtained based on the fire source location, and the alarm system is controlled to activate the loudspeaker and lighting system of the target venue to sound an alarm.

2. The method according to claim 1, characterized in that The sensor data is sent to the Internet of Things gateway via a low-power communication module equipped on the sensor node using a wireless signal.

3. The method according to claim 1, characterized in that The method further comprises: Dividing the distributed sensor network into multiple functional modules; The configuration information of the plurality of functional modules is adjusted according to the scale information of the target venue.

4. The method according to claim 2, characterized in that The method further comprises: The Internet of Things gateway obtains the transmission information of the sensor node and adjusts the data transmission path based on the transmission information; or The Internet of Things gateway and the sensor node are authenticated based on a preset encryption authentication method, and the Internet of Things gateway and the sensor node perform data transmission after being authenticated.

5. The method according to claim 1, wherein The step of inputting the to-be-processed sensor data into a pre-built artificial neural network to obtain position estimation and speed estimation comprises: Performing feature extraction on the sensor data to be processed based on a convolutional neural network to obtain a first feature vector; Using a recurrent neural network to extract features from the sensor data to be processed input in time series to obtain a second feature vector; The first feature vector and the second feature vector are fused by feature concatenation to obtain a fused feature vector, and the fused feature vector is converted into the position estimate and the speed estimate through a fully connected layer.

6. The method according to claim 1, characterized in that The state equation has a corresponding process noise covariance matrix and a state transfer matrix, and the observation equation has a corresponding observation noise covariance matrix and an observation matrix. The method further includes: Adjusting the process noise covariance matrix and / or the observation noise covariance matrix based on the environmental information and / or sensor information of the target venue; The state transfer matrix and / or the observation matrix are adjusted based on the difference information between the fire source location and the actual fire source information.

7. An intelligent fire location system based on a large model, characterized in that: Environmental monitoring sensors are deployed at key locations in the target venue to form a distributed sensor network. The environmental monitoring sensors include multiple types of sensors. The system includes: A data acquisition module is used to collect sensor data from each sensor in the environmental monitoring sensor according to a preset acquisition rule and send it to the Internet of Things gateway; The IoT gateway is configured to clean and format all the sensor data, obtain the sensor data to be processed, and send it to the central processing unit; The central processing unit is used to send the sensor data to be processed to a data processing and analysis server; The data processing and analysis server is used to input the sensor data to be processed into a pre-built target detection network to obtain an output result, and to construct a state equation at the target moment based on the output result, and to determine the observation equation according to the state equation and the extended Kalman filter algorithm; wherein, the pre-built target detection network includes a neck module, a feature conversion module and a multi-scale feature fusion module; the neck module includes a convolution module of a convolution layer and a hybrid convolution layer, and an output convolution layer; wherein, the hybrid convolution layer includes a depth-separable convolution and a standard convolution; the feature conversion module includes a splitting layer, a feature extraction and enhancement module, a splicing module, and a multi-scale feature fusion module. A convolution layer and a channel compression convolution layer; wherein the input to the feature conversion module is split into two features through the splitting layer; wherein one feature is input to n series-connected feature extraction and enhancement modules, the output of the last feature extraction and enhancement module is spliced ​​with the other feature of the splitting layer through the splicing layer, and the splicing result is output after channel compression processing through the channel compression convolution layer; wherein n is a positive integer greater than 0; the multi-scale feature fusion module determines the weight information corresponding to the different scale features according to the target task, and performs feature fusion on the different scale features based on the weight information corresponding to each scale feature; The data processing and analysis server is further configured to input the sensor data to be processed into a pre-built artificial neural network to obtain a position estimate and a velocity estimate, fuse the position estimate and the velocity estimate with the output of the observation equation to obtain an observation vector, and perform a prediction based on the observation vector using the extended Kalman filter algorithm to obtain a fire source location; The alarm module is used to obtain fire confirmation information based on the fire source location and control the alarm system to activate the loudspeaker and lighting system of the target venue to issue an alarm.

8. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing instructions executable by the processor; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the large model-based intelligent fire location method described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and the computer program is used to execute the large model-based intelligent fire location method described in any one of claims 1 to 6.

Images