A multi-phase lithium battery special shed intelligent tracking jet flow fire extinguishing method and system

Abstract

The application relates to the fire-fighting technical field, in particular to a multi-phase lithium battery special vehicle shed intelligent tracking jet fire extinguishing method and system. Double-light image data of a covered area are first collected, suspected fire area positions and initial confidence are output through feature detection, and the positions are moved to corresponding areas when reaching a first set value; double-light images and pose data are collected at multiple observation positions, secondary confidence and fire area contours are output through view angle alignment and feature fusion processing when reaching a second set value, and the positions are moved to close-range observation positions; multi-pitch-angle images and gas data are collected at the close-range positions, final fire confidence and fire grades are output through multi-source evidence fusion when reaching a third set value, and a mobile jet scheme containing a jet machine position, a jet angle, a jet mode and a jet time length is generated and executed. The method can improve fire determination accuracy and fire extinguishing treatment pertinence.

Description

Technical Field

[0001] This application relates to the field of fire protection technology, specifically to a method and system for intelligent tracking jet fire extinguishing for multiphase lithium battery-specific carports. Background Technology

[0002] Current fire suppression systems in concentrated vehicle parking areas mostly employ a combination of fixed-point detection devices and fixed sprinkler structures. The fixed-position detection devices can only collect data from a single perspective, making them susceptible to interference from ambient light and localized heat, resulting in a high false alarm rate for fire detection. Furthermore, the lack of a multi-level, progressive confirmation mechanism makes it difficult to improve accuracy while maintaining response speed. The sprinkler structures use fixed-position nozzles, and the spray position and angle cannot be dynamically adjusted according to the fire source location. A single fire can easily trigger a full-area spraying action, leading to unnecessary consumption of extinguishing agent and making it difficult to achieve targeted spraying at the fire source. In existing systems, the detection and spraying units are independent, making it impossible to achieve coordinated detection and spraying via mobile carriers. This makes it difficult to adapt to different parking area layouts, and the dynamic tracking and response capabilities for lithium battery-related fires are significantly insufficient. Summary of the Invention

[0003] The purpose of this application is to provide a method and system for intelligent tracking jet fire extinguishing for multiphase lithium battery-specific carports, so as to adapt to parking areas with different layouts and improve the dynamic tracking and handling capabilities for lithium battery fires.

[0004] In a first aspect, embodiments of this application provide an intelligent tracking jet fire extinguishing method for multiphase lithium battery-specific carports, comprising the following steps:

[0005] S1. Collect dual-light image data of the coverage area, perform feature extraction and target detection processing on the dual-light image data, and output the location information and preliminary confidence level of the suspected fire area; when the preliminary confidence level reaches the first set value, move to the location corresponding to the suspected fire area.

[0006] S2. Move along the set path to the observation position and collect the dual-light image data and pose data corresponding to each observation position; perform viewpoint alignment and feature fusion processing on the dual-light image data of multiple observation positions, and output the secondary confidence level and fire area contour data; when the secondary confidence level reaches the second set value, move to the closer observation position.

[0007] S3. Collect multiple sets of dual-light image data and gas concentration data at close-range observation positions, and calculate the rate of temperature change; perform multi-source parameter cross-validation processing on visual feature data, gas concentration data and temperature change rate, and output the final fire confidence and fire level data; when the final fire confidence reaches the third set value, trigger the fire extinguishing plan generation process.

[0008] S4. Based on the fire area contour data and fire level data, generate a mobile jet plan that includes the jet position, spray angle, spray mode and spray duration; move to the corresponding jet position in sequence according to the mobile jet plan and execute the fire extinguishing agent spraying operation.

[0009] In some embodiments, step S1 includes the following sub-steps:

[0010] S11. Acquire visible light image data and infrared temperature image data, and perform channel stitching processing on the visible light image data and infrared temperature image data to generate dual-light image data;

[0011] S12. Input the dual-light image data into the first-level fire detection model, extract the basic features of the image and identify the suspected target area, and output the planar coordinate data and preliminary confidence of the suspected fire area;

[0012] S13. When the initial confidence level is greater than or equal to the first set value, a position movement command is generated, and the position of the moving target is determined based on the planar coordinate data of the suspected fire area.

[0013] In some embodiments, step S2 includes the following sub-steps:

[0014] S21. Move along the linear track to the observation position in sequence, record the track position data and gimbal angle data corresponding to each observation position, and generate pose data;

[0015] S22. Collect dual-light image data from the corresponding viewpoint at each observation position, and perform pixel spatial alignment processing on multiple sets of dual-light image data based on pose data;

[0016] S23. Input the spatially aligned features of multiple sets of dual-light images into the second-level fire detection model, perform fusion verification of common features from multiple perspectives, and output the second-level confidence score and fire area contour data.

[0017] In some embodiments, step S3 includes the following sub-steps:

[0018] S31. Move to a close-up observation position, adjust the gimbal pitch angle, and collect multiple sets of dual-light image data at different pitch angles;

[0019] S32. Collect environmental smoke concentration data and carbon monoxide concentration data, and calculate the temperature change rate of the fire area based on continuous frame infrared temperature data;

[0020] S33. Extract the visual confidence level corresponding to the dual-light image data, combine it with smoke concentration data, carbon monoxide concentration data and temperature change rate, calculate the final fire confidence level through a weighted evidence fusion algorithm, and output the fire level data.

[0021] In some embodiments, step S4 includes the following sub-steps:

[0022] S41. Collect dual-light image data from multiple calibration locations and reconstruct the three-dimensional temperature field data of the fire area by combining the pose data;

[0023] S42. Based on three-dimensional temperature field data and fire level data, determine the number of jet nozzles and the coordinate position of each jet nozzle;

[0024] S43. Calculate the jet angle parameters for each jet position, match the jet mode and jet duration parameters, and generate a moving jet scheme.

[0025] In some embodiments, the first-level fire detection model adopts a lightweight convolutional neural network architecture. The first-level fire detection model includes a shared feature extraction backbone and a detection task head. The shared feature extraction backbone is used to extract multi-scale basic features of dual-light image data, and the detection task head is used to output the coordinates and confidence of the suspected target area. The first-level fire detection model is trained using a quantitative perception training method. The training dataset includes simulated fire data, measured fire data, and interference scene data.

[0026] In some embodiments, in step S23, the second-level fire detection model includes a viewpoint alignment attention module; the viewpoint alignment attention module calculates the pixel correspondence of images from different viewpoints based on pose data, strengthens the common features of multiple viewpoints and suppresses the interference features of single viewpoints through attention weights; the second-level fire detection model outputs the segmentation mask of the fire area as the contour data of the fire area.

[0027] In some embodiments, during step S4, when performing the fire extinguishing agent spraying operation, dual-light image data of the fire area are collected at set time intervals, and the three-dimensional temperature field data is updated; the spray angle parameters, spray mode parameters, and spray duration parameters are adjusted based on the updated three-dimensional temperature field data; when the temperature change rate is detected to exceed the set threshold, the jet position is adjusted and a supplementary spraying operation is performed.

[0028] In some embodiments, after step S4, the fire extinguishing agent spraying operation is stopped, and dual-light image data and gas concentration data of the fire area are collected at set time intervals; the multi-source parameter cross-validation process is repeatedly executed to determine the fire reignition status; when reignition characteristics are detected, the fire extinguishing plan generation process is triggered again; when no reignition characteristics are detected and the duration reaches the set value, the system is reset to the initial position.

[0029] Secondly, this application provides a multiphase lithium battery-specific intelligent tracking jet fire extinguishing system for carports, used to execute the above-mentioned method. The system includes a track-supporting assembly, an integrated mobile terminal, a control assembly, and a fire extinguishing agent supply assembly. The track-supporting assembly is installed on the top of the carport and extends along the length of the carport. The integrated mobile terminal is slidably mounted on the track-supporting assembly and integrates a sensing module, a jet module, and an edge computing module. The sensing module includes a visible light imaging unit, an infrared thermal imaging unit, and a gas detection unit. The jet module includes an adjustable nozzle, an electric pan-tilt unit, and a branch control valve. The control assembly includes an electronic control unit and a frequency converter drive unit, and is electrically connected to the integrated mobile terminal. The fire extinguishing agent supply assembly includes a storage tank, a liquid supply main pipe, and a pressure pump. The liquid supply main pipe is arranged along the track-supporting assembly and is connected to the jet module of the integrated mobile terminal through a flexible pipeline.

[0030] The beneficial effects of this application are as follows: This application adopts a multi-level, mobile, progressive fire assessment process. It acquires multi-view and multi-dimensional sensor data through mobile observation, and combines feature fusion and multi-source evidence fusion processing to effectively reduce the impact of false alarms caused by environmental interference factors, thereby improving the reliability of fire assessment results. Relying on a mobile integrated terminal to achieve integrated detection and spraying functions, the observation and spraying points can be flexibly adjusted according to the fire location, adapting to different parking areas and improving the targeting of fire suppression. A multi-position mobile jet scheme is generated based on the fire distribution status, and corresponding spraying parameters can be matched according to the fire level, improving the utilization efficiency of the extinguishing agent. Combined with a real-time feedback adjustment mechanism, it can adapt to the dynamic changes in the fire situation. Simultaneously, the accompanying reignition monitoring process reduces the possibility of fire reignition, comprehensively improving the fire protection effect of the parking area. Attached Figure Description

[0031] Figure 1 This is a flowchart of an intelligent tracking jet fire extinguishing method for a multiphase lithium battery-specific carport according to this application;

[0032] Figure 2 This is a flowchart of step S1 in this application;

[0033] Figure 3 This is a flowchart of step S2 in this application;

[0034] Figure 4 This is a flowchart of step S3 in this application;

[0035] Figure 5 This is a flowchart of step S4 in this application;

[0036] Figure 6 This is a schematic diagram of the module structure of an intelligent tracking jet fire extinguishing system for a multiphase lithium battery-specific carport according to this application.

[0037] Figure 7 This is a block diagram of a multiphase lithium battery-specific intelligent tracking jet fire extinguishing system for carports, as described in this application. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0039] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0040] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0041] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0042] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0043] Current fire suppression systems in concentrated vehicle parking areas mostly employ a combination of fixed-point detection devices and fixed sprinkler structures. The fixed-position detection devices can only collect data from a single perspective, making them susceptible to interference from ambient light and localized heating, resulting in a high false alarm rate for fire detection. Furthermore, the lack of a multi-level, progressive confirmation mechanism makes it difficult to improve accuracy while maintaining response speed. The sprinkler structure uses fixed-position nozzles, and the spray position and angle cannot be dynamically adjusted according to the fire source location. A single fire can easily trigger a full-area spraying action, causing unnecessary consumption of extinguishing agent and making it difficult to achieve targeted spraying at the fire source. In existing systems, the detection and spraying units are independent, making it impossible to achieve coordinated detection and spraying via a mobile platform. This makes it difficult to adapt to parking areas with different layouts, and the dynamic tracking and handling capabilities for lithium battery fires are significantly insufficient. Therefore, this application provides a method and system for intelligent tracking jet fire suppression in multiphase lithium battery-specific vehicle sheds.

[0044] This embodiment provides a specific implementation process for an intelligent tracking jet fire extinguishing method for a multiphase lithium battery-specific carport. This method relies on an intelligent tracking jet fire extinguishing system for the carport and is described in reference to... Figure 6 and Figure 7 The intelligent tracking jet fire extinguishing system for carports includes a track-supported assembly, an integrated mobile terminal, a control assembly, and a fire extinguishing agent supply assembly.

[0045] The track-supporting assembly is installed on the top of the carport and extends along the length of the carport. The integrated mobile terminal is slidably installed on the track-supporting assembly and integrates a sensing module, a jet module, and an edge computing module. The sensing module includes a visible light imaging unit, an infrared thermal imaging unit, and a gas detection unit. The jet module includes an adjustable nozzle, an electric pan-tilt unit, and a branch control valve. The control assembly includes an electrical control unit and a frequency converter drive unit, and is electrically connected to the integrated mobile terminal. The extinguishing agent supply assembly includes a storage tank, a liquid supply main pipe, and a pressure pump. The liquid supply main pipe is arranged along the track-supporting assembly and is connected to the jet module of the integrated mobile terminal through a flexible pipeline.

[0046] The fire extinguishing method provided in this application identifies the fire status through a multi-level progressive mobile observation and AI judgment process, and generates a dynamic jet treatment plan based on the fire distribution, referring to... Figure 1 The specific implementation process is as follows:

[0047] Step 1. Global inspection and preliminary fire detection: Collect dual-light image data of the covered area, perform feature extraction and target detection processing on the dual-light image data, and output the location information and preliminary confidence level of the suspected fire area; when the preliminary confidence level reaches a first set value, move to the location corresponding to the suspected fire area, referring to... Figure 2 Specifically, it includes:

[0048] Step 1.1. Dual-light image data acquisition and stitching: Visible light image data and infrared temperature image data of the covered area are acquired. After spatial registration of the visible light image data and infrared temperature image data, channel stitching is performed to generate dual-light image data. The dual-light image data contains texture information in the visible light band and temperature information in the infrared band, which can simultaneously provide the visual morphological features and temperature distribution features of the fire.

[0049] Spatial registration is performed using pre-calibrated intrinsic parameters of the visible light imaging unit and the infrared thermal imaging unit, as well as an extrinsic parameter matrix between the two imaging units. This aligns the images acquired by the two imaging units to the same pixel coordinate system, ensuring a one-to-one correspondence of pixel positions for the same object in both images. The spatial registration process first corrects distortion in the infrared image, then maps the pixel coordinates of the infrared image to the pixel coordinate system of the visible light image using the extrinsic parameter matrix, ultimately outputting two-light image data with consistent size and pixel alignment.

[0050] Step 1.2. First-level fire detection model inference: The dual-light image data is input into the first-level fire detection model. The first-level fire detection model extracts multi-scale basic features from the dual-light image data, identifies suspected target areas in the image, and outputs the planar coordinate data and preliminary confidence scores of the suspected fire areas. The first-level fire detection model adopts a lightweight convolutional neural network architecture for rapid initial fire screening in front-end inspection scenarios. The first-level fire detection model includes a shared feature extraction backbone and a detection task head, which are connected sequentially. The shared feature extraction backbone is used to extract multi-level visual features from the input image, and the detection task head is used to output target detection results based on the features.

[0051] The shared feature extraction backbone uses depthwise separable convolution instead of standard convolution, which breaks down the standard convolution into two steps: depthwise convolution and pointwise convolution. The detection task head computes based on multi-scale feature map tensors, outputting the class confidence and bounding box coordinates for each candidate region.

[0052] The first-level fire detection model is trained using a quantitative perception training method. The training dataset includes simulated fire data, measured fire data, and interference scenario data. The simulated fire data is generated by fire dynamics simulation software and covers different combustion states and environmental conditions. The measured fire data comes from fire data collected in real scenarios. The interference scenario data includes various non-fire scenario data that are prone to misjudgment.

[0053] Step 1.3. Preliminary Judgment and Location Movement Trigger: The preliminary confidence level is compared with a first preset value. If the preliminary confidence level is greater than or equal to the first preset value, a location movement command is generated. The target location is determined based on the planar coordinate data of the suspected fire area. The first preset value is a pre-set initial screening threshold used to filter out obvious non-fire scenarios and reduce the triggering frequency of subsequent secondary judgment processes. The location movement command drives the integrated mobile terminal to move along the linear track on the roof of the carport towards the location corresponding to the suspected fire area. During the movement, the sensing module maintains its acquisition status and continuously updates the location information of the suspected fire area to ensure the correspondence between the movement endpoint and the suspected area.

[0054] In some embodiments, a temperature threshold constraint is introduced during the inference process of the first-level fire detection model. Candidate regions with infrared temperature values ​​lower than the set temperature value are directly filtered out and do not enter the subsequent confidence calculation process, thereby reducing the amount of calculation for invalid candidate regions and improving the overall running speed of the first-level detection process.

[0055] Step 2. Multi-view mobile observation and level-two fire assessment, refer to Figure 3The system moves along a predetermined path to the observation position, collecting dual-light image data and pose data for each position. It then performs viewpoint alignment and feature fusion processing on the dual-light image data from multiple observation positions, outputting secondary confidence level and fire area contour data. When the secondary confidence level reaches a second predetermined value, the system moves to a closer observation position, specifically including:

[0056] Step 2.1. Multi-Observation Point Movement and Pose Data Acquisition: Move sequentially along the linear track to the observation positions, record the track position data and gimbal angle data corresponding to each observation position, and generate pose data. The pose data includes the one-dimensional position coordinates of the integrated mobile terminal on the track, as well as the horizontal and pitch angles of the motorized gimbal, used to characterize the spatial attitude of the current observation viewpoint. Each observation position corresponds to an independent observation viewpoint, and multiple observation positions are distributed at intervals along the linear track to form a multi-view observation array in the horizontal direction. The distance between adjacent observation positions ranges from 0.5m to 1.5m, and in this embodiment, the specific value can be 1m. After moving to each observation position, trigger the gimbal angle locking and image acquisition commands to ensure the stability of the pose state when acquiring images.

[0057] Step 2.2. Multi-view dual-light image spatial alignment processing: Dual-light image data of the corresponding viewpoint are acquired at each observation position. Based on the pose data, pixel spatial alignment processing is performed on multiple sets of dual-light image data. Pixel spatial alignment processing is achieved through homography transformation, which maps images acquired from different viewpoints to the same reference plane, realizing pixel-level alignment of multi-view images.

[0058] The homography transformation matrix can be calculated from pose data and camera intrinsic parameters, and the homography transformation matrix varies for different observation positions. The calculation process first calculates the camera's extrinsic parameter matrix based on the orbit position and gimbal angle, and then derives the homography transformation matrix between the two viewpoints by combining the camera's intrinsic parameter matrix. After spatial alignment processing, the same spatial point in multiple sets of dual-light images corresponds to the same pixel coordinates, which are used to provide data for subsequent feature fusion.

[0059] Step 2.3. Multi-view Feature Fusion and Secondary Confidence Output: Multiple sets of spatially aligned dual-light image features are input into the second-level fire detection model. Common features from multiple views are fused and verified, and the secondary confidence score and fire area contour data are output. The second-level fire detection model includes a view alignment attention module, which is used to perform weighted fusion of multi-view features, strengthening features common to different views and suppressing interfering features that only appear in a single view. The calculation process of the view alignment attention module is as follows:

[0060]

[0061] in Let represent the feature map tensor corresponding to the i-th viewpoint, Concat represents the concatenation operation along the channel dimension, W and b represent the learnable weight and bias parameters, and Softmax represents the normalized exponential function used to normalize the weight values ​​into a probability distribution. This represents the attention weight corresponding to the i-th view feature. This represents the fused global feature map tensor. The second-level fire detection model calculates based on the fused global feature map tensor, outputting a second-level confidence score and fire area contour data. The fire area contour data is presented as a segmentation mask, which is a binary matrix with the same size as the input image, used to identify the pixel boundaries between fire areas and non-fire areas. When the second-level confidence score is greater than or equal to a second preset value, the integrated mobile terminal is driven to move towards a closer observation position.

[0062] In some embodiments, the number and spacing of observation points can be automatically adjusted according to the size of the suspected fire area. When the suspected fire area is large, the number of observation points is increased and the spacing is widened; when the suspected fire area is small, the number of observation points is reduced and the spacing is narrowed. Dynamically adjusted observation points can reduce unnecessary observation procedures while ensuring the accuracy of the judgment.

[0063] Step 3. Close-range multi-source fusion and final fire assessment: Multiple sets of dual-light image data and gas concentration data at elevation angles are acquired at close-range observation locations, and the temperature change rate is calculated. Multi-source parameter cross-validation processing is performed on the visual feature data, gas concentration data, and temperature change rate to output the final fire confidence score and fire level data. When the final fire confidence score reaches the third preset value, the fire extinguishing plan generation process is triggered, referring to… Figure 4 Specifically, it includes:

[0064] Step 3.1. Multi-angle Image and Gas Data Acquisition: Move to a close-range observation position, adjust the gimbal's pitch angle, and acquire multiple sets of dual-light image data at different pitch angles. Simultaneously acquire environmental smoke concentration data and carbon monoxide concentration data, and calculate the temperature change rate of the fire area based on continuous frame infrared temperature data. By moving to a closer observation position to the suspected fire area, higher resolution image data and more accurate gas concentration data can be obtained. Multiple pitch angle observations can form multi-dimensional observation data in the vertical direction, supplementing the information dimension of the horizontal direction. The gas detection unit uses electrochemical sensing principles to continuously acquire smoke particle concentration and carbon monoxide gas concentration in the environment. The acquired gas data is filtered and used for subsequent evidence fusion calculations.

[0065] Step 3.2. Temperature Change Rate Calculation and Multi-Source Evidence Extraction: The temperature change rate of the fire area is calculated based on continuous time-series infrared temperature data. Basic probability allocation values ​​are extracted for visual evidence, smoke concentration evidence, carbon monoxide concentration evidence, and temperature change rate evidence. The basic probability allocation value for visual evidence is obtained by converting the secondary confidence level output from the second-level fire detection model; the basic probability allocation value for smoke concentration evidence is calculated based on the correspondence between smoke concentration data and a set concentration value; the basic probability allocation value for carbon monoxide concentration evidence is calculated based on the correspondence between carbon monoxide concentration data and a set concentration value; and the basic probability allocation value for temperature change rate evidence is calculated based on the correspondence between the temperature change rate and a set rate value. Each piece of evidence corresponds to two basic probability allocation values: a fire proposition and a non-fire proposition, representing the degree to which the evidence supports or does not support the occurrence of a fire.

[0066] Step 3.3. Weighted Evidence Fusion and Fire Situation Level Output: The basic probability allocation values ​​of multiple sets of evidence are weighted and corrected. The final fire confidence level is calculated using an evidence fusion algorithm, and the fire situation level data is output. The evidence fusion algorithm adopts the Dempster-Shafer evidence theory. In this embodiment, it is used to fuse visual and non-visual multi-source sensor data to improve the reliability of the fire situation determination result. The weighting correction process assigns a corresponding weight coefficient to each piece of evidence. The weight coefficient is preset according to the reliability of the evidence. The basic probability allocation value of each piece of evidence is multiplied by the corresponding weight coefficient to obtain the corrected basic probability allocation value. The mathematical expression of the Dempster combination rule is:

[0067]

[0068] Where m1 and m2 represent the basic probability assignment functions corresponding to the two pieces of evidence. and Let K represent the set of propositions corresponding to two pieces of evidence, K represent the evidence conflict coefficient, used to characterize the degree of conflict between the two pieces of evidence, and m(A) represent the basic probability assignment value corresponding to proposition A after fusion. Multiple sets of evidence are fused sequentially using the Dempster combination rule to finally obtain the final fire confidence level corresponding to the fire proposition. Fire level data is divided according to the value range of the final fire confidence level. The fire level data is used to characterize the severity of the fire and provide a basis for the subsequent jetting scheme generation. When the final fire confidence level is greater than or equal to a third set value, the fire extinguishing scheme generation process is triggered.

[0069] In some embodiments, the type of evidence can be expanded according to the sensor configuration of the deployment scenario, adding other types of gas detection data or ultraviolet flame detection data as supplementary evidence. Expanded evidence can further enrich the information dimensions of multi-source fusion and adapt to the fire detection needs in different scenarios.

[0070] The weighted evidence fusion algorithm used in the above multi-source evidence fusion processing flow adaptively adjusts the weight coefficients of each piece of evidence based on the current scenario. Specifically, during operation, the system continuously monitors the signal noise level of each sensor channel and the consistency index with historical judgments. When the noise variance of a piece of evidence exceeds the set range or its judgment result continuously conflicts with most pieces of evidence, the system automatically reduces the weight coefficient corresponding to that piece of evidence; conversely, it appropriately increases it.

[0071] In addition, the viewpoint alignment attention module in the second-level fire detection model introduces a contrastive learning strategy during the training phase. By constructing positive and negative sample pairs, positive samples are image features of the same fire from different viewpoints, and negative samples are features of different fire or non-fire scenarios, the model's ability to express common fire features is enhanced. This allows the model to maintain stable segmentation mask output accuracy when facing different environments, further improving the model's adaptability.

[0072] Step 4. Three-dimensional thermal field reconstruction and jet plan execution: Based on the fire area contour data and fire level data, generate a moving jet plan including jet position, spray angle, spray mode, and spray duration; move to the corresponding jet position according to the moving jet plan and execute the extinguishing agent spraying operation, referring to... Figure 5 Specifically, it includes:

[0073] Step 4.1. Calibration Point Acquisition and 3D Temperature Field Reconstruction: Move along a linear track to multiple calibration positions, acquire dual-light image data and pose data corresponding to each calibration position, and combine the pose data to reconstruct the 3D temperature field data of the fire area, that is, calculate the temperature distribution plane of the fire area and locate the position of the highest temperature point. 3D temperature field data refers to field data containing spatial 3D coordinates and corresponding temperature values, which can characterize the spatial distribution of temperature in the fire area. The reconstruction process first performs depth estimation on the thermal infrared image of each calibration position to obtain the relative depth value corresponding to each pixel; then, combined with the pose data, the pixel coordinates are converted into 3D coordinates in the world coordinate system to generate 3D temperature point cloud data; finally, continuous 3D temperature field data is generated through a surface reconstruction algorithm.

[0074] Step 4.2. Jet Position and Spray Parameter Planning: Based on three-dimensional temperature field data and fire level data, determine the number of jet positions and the corresponding coordinates of each jet position; calculate the spray angle parameters for each jet position, match the spray mode and spray duration parameters, and generate a moving jet scheme. The selection of jet positions is based on the coverage weight of temperature distribution to ensure that each location in the fire area can be covered by the jet from at least one jet position. The selection process first generates a set of candidate positions within the feasible range of the trajectory, calculates the coverage area and temperature-weighted coverage value of each candidate position for the fire area, and selects the candidate positions with the highest weighted coverage values ​​as the final jet positions. The calculation of the spray angle parameters is corrected in conjunction with the ballistic characteristics of the extinguishing agent jet. The extinguishing agent jet exhibits a parabolic trajectory under the action of gravity. The trajectory correction calculation is based on the projectile motion model. The trajectory formula of projectile motion is:

[0075]

[0076] Where x represents the horizontal distance, y represents the vertical height difference, and θ represents the jet elevation angle. The initial velocity of the extinguishing agent jet is represented by , and g represents the acceleration due to gravity. Based on the horizontal distance and vertical height difference from the target point, combined with the initial velocity parameter, the corresponding jet elevation angle can be calculated to ensure the jet's impact point coincides with the target location. The spray modes include direct current spray mode and atomized spray mode, each corresponding to different extinguishing agent flow rates and coverage areas. The spray duration parameter is set based on fire severity and temperature distribution data; the higher the fire severity, the longer the spray duration. The mobile jet plan includes the sequence of jet operator movement, the spray angle for each operator, the spray mode, and the spray duration.

[0077] Step 4.3. Execution of the Moving Jet Scheme and Real-time Feedback Adjustment: Following the moving jet scheme, move sequentially to the corresponding jet position and execute the extinguishing agent spraying operation. During the extinguishing agent spraying operation, acquire dual-light image data of the fire area at set time intervals and update the three-dimensional temperature field data; adjust the spray angle parameters, spray mode parameters, and spray duration parameters based on the updated three-dimensional temperature field data; when the temperature change rate exceeds the set threshold, adjust the jet position and execute a supplementary spraying operation. The real-time feedback adjustment mechanism can adjust the spraying strategy in a timely manner according to the dynamic changes of the fire, adapting to different stages of fire development. During spraying, the spray mode can be switched according to the temperature drop status; if the temperature drop rate meets expectations, maintain the current spray parameters; if the temperature drop rate does not meet expectations, adjust the spray parameters or change the jet position. Supplementary spraying operations prioritize the jet position with the highest coverage weight for the temperature anomaly area, adjust the spray parameters, and execute supplementary spraying until the temperature change rate returns to the set range.

[0078] In some embodiments, a fixedly arranged spray assembly can be activated during the spraying process to assist in cooling. The spray assembly is arranged at intervals along the main liquid supply pipe, forming a planar area covered by the extinguishing agent. When the fire area is detected to exceed a set range, the spray assembly is activated to cool the entire area, and coordinated with an integrated mobile terminal for collaborative handling.

[0079] After the extinguishing agent spraying operation is completed, the spraying operation is stopped, and dual-light image data and gas concentration data of the fire area are collected at set time intervals. The multi-source evidence fusion processing flow is repeatedly executed to determine the fire reignition status. If reignition characteristics are detected, the fire extinguishing plan generation process is triggered again. If no reignition characteristics are detected and the duration reaches the set value, the integrated mobile terminal is driven to reset to the initial position. During reignition monitoring, a low-power operation state is maintained. The complete multi-source evidence fusion calculation flow is only initiated when characteristics such as temperature rise or gas concentration increase are detected, balancing the continuity of monitoring and the energy consumption of the equipment. During the reset process, the device returns to the initial docking position along the track, shuts down unnecessary sensing and computing units, and returns to the standby inspection state.

[0080] The intelligent tracking jet fire extinguishing method for carports provided in this embodiment effectively reduces the false alarm probability of fire identification and improves the reliability of fire assessment results through a multi-level, progressive fire assessment process combined with multi-view observation and multi-source data fusion. A mobile, integrated terminal integrates fire detection and jet treatment, allowing for flexible adjustment of observation and spray positions based on fire location, adapting to different layouts of carport parking areas and enhancing the targeted nature of fire extinguishing. The combination of three-dimensional temperature field reconstruction and dynamic jet planning allows for adjustment of spray positions and parameters according to the spatial distribution of the fire, helping to improve the utilization efficiency of extinguishing agents and reduce unnecessary consumption. A real-time feedback adjustment mechanism and reignition monitoring process adapt to dynamic changes in the fire situation, allowing for timely adjustments to the response strategy and reducing the possibility of reignition. The supporting system adopts a modular structural design, with detachable connections between the integrated terminal and the track and pipelines, facilitating installation and subsequent maintenance. The number and placement of terminals can be flexibly adjusted according to actual usage needs. The overall solution balances the accuracy of fire detection with the dynamism of fire suppression, and can be applied to fire protection scenarios in various electric vehicle parking areas.

[0081] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0082] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for intelligent tracking jet fire extinguishing in a multiphase lithium battery-specific carport, characterized in that, Includes the following steps: S1. Collect dual-light image data of the coverage area, perform feature extraction and target detection processing on the dual-light image data, and output the location information and preliminary confidence level of the suspected fire area; when the preliminary confidence level reaches the first set value, move to the location corresponding to the suspected fire area; S2. Move along the set path to the observation position, and collect the dual-light image data and pose data corresponding to each observation position; perform viewpoint alignment and feature fusion processing on the dual-light image data of multiple observation positions, and output the secondary confidence level and fire area contour data; when the secondary confidence level reaches the second set value, move to the closer observation position; S3. Collect multiple sets of dual-light image data and gas concentration data at close-range observation positions, and calculate the temperature change rate; perform multi-source parameter cross-validation processing on visual feature data, gas concentration data and temperature change rate, and output the final fire confidence and fire level data; when the final fire confidence reaches the third set value, trigger the fire extinguishing plan generation process. S4. Based on the fire area contour data and fire level data, generate a mobile jet plan that includes the jet position, spray angle, spray mode and spray duration; move to the corresponding jet position in sequence according to the mobile jet plan and execute the fire extinguishing agent spraying operation.

2. The method according to claim 1, characterized in that, Step S1 includes the following sub-steps: S11. Acquire visible light image data and infrared temperature image data, and perform channel stitching processing on the visible light image data and infrared temperature image data to generate dual-light image data; S12. Input the dual-light image data into the first-level fire detection model, extract the basic features of the image and identify the suspected target area, and output the planar coordinate data and preliminary confidence of the suspected fire area; S13. When the initial confidence level is greater than or equal to the first set value, a position movement command is generated, and the position of the moving target is determined based on the planar coordinate data of the suspected fire area.

3. The method according to claim 1, characterized in that, Step S2 includes the following sub-steps: S21. Move along the linear track to the observation position in sequence, record the track position data and gimbal angle data corresponding to each observation position, and generate pose data; S22. Collect dual-light image data from the corresponding viewpoint at each observation position, and perform pixel spatial alignment processing on multiple sets of dual-light image data based on pose data; S23. Input the spatially aligned features of multiple sets of dual-light images into the second-level fire detection model, perform fusion verification of common features from multiple perspectives, and output the second-level confidence score and fire area contour data.

4. The method according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31. Move to a close-up observation position, adjust the gimbal pitch angle, and collect multiple sets of dual-light image data at different pitch angles; S32. Collect environmental smoke concentration data and carbon monoxide concentration data, and calculate the temperature change rate of the fire area based on continuous frame infrared temperature data; S33. Extract the visual confidence level corresponding to the dual-light image data, combine it with smoke concentration data, carbon monoxide concentration data and temperature change rate, calculate the final fire confidence level through a weighted evidence fusion algorithm, and output the fire level data.

5. The method according to claim 1, characterized in that, Step S4 includes the following sub-steps: S41. Collect dual-light image data from multiple calibration locations and reconstruct the three-dimensional temperature field data of the fire area by combining the pose data; S42. Based on three-dimensional temperature field data and fire level data, determine the number of jet nozzles and the coordinate position of each jet nozzle; S43. Calculate the jet angle parameters for each jet position, match the jet mode and jet duration parameters, and generate a moving jet scheme.

6. The method according to claim 2, characterized in that, In step S12, the first-level fire detection model adopts a lightweight convolutional neural network architecture, which includes a shared feature extraction backbone and a detection task head. The shared feature extraction backbone is used to extract multi-scale basic features from dual-light image data, and the detection task head is used to output the coordinates and confidence of suspected target areas. The first-level fire detection model is trained using a quantitative perception training method, and the training dataset includes simulated fire data, measured fire data, and interference scene data.

7. The method according to claim 3, characterized in that, In step S23, the second-level fire detection model includes a view alignment attention module; The viewpoint alignment attention module calculates the pixel correspondence between images from different viewpoints based on pose data, and strengthens the common features of multiple viewpoints and suppresses the interference features of single viewpoints through attention weights. The second-level fire detection model outputs a segmentation mask of the fire area as the fire area contour data.

8. The method according to claim 1, characterized in that, In step S4, during the execution of the extinguishing agent spraying operation, dual-light image data of the fire area are collected at set time intervals, and the three-dimensional temperature field data is updated; the spray angle parameters, spray mode parameters, and spray duration parameters are adjusted based on the updated three-dimensional temperature field data; when the temperature change rate is detected to exceed the set threshold, the jet position is adjusted and a supplementary spraying operation is performed.

9. The method according to claim 1, characterized in that, After step S4, the fire extinguishing agent spraying operation is stopped, and dual-light image data and gas concentration data of the fire area are collected at set time intervals; the multi-source parameter cross-validation process is repeated to determine the fire reignition status; when reignition characteristics are detected, the fire extinguishing plan generation process is triggered again; when no reignition characteristics are detected and the duration reaches the set value, the system is reset to the initial position.

10. A multiphase lithium battery-specific intelligent tracking jet fire extinguishing system for carports, used to execute the method as described in any one of claims 1-9, characterized in that, The system includes a track-supported assembly, an integrated mobile terminal, a control assembly, and a fire extinguishing agent supply assembly. The track-supported assembly is located on the top of the vehicle shed and extends along the length of the shed. The integrated mobile terminal is slidably mounted on the track-supported assembly and integrates a sensing module, a jet module, and an edge computing module. The sensing module includes a visible light imaging unit, an infrared thermal imaging unit, and a gas detection unit. The jet module includes an adjustable nozzle, an electric pan-tilt unit, and a branch control valve. The control assembly includes an electronic control unit and a frequency converter drive unit, and is electrically connected to the integrated mobile terminal. The fire extinguishing agent supply assembly includes a storage tank, a main supply pipe, and a pressure pump. The main supply pipe is arranged along the track-supported assembly and is connected to the jet module of the integrated mobile terminal through a flexible pipeline.

Images