Smart park multi-source fire-fighting data dynamic monitoring and real-time analysis system

Abstract

The invention relates to the technical field of park monitoring, in particular to a smart park multi-source fire-fighting data dynamic monitoring and real-time analysis system, which comprises a risk early warning unit, a rescue request unit, a visual large screen unit, a rescue planning unit, a diffusion analysis unit and a fire-fighting evacuation unit, and further comprises a state monitoring unit, a real-time analysis unit and a fire-fighting evacuation unit, the state monitoring unit is used for performing state monitoring on a plurality of monitoring areas in the target smart park through the sensor group to obtain first fire-fighting data of the monitoring areas, and the first fire-fighting data of the monitoring areas comprises temperature, smoke concentration and gas concentration of the monitoring areas. Through the combination of the state monitoring unit and the threshold value judgment unit, the system can obtain and analyze fire-fighting data in a park in real time, and timely identify a potential fire risk area, so that the system can quickly respond to a possible fire event, and the fire response speed and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of park monitoring technology, and in particular to a smart park multi-source fire data dynamic monitoring and real-time analysis system. Background Art

[0002] The Textile Smart Park is a modern park that combines advanced information technology and the needs of the textile industry. Its core goal is to improve the production efficiency, management level and operational benefits of the textile industry through intelligent means.

[0003] Traditional systems may rely on manual inspections and manual reporting, which can lead to delays in fire alarms and responses. Traditional systems may lack advanced analytical tools and models, resulting in low accuracy in fire warnings and prone to false alarms or missed alarms. Traditional systems often lack real-time data and intelligent analysis, resulting in inefficient scheduling and allocation of rescue resources, which may lead to prolonged response times. Traditional systems may not have effective diffusion analysis tools and cannot accurately predict the spread of smoke, thereby affecting the formulation of evacuation and rescue plans. Traditional systems may lack modern visualization tools, making it difficult for campus managers and emergency response teams to quickly understand the fire situation and locate problem areas. Traditional systems may not be able to integrate and analyze data from multiple sensors in real time, resulting in an inaccurate overall assessment of fire risks. Summary of the Invention

[0004] The purpose of the present invention is to address the problems existing in the background technology and propose a smart park multi-source fire data dynamic monitoring and real-time analysis system.

[0005] The technical solution of the present invention is a smart park multi-source fire data dynamic monitoring and real-time analysis system, which includes a risk warning unit, a rescue request unit, a visualization large screen unit, a rescue planning unit, a diffusion analysis unit, and a fire evacuation unit, and also includes: a status monitoring unit, the status monitoring unit being configured to monitor the status of multiple monitoring areas within the target smart park through a sensor group to obtain first fire protection data of the monitoring areas, the first fire protection data of the monitoring areas including the temperature, smoke concentration, and gas concentration of the monitoring areas, and transmit the first fire protection data of the monitoring areas to the threshold judgment unit; a threshold judgment unit, the threshold judgment unit receiving the first fire data of the monitoring area transmitted by the status monitoring unit, and comparing the first fire data of the monitoring area with a preset first fire data threshold; if the first fire data of the monitoring area is less than the preset first fire data threshold, the monitoring area is judged as a suspicious monitoring area, and basic information of the suspicious monitoring area is transmitted to the image monitoring unit; An image monitoring unit receives the basic information of the suspicious monitoring area transmitted by the threshold judgment unit, and calls the monitoring camera of the suspicious monitoring area based on the basic information of the suspicious monitoring area, and performs image monitoring of the suspicious monitoring area through the monitoring camera to obtain second fire data of the suspicious monitoring area, wherein the second fire data includes the monitoring image and monitoring video of the suspicious monitoring area, and transmits the second fire data of the suspicious monitoring area to the risk warning unit.

[0006] Preferably, the risk warning unit receives the second fire data of the suspicious monitoring area transmitted by the image monitoring unit, and inputs the monitoring image in the second fire data into the trained fire risk warning model, outputs the fire level and warning information corresponding to the monitoring image through the trained fire risk warning model to obtain the fire level and warning information of the suspicious monitoring area, and transmits the fire level and warning information of the suspicious monitoring area to the rescue request unit and the visualization large screen unit; The rescue request unit receives the fire level and warning information of the suspicious monitoring area transmitted by the risk warning unit, and compares the fire level of the suspicious monitoring area with the preset fire level. If the fire level of the suspicious monitoring area is greater than the preset fire level, the fire level and warning information of the suspicious monitoring area are matched with a pre-constructed fire knowledge spectrum to obtain a fire description of the suspicious monitoring area, and a rescue request is generated based on the fire level, warning information and fire description of the suspicious monitoring area, and the rescue request is transmitted to the rescue planning unit and the diffusion analysis unit.

[0007] Preferably, the rescue planning unit receives the rescue request transmitted by the rescue request unit, and performs fire station matching based on the suspicious monitoring area corresponding to the rescue request, so as to match the rescue request to the corresponding fire station; The diffusion analysis unit receives the rescue request transmitted by the rescue request unit, and performs smoke concentration analysis based on the basic information of the suspicious monitoring area corresponding to the rescue request to obtain the smoke concentration of the suspicious monitoring area, matches the smoke concentration of the suspicious monitoring area with the preset smoke concentration diffusion table to obtain the smoke diffusion range of the suspicious monitoring area, and transmits the smoke diffusion range of the suspicious monitoring area to the fire evacuation unit.

[0008] Preferably, the fire evacuation unit receives the smoke diffusion range of the suspicious monitoring area transmitted by the diffusion analysis unit, matches the smoke diffusion range of the suspicious monitoring area with the multiple monitoring areas to obtain a matching monitoring area, and issues a fire alarm through the sound and light equipment in the matching monitoring area; The visualization large screen unit receives the fire level and warning information of the suspicious monitoring area transmitted by the risk warning unit, and marks the fire level and warning information of the suspicious monitoring area on the electronic map of the target smart park as abnormal, and displays the electronic map of the target smart park after abnormal marking in the visualization terminal.

[0009] Preferably, the training method of the fire risk warning model includes: Based on the fire images of the target smart park and the corresponding fire keywords, a fire risk warning dataset is constructed. The fire risk warning model includes an encoder and a decoder; The encoder performs feature extraction on the fire images in the fire risk warning dataset through a feature extraction module to obtain an image risk feature vector set, and performs feature extraction on the fire keywords in the fire risk warning dataset through a keyword feature extraction module to obtain a keyword risk feature vector set; Decoding the image risk feature vector set and the keyword risk feature vector set corresponding to the fire risk warning dataset through a decoder, and calculating the probability distribution of the image risk feature vector set and the keyword risk feature vector set through a position feedforward network; The encoder and the decoder are optimized and trained through an alternating training mechanism, an encoding-decoding joint cross entropy loss function is constructed, image risk information and keyword risk information are calculated, and the feature differences between different modalities are minimized; The fire risk warning model is trained until it converges to obtain a trained fire risk warning model.

[0010] Preferably, the feature extraction module is used to extract features from the fire images in the fire risk warning dataset to obtain an image risk feature vector set, including: Input a fire image, adjust the size of the fire image, and perform feature extraction on the fire image using a ResNet-101 model; Using the output of the ResNet-101 model as the input of the RPN model, and extracting the region of interest of the fire image through the RPN model; The new vector convolution operation is used to reduce the dimension of the 5-channel vector convolution output of the ResNet-101 model to 1024, and the feature map with a dimension of 1024 is input to the classification convolution layer and the regression convolution layer respectively; The classification convolutional layer output The position-sensitive score map of the dimension is combined with the RPN model to obtain the classification result of the region of interest, where: represents the number of convolution kernels, Indicates the number of input channels, the regression convolution layer output The dimensional position sensitive score map is combined with the RPN model to obtain the regression result of the region of interest, and the image risk feature vector set is obtained based on the classification result and regression result of the region of interest.

[0011] Preferably, the keyword feature extraction module adopts the BERT model.

[0012] Preferably, performing fire station matching based on the suspicious monitoring area corresponding to the rescue request to match the rescue request to the corresponding fire station includes: Define the real-time waiting time for rescue within the jurisdiction of the fire station. The calculation formula for the real-time waiting time for rescue is as follows: ; in, Indicates the The day corresponding to time periods, Indicates the fire station's jurisdiction A rescue request, Indicates the fire station's jurisdiction Rescue request in sky Real-time waiting time for rescue within a time period, Indicates the Fire station within the jurisdiction of Rescue request in Real-time waiting time for rescue in each time period, Indicates the total number of days; Define the real-time driving rescue distance within the fire station's jurisdiction. The real-time driving rescue distance calculation formula is as follows: ; in, Indicates the fire station's jurisdiction Rescue request in sky Real-time driving rescue distance within a time period, Indicates the Fire station within its jurisdiction Rescue request in Real-time driving rescue distance for each time period.

[0013] Preferably, performing fire station matching based on the suspicious monitoring area corresponding to the rescue request to match the rescue request to the corresponding fire station further includes: The average travel time to rescue of the fire station is calculated based on the real-time waiting time for rescue within the jurisdiction of the fire station. The calculation formula of the average travel time to rescue is as follows: ; in, Indicates the The average rescue time of each fire station is Indicates the total time period, Indicates the total number of rescue requests; The average driving and rescue distance of the fire station is calculated based on the real-time driving and rescue distance within the fire station's jurisdiction. The calculation formula for the average driving and rescue distance is as follows: ; in, Indicates the Average driving rescue distance per fire station; The rescue request is matched to a corresponding fire station based on the average travel rescue time and average driving rescue distance of the plurality of fire stations.

[0014] Preferably, performing smoke concentration analysis based on basic information of the suspicious monitoring area corresponding to the rescue request to obtain the smoke concentration of the suspicious monitoring area includes: The smoke plume concentration is calculated based on the basic information of the suspicious monitoring area corresponding to the rescue request. The smoke plume concentration calculation formula is as follows: in, The spatial coordinates of the suspicious monitoring area corresponding to the rescue request are The plume concentration, represents the gas flow mass flow rate, Indicates the wind speed measured by the wind speed sensor, Indicates the height of the smoke plume from the ground. and represent the diffusion coefficients in the crosswind direction and vertical wind direction, respectively; The smoke puff concentration is calculated based on the basic information of the suspicious monitoring area corresponding to the rescue request. The smoke puff concentration calculation formula is as follows: in, The spatial coordinates of the suspicious monitoring area corresponding to the rescue request are The smoke density, Indicates the instantaneous discharge material mass, represents the downwind diffusion coefficient; The smoke plume concentration and the smoke puff concentration of the suspicious monitoring area corresponding to the rescue request are weighted averagely to obtain the smoke concentration of the suspicious monitoring area.

[0015] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects: 1. Through the combination of a status monitoring unit and a threshold judgment unit, the present invention enables the system to acquire and analyze fire data within the park in real time and promptly identify potential fire risk areas. This enables the system to quickly respond to possible fire incidents, thereby improving the speed and efficiency of fire response. The risk warning unit uses a trained fire risk warning model to analyze monitoring images and accurately determine the fire level and warning information. This precise warning helps avoid false alarms and missed alarms, and improves the reliability and accuracy of fire warnings.

[0016] 2. The present invention uses a rescue request unit to compare the fire level with the preset standard and generate a detailed rescue request based on the fire knowledge spectrum, which can ensure the rational scheduling and efficient use of rescue resources. The rescue planning unit matches the fire station according to the rescue request to ensure that the fire brigade can arrive at the scene quickly. The diffusion analysis unit can predict the diffusion range of smoke and transmit the result to the fire evacuation unit, which can help formulate effective evacuation plans and fire alarm strategies, reduce safety hazards caused by smoke diffusion, and provide a visual display of fire level and warning information through the visual large-screen unit, which can mark anomalies on the electronic map. This intuitive display method helps park managers and emergency response teams to quickly understand the fire situation and locate problem areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the system architecture of the overall system in an embodiment of the present invention.

[0018] Figure numerals: 1. Status monitoring unit; 2. Threshold judgment unit; 3. Image monitoring unit; 4. Risk warning unit; 5. Rescue request unit; 6. Visualization large screen unit; 7. Rescue planning unit; 8. Diffusion analysis unit; 9. Fire evacuation unit. DETAILED DESCRIPTION

[0019] Example 1, as Figure 1 As shown, the present invention proposes a smart park multi-source fire data dynamic monitoring and real-time analysis system, including a risk warning unit 4, a rescue request unit 5, a visualization large screen unit 6, a rescue planning unit 7, a diffusion analysis unit 8 and a fire evacuation unit 9, and also includes: Status monitoring unit 1, which is used to monitor the status of multiple monitoring areas within the target smart park through a sensor group to obtain first fire protection data of the monitoring areas, including the temperature, smoke concentration, and gas concentration of the monitoring areas, and transmit the first fire protection data of the monitoring areas to threshold judgment unit 2; The threshold judgment unit 2 receives the first fire data of the monitoring area transmitted by the status monitoring unit 1, and compares the first fire data of the monitoring area with a preset first fire data threshold. If the first fire data of the monitoring area is less than the preset first fire data threshold, the monitoring area is judged as a suspicious monitoring area, and basic information of the suspicious monitoring area is transmitted to the image monitoring unit 3; The image monitoring unit 3 receives the basic information of the suspicious monitoring area transmitted by the threshold judgment unit 2, and calls the monitoring camera of the suspicious monitoring area based on the basic information of the suspicious monitoring area, and performs image monitoring of the suspicious monitoring area through the monitoring camera to obtain the second fire data of the suspicious monitoring area. The second fire data includes the monitoring image and monitoring video of the suspicious monitoring area, and transmits the second fire data of the suspicious monitoring area to the risk warning unit 4.

[0020] In the present invention, the sensor group includes a temperature sensor, a smoke sensor and a gas sensor.

[0021] In an optional embodiment, the risk warning unit 4 receives the second fire data of the suspicious monitoring area transmitted by the image monitoring unit 3, and inputs the monitoring image in the second fire data into the trained fire risk warning model, and outputs the fire level and warning information corresponding to the monitoring image through the trained fire risk warning model to obtain the fire level and warning information of the suspicious monitoring area, and transmits the fire level and warning information of the suspicious monitoring area to the rescue request unit 5 and the visualization large screen unit 6.

[0022] In an optional embodiment, the rescue request unit 5 receives the fire level and warning information of the suspicious monitoring area transmitted by the risk warning unit 4, and compares the fire level of the suspicious monitoring area with the preset fire level. If the fire level of the suspicious monitoring area is greater than the preset fire level, the fire level and warning information of the suspicious monitoring area are matched with a pre-constructed fire knowledge spectrum to obtain a fire description of the suspicious monitoring area, and a rescue request is generated based on the fire level, warning information and fire description of the suspicious monitoring area, and the rescue request is transmitted to the rescue planning unit 7 and the diffusion analysis unit 8.

[0023] It should be noted that the pre-built fire knowledge graph is a systematic knowledge framework that contains various types of information about fires, such as fire type, severity, scope of impact, emergency response measures, etc. This information is usually organized in the form of a graph, enabling the system to accurately describe fires and handle emergencies based on real-time data.

[0024] In an optional embodiment, the rescue planning unit 7 receives the rescue request transmitted by the rescue request unit 5, and performs fire station matching based on the suspicious monitoring area corresponding to the rescue request to match the rescue request to the corresponding fire station.

[0025] In an optional embodiment, the diffusion analysis unit 8 receives the rescue request transmitted by the rescue request unit 5, and performs a smoke concentration analysis based on the basic information of the suspicious monitoring area corresponding to the rescue request to obtain the smoke concentration of the suspicious monitoring area, matches the smoke concentration of the suspicious monitoring area with a preset smoke concentration diffusion table to obtain the smoke diffusion range of the suspicious monitoring area, and transmits the smoke diffusion range of the suspicious monitoring area to the fire evacuation unit 9.

[0026] It should be noted that the smoke concentration diffusion table is a tool used to describe and predict the pattern and concentration distribution of smoke diffusion in a specific environment. It estimates the concentration of smoke at different locations based on meteorological conditions, terrain and smoke source characteristics to help assess and respond to the impact of smoke on health and the environment.

[0027] In an optional embodiment, the fire evacuation unit 9 receives the smoke diffusion range of the suspicious monitoring area transmitted by the diffusion analysis unit 8, and matches the smoke diffusion range of the suspicious monitoring area with multiple monitoring areas to obtain a matching monitoring area, and performs a fire alarm through the sound and light equipment in the matching monitoring area.

[0028] In an optional embodiment, the visualization large screen unit 6 receives the fire level and warning information of the suspicious monitoring area transmitted by the risk warning unit 4, and marks the fire level and warning information of the suspicious monitoring area on the electronic map of the target smart park as abnormal, and displays the electronic map of the target smart park after abnormal marking in the visualization terminal.

[0029] It should be noted that the electronic map of the target smart park is a digital, real-time updated map system that displays various information within the park, including geographical layout, buildings, facility locations and other key data. It is usually used to monitor and manage conditions within the park, such as fire, security or environmental changes, and supports visualization and decision support functions.

[0030] In the second embodiment, the present invention proposes a system for dynamic monitoring and real-time analysis of multi-source fire data in a smart park. Compared with the first embodiment, this embodiment further includes: a method for training a fire risk warning model, including: Based on the fire images and corresponding fire keywords of the target smart park, a fire risk warning dataset is constructed. The fire risk warning model includes an encoder and a decoder. The encoder uses the feature extraction module to extract features from the fire images in the fire risk warning dataset to obtain an image risk feature vector set, and uses the keyword feature extraction module to extract features from the fire keywords in the fire risk warning dataset to obtain a keyword risk feature vector set; The image risk feature vector set and keyword risk feature vector set corresponding to the fire risk warning dataset are decoded through the decoder, and the probability distribution of the image risk feature vector set and keyword risk feature vector set is calculated through the position feedforward network; By optimizing the encoder and decoder through an alternating training mechanism, a joint encoding-decoding cross-entropy loss function is constructed to calculate image risk information and keyword risk information, minimizing the feature differences between different modalities. The fire risk warning model is trained until it converges to obtain a trained fire risk warning model.

[0031] In this embodiment, the encoder refers to a part of the model that converts input data (such as fire images and keywords) into feature vectors; the decoder refers to another part of the model that converts the feature vectors generated by the encoder back into interpretable outputs; the position feedforward network is a neural network used to calculate the probability distribution of feature vectors, which can predict the risk probability of feature vectors in certain positions or conditions; the alternating training mechanism is a training method in which the encoder and decoder are trained alternately. Through alternating training, the encoder and decoder of the model can adjust to each other and optimize overall performance; the encoding-decoding joint cross entropy loss function is a loss function that combines the losses of the encoder and decoder. The cross entropy loss function is used to measure the difference between the predicted probability distribution and the true distribution. The joint loss function takes into account the differences in image and keyword features to minimize the difference between different modalities.

[0032] In an optional embodiment, a feature extraction module is used to extract features from fire images in a fire risk warning dataset to obtain an image risk feature vector set, including: Input a fire image, adjust the size of the fire image, and extract features of the fire image using the ResNet-101 model; The output of the ResNet-101 model is used as the input of the RPN model, and the region of interest of the fire image is extracted through the RPN model; The new vector convolution operation is used to reduce the dimension of the 5-channel output of the vector convolution of the ResNet-101 model to 1024. The feature map with a dimension of 1024 is input to the classification convolution layer and the regression convolution layer respectively. Classification convolutional layer output The position-sensitive score map of the dimensional space is combined with the RPN model to obtain the classification results of the region of interest, where represents the number of convolution kernels, Indicates the number of input channels, regression convolution layer output The dimensional position-sensitive score map is combined with the RPN model to obtain the regression result of the region of interest, and the image risk feature vector set is obtained based on the classification result and regression result of the region of interest.

[0033] It should be noted that the ResNet-101 model refers to a deep convolutional neural network architecture for image feature extraction. ResNet-101 is a variant of ResNet (residual network) with 101 layers. It avoids the gradient vanishing problem in deep network training by introducing residual connections, thereby effectively extracting image features; the RPN model refers to a network structure used to generate regions of interest in images. RPN generates region candidate boxes that may contain targets (such as fires) by performing sliding window operations on feature maps; regions of interest refer to region candidate boxes obtained from the RPN model, which are considered to contain fire-related content; the classification result of the region of interest refers to the result of whether each region contains a fire based on the output of the classification convolution layer combined with the region of interest generated by the RPN model; the regression result of the region of interest refers to the position correction result of each region based on the output of the regression convolution layer combined with the region of interest generated by the RPN model.

[0034] In an optional embodiment, the keyword feature extraction module adopts the BERT model.

[0035] It should be noted that the BERT model is a pre-trained model based on the Transformer architecture, which is used to generate context-dependent representations of text.

[0036] In an optional embodiment, performing fire station matching based on the suspicious monitoring area corresponding to the rescue request to match the rescue request to the corresponding fire station includes: Define the real-time waiting time for rescue within the jurisdiction of the fire station. The calculation formula for the real-time waiting time for rescue is as follows: ; in, Indicates the The day corresponding to time periods, Indicates the fire station's jurisdiction A rescue request, Indicates the fire station's jurisdiction Rescue request in sky Real-time waiting time for rescue within a time period, Indicates the Fire station within its jurisdiction Rescue request in Real-time waiting time for rescue in each time period, Indicates the total number of days; Define the real-time driving rescue distance within the fire station's jurisdiction. The real-time driving rescue distance calculation formula is as follows: ; in, Indicates the fire station's jurisdiction Rescue request in sky Real-time driving rescue distance within a time period, Indicates the Fire station within its jurisdiction Rescue request in Real-time driving rescue distance for each time period.

[0037] In an optional embodiment, matching fire stations based on the suspicious monitoring area corresponding to the rescue request to match the rescue request to the corresponding fire station further includes: The average travel time to rescue at a fire station is calculated based on the real-time waiting time for rescue within the fire station's jurisdiction. The formula for calculating the average travel time to rescue is as follows: ; in, Indicates the The average rescue time of each fire station is Indicates the total time period, Indicates the total number of rescue requests; The average driving rescue distance of a fire station is calculated based on the real-time driving rescue distance within the fire station's jurisdiction. The formula for calculating the average driving rescue distance is as follows: ; in, Indicates the Average driving rescue distance per fire station; The rescue request is matched to the corresponding fire station based on the average travel rescue time and average driving rescue distance of multiple fire stations.

[0038] In an optional embodiment, performing smoke concentration analysis based on basic information of the suspicious monitoring area corresponding to the rescue request to obtain the smoke concentration of the suspicious monitoring area includes: The smoke plume concentration is calculated based on the basic information of the suspicious monitoring area corresponding to the rescue request. The smoke plume concentration calculation formula is as follows: in, The spatial coordinates of the suspicious monitoring area corresponding to the rescue request are The plume concentration, represents the gas flow mass flow rate, Indicates the wind speed measured by the wind speed sensor, Indicates the height of the smoke plume from the ground. and represent the diffusion coefficients in the crosswind direction and vertical wind direction, respectively; The smoke puff density is calculated based on the basic information of the suspicious monitoring area corresponding to the rescue request. The smoke puff density calculation formula is as follows: in, The spatial coordinates of the suspicious monitoring area corresponding to the rescue request are The smoke density, Indicates the instantaneous discharge material mass, represents the downwind diffusion coefficient; The smoke plume concentration and smoke puff concentration in the suspicious monitoring area corresponding to the rescue request are weighted averagely to obtain the smoke concentration in the suspicious monitoring area.

[0039] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A smart park multi-source fire data dynamic monitoring and real-time analysis system, comprising a risk warning unit (4), a rescue request unit (5), a visualization large screen unit (6), a rescue planning unit (7), a diffusion analysis unit (8) and a fire evacuation unit (9), characterized in that: A state monitoring unit (1), the state monitoring unit (1) is used to monitor the state of multiple monitoring areas in the target smart park through a sensor group to obtain first fire data of the monitoring area, the first fire data of the monitoring area including the temperature, smoke concentration and gas concentration of the monitoring area, and transmit the first fire data of the monitoring area to a threshold judgment unit (2); A threshold judgment unit (2), wherein the threshold judgment unit (2) receives the first fire data of the monitoring area transmitted by the status monitoring unit (1), and compares the first fire data of the monitoring area with a preset first fire data threshold value; if the first fire data of the monitoring area is less than the preset first fire data threshold value, the monitoring area is judged as a suspicious monitoring area, and basic information of the suspicious monitoring area is transmitted to the image monitoring unit (3); An image monitoring unit (3), the image monitoring unit (3) receives the basic information of the suspicious monitoring area transmitted by the threshold judgment unit (2), and calls the monitoring camera of the suspicious monitoring area based on the basic information of the suspicious monitoring area, performs image monitoring of the suspicious monitoring area through the monitoring camera to obtain second fire protection data of the suspicious monitoring area, the second fire protection data including the monitoring image and monitoring video of the suspicious monitoring area, and transmits the second fire protection data of the suspicious monitoring area to the risk warning unit (4); The risk warning unit (4) receives the second fire data of the suspicious monitoring area transmitted by the image monitoring unit (3), and inputs the monitoring image in the second fire data into the trained fire risk warning model, outputs the fire level and warning information corresponding to the monitoring image through the trained fire risk warning model, so as to obtain the fire level and warning information of the suspicious monitoring area, and transmits the fire level and warning information of the suspicious monitoring area to the rescue request unit (5) and the visualization large screen unit (6); The training method of the fire risk early warning model includes: Based on the fire images of the target smart park and the corresponding fire keywords, a fire risk warning dataset is constructed. The fire risk warning model includes an encoder and a decoder; The encoder performs feature extraction on the fire images in the fire risk warning dataset through a feature extraction module to obtain an image risk feature vector set, and performs feature extraction on the fire keywords in the fire risk warning dataset through a keyword feature extraction module to obtain a keyword risk feature vector set; Decoding the image risk feature vector set and the keyword risk feature vector set corresponding to the fire risk warning dataset through a decoder, and calculating the probability distribution of the image risk feature vector set and the keyword risk feature vector set through a position feedforward network; The encoder and the decoder are optimized and trained through an alternating training mechanism, an encoding-decoding joint cross entropy loss function is constructed, image risk information and keyword risk information are calculated, and the feature differences between different modalities are minimized; The fire risk warning model is trained until it converges to obtain a trained fire risk warning model.

2. The smart park multi-source fire data dynamic monitoring and real-time analysis system according to claim 1 is characterized by: The rescue request unit (5) receives the fire level and warning information of the suspicious monitoring area transmitted by the risk warning unit (4), and compares the fire level of the suspicious monitoring area with the preset fire level. If the fire level of the suspicious monitoring area is greater than the preset fire level, the fire level and warning information of the suspicious monitoring area are matched with a pre-constructed fire knowledge spectrum to obtain a fire description of the suspicious monitoring area. A rescue request is generated based on the fire level, warning information and fire description of the suspicious monitoring area, and the rescue request is transmitted to the rescue planning unit (7) and the diffusion analysis unit (8).

3. The smart park multi-source fire data dynamic monitoring and real-time analysis system according to claim 2 is characterized by: The rescue planning unit (7) receives the rescue request transmitted by the rescue request unit (5), and performs fire station matching based on the suspicious monitoring area corresponding to the rescue request, so as to match the rescue request to the corresponding fire station; The diffusion analysis unit (8) receives the rescue request transmitted by the rescue request unit (5), and performs smoke concentration analysis based on basic information of the suspicious monitoring area corresponding to the rescue request to obtain the smoke concentration of the suspicious monitoring area, matches the smoke concentration of the suspicious monitoring area with the preset smoke concentration diffusion table to obtain the smoke diffusion range of the suspicious monitoring area, and transmits the smoke diffusion range of the suspicious monitoring area to the fire evacuation unit (9).

4. The smart park multi-source fire data dynamic monitoring and real-time analysis system according to claim 3 is characterized by: The fire evacuation unit (9) receives the smoke diffusion range of the suspicious monitoring area transmitted by the diffusion analysis unit (8), matches the smoke diffusion range of the suspicious monitoring area with the multiple monitoring areas to obtain a matching monitoring area, and issues a fire alarm through the sound and light equipment in the matching monitoring area; The visualization large screen unit (6) receives the fire level and warning information of the suspicious monitoring area transmitted by the risk warning unit (4), and marks the fire level and warning information of the suspicious monitoring area on the electronic map of the target smart park as abnormal, and displays the electronic map of the target smart park after abnormal marking on the visualization terminal.

5. The smart park multi-source fire data dynamic monitoring and real-time analysis system according to claim 1 is characterized by: The feature extraction module is used to extract features from the fire images in the fire risk warning dataset to obtain an image risk feature vector set, including: Input a fire image, adjust the size of the fire image, and perform feature extraction on the fire image using a ResNet-101 model; Using the output of the ResNet-101 model as the input of the RPN model, and extracting the region of interest of the fire image through the RPN model; The new vector convolution operation is used to reduce the dimension of the 5-channel vector convolution output of the ResNet-101 model to 1024, and the feature map with a dimension of 1024 is input to the classification convolution layer and the regression convolution layer respectively; The classification convolutional layer output The position-sensitive score map of the dimension is combined with the RPN model to obtain the classification result of the region of interest, where: represents the number of convolution kernels, Indicates the number of input channels, the regression convolution layer output The dimensional position sensitive score map is combined with the RPN model to obtain the regression result of the region of interest, and the image risk feature vector set is obtained based on the classification result and regression result of the region of interest.

6. The smart park multi-source fire data dynamic monitoring and real-time analysis system according to claim 1 is characterized by: The keyword feature extraction module adopts the BERT model.

7. The smart park multi-source fire data dynamic monitoring and real-time analysis system according to claim 3 is characterized by: Performing fire station matching based on the suspicious monitoring area corresponding to the rescue request to match the rescue request to a corresponding fire station includes: Define the real-time waiting time for rescue within the jurisdiction of the fire station. The calculation formula for the real-time waiting time for rescue is as follows: ； in, Indicates the The day corresponding to time periods, Indicates the fire station's jurisdiction A rescue request, Indicates the fire station's jurisdiction Rescue request in sky Real-time waiting time for rescue within a time period, Indicates the Fire station within its jurisdiction Rescue request in Real-time waiting time for rescue in each time period, Indicates the total number of days; Define the real-time driving rescue distance within the fire station's jurisdiction. The real-time driving rescue distance calculation formula is as follows: ； in, Indicates the fire station's jurisdiction Rescue request in sky Real-time driving rescue distance within a time period, Indicates the Fire station within its jurisdiction Rescue request in Real-time driving rescue distance for each time period.

8. The smart park multi-source fire data dynamic monitoring and real-time analysis system according to claim 7 is characterized by: Performing fire station matching based on the suspicious monitoring area corresponding to the rescue request to match the rescue request to a corresponding fire station also includes: The average travel time to rescue of the fire station is calculated based on the real-time waiting time for rescue within the jurisdiction of the fire station. The calculation formula of the average travel time to rescue is as follows: ； in, Indicates the The average rescue time of each fire station is Indicates the total time period, Indicates the total number of rescue requests; The average driving and rescue distance of the fire station is calculated based on the real-time driving and rescue distance within the fire station's jurisdiction. The calculation formula for the average driving and rescue distance is as follows: ； in, Indicates the Average driving rescue distance per fire station; The rescue request is matched to a corresponding fire station based on the average travel rescue time and average driving rescue distance of the plurality of fire stations.

9. The smart park multi-source fire data dynamic monitoring and real-time analysis system according to claim 3 is characterized by: Performing smoke concentration analysis based on basic information of the suspicious monitoring area corresponding to the rescue request to obtain the smoke concentration of the suspicious monitoring area includes: The smoke plume concentration is calculated based on the basic information of the suspicious monitoring area corresponding to the rescue request. The smoke plume concentration calculation formula is as follows: in, The spatial coordinates of the suspicious monitoring area corresponding to the rescue request are The plume concentration, represents the gas flow mass flow rate, Indicates the wind speed measured by the wind speed sensor, Indicates the height of the smoke plume from the ground. and represent the diffusion coefficients in the crosswind direction and vertical wind direction, respectively; The smoke puff concentration is calculated based on the basic information of the suspicious monitoring area corresponding to the rescue request. The smoke puff concentration calculation formula is as follows: in, The spatial coordinates of the suspicious monitoring area corresponding to the rescue request are The smoke density, Indicates the instantaneous discharge material mass, represents the downwind diffusion coefficient; The smoke plume concentration and the smoke puff concentration of the suspicious monitoring area corresponding to the rescue request are weighted averagely to obtain the smoke concentration of the suspicious monitoring area.

Images