New energy battery thermal runaway multi-parameter coupling positioning hierarchical early warning system and method

Abstract

The application discloses a new energy battery thermal runaway multi-parameter coupling positioning hierarchical early warning system and method, belongs to the technical field of new energy battery safety monitoring and fire early warning, and the system is networked and interacted by four modules of front-end multi-parameter sensing, edge computing and transmission, background analysis and management, display and control. The front end synchronously collects thermal imaging, visible light video and battery thermal runaway characteristic gas multi-class parameters and wirelessly converges; the edge side analyzes a temperature matrix to construct four-stage progressive thermal imaging early warning, identifies local anomalies by suppressing light interference through pixel-by-pixel brightness time series analysis, encapsulates multi-parameter data and executes on-site control instructions; the background relies on an industrial computer IP to preferentially determine a monitoring area node, decouples to realize regional attribution and hierarchical management of sensing instances, pairs double-camera hotspot coordinates within a time window, solves fire three-dimensional coordinates by means of regional calibration full connection neural network and quantifies position reliability, and uniformly manages multi-source alarms, dynamically updates fire positioning records.

Description

Technical Field

[0001] This invention relates to the field of new energy battery safety monitoring and fire early warning technology, specifically to a multi-parameter coupled positioning and hierarchical early warning system and method for thermal runaway of new energy batteries. Background Technology

[0002] With the rapid development of the new energy industry, lithium-ion batteries and other new energy battery systems have been widely used in various scenarios such as grid energy storage, electric vehicles, ship power supply, and stationary backup power. During long-term charge and discharge operation, new energy batteries are susceptible to thermal runaway caused by factors such as internal short circuits, overcharging and over-discharging, thermal abuse, and mechanical damage. Thermal runaway is characterized by insidious early signs, rapid temperature rise, rapid fire spread, and a high risk of reignition. If abnormal temperature rise cannot be identified and accurately located at an early stage, it can easily escalate into a fire or explosion, causing significant property damage and personal safety risks. Therefore, building a highly reliable and timely battery thermal runaway monitoring and early warning system has become a core issue that urgently needs to be addressed in the field of new energy battery safety protection.

[0003] Currently, various monitoring and early warning technologies exist for battery thermal runaway and fires in energy storage sites, but they still have many shortcomings in practical engineering applications:

[0004] First, single-parameter monitoring methods suffer from delayed early warning and lack a tiered, progressive assessment mechanism. Current mainstream solutions rely heavily on single-point temperature or smoke detectors or single-threshold thermal imaging temperature measurements, triggering alarms only when temperature or smoke reaches fixed thresholds. This fails to match the progressive development of battery thermal runaway, from abnormal temperature rise, thermal diffusion, open flame generation to fire expansion, making early anomaly identification difficult. Furthermore, single-parameter monitoring is susceptible to interference from ambient residual heat, ventilation conditions, and partial obstruction, resulting in a high risk of false alarms and missed alarms, thus failing to provide a basis for tiered response.

[0005] Secondly, visual anomaly detection has weak anti-interference capabilities and struggles to capture weak early signals. Existing visible light-based flame recognition and anomaly detection technologies largely rely on global image features and template matching. In complex lighting conditions in industrial settings, such as equipment reflections, indicator light flickering, and overall fluctuations in ambient light, the false positive rate increases significantly. Most solutions do not conduct pixel-by-pixel brightness temporal analysis, failing to effectively suppress global brightness synchronization disturbances. Small-sized, low-amplitude local brightness anomalies in the early stages of battery thermal runaway are easily masked by environmental noise, making accurate identification difficult.

[0006] Third, the data attribution and sensor instance management mechanisms in a multi-level wireless transmission architecture lack adaptability. In a large-scale deployment architecture of "wireless sensor node - wireless gateway - edge industrial control computer - backend", existing systems typically use sensor IP or gateway IP as the basis for determining area attribution. When the same edge industrial control computer connects to multiple wireless gateways and multiple sensor nodes, area attribution errors are likely to occur. Furthermore, existing solutions do not decouple "monitoring area attribution determination" from "sensor instance differentiation within the area", making it impossible to accurately distinguish multiple wireless sensor nodes within the same area and aggregate their various parameters under the premise of stable attribution area. This is not conducive to the large-scale deployment and operation and maintenance management of multiple nodes in the field.

[0007] Fourth, the accuracy of multi-camera fire source localization is limited, and its on-site adaptability and maintainability are poor. Existing multi-view heat source localization solutions mostly adopt traditional binocular geometric ranging models, which rely on pre-calibrated camera intrinsic and extrinsic parameters. When the size of the monitoring area, the camera installation angle, or the position is adjusted, it is difficult to correct the model parameters and the operation and maintenance costs are high, making it difficult to adapt to the differentiated deployment needs of different sites. At the same time, most localization solutions do not conduct quantitative confidence assessment of the localization results, which cannot intuitively reflect the reliability of the localization results, and it is difficult to balance localization accuracy and engineering adaptability.

[0008] Fifth, alarm management and linkage control are disconnected, failing to form a complete closed-loop response mechanism. Most existing early warning systems only display and push alarm information. Multi-source alarm data, fire source location results, and linkage control processes are independent of each other, lacking a unified node-based correlation management mechanism. Control command distribution is not strictly isolated according to monitoring areas, posing a safety hazard of cross-regional erroneous execution. Furthermore, command execution results are not linked to alarm triggering and regional nodes for traceability, failing to form a complete control loop from anomaly detection, tiered early warning, precise location, linkage response, and historical auditing, making it difficult to meet the standardized requirements of industrial-grade safety management.

[0009] To address this, a multi-parameter coupled positioning and hierarchical early warning system and method for thermal runaway of new energy batteries is proposed. Summary of the Invention

[0010] The purpose of this invention is to provide a multi-parameter coupled positioning and hierarchical early warning system and method for thermal runaway of new energy batteries, so as to solve the problems mentioned in the background art.

[0011] To achieve the above objectives, the present invention provides the following technical solution: a multi-parameter coupled positioning and hierarchical early warning system for thermal runaway of new energy batteries, including a front-end multi-parameter sensing module, an edge computing and transmission module, a back-end analysis and management module, and a display and control module, wherein each module interacts with data through a wired or wireless network;

[0012] The front-end multi-parameter sensing module includes a dual-light image acquisition unit, a thermal imaging temperature measurement unit, a visible light video acquisition unit, a gas sensing unit, a wireless acquisition unit, and a wireless gateway unit. The dual-light image acquisition unit is used to simultaneously acquire thermal imaging data and visible light video data of the same monitoring area. The gas sensing unit is used to acquire multiple types of characteristic gas parameters of battery thermal runaway. The wireless acquisition unit is used to assign wireless node identifiers to the gas sensing units and encapsulate the acquired data. The wireless gateway unit is used to aggregate data from at least one wireless acquisition unit and forward it to the edge computing and transmission module.

[0013] The edge computing and transmission module includes a temperature matrix analysis unit, a thermal imaging hierarchical early warning unit, a visible light anomaly identification unit, a wireless data receiving and packetizing unit, a data reporting unit, and a control execution unit. The temperature matrix analysis unit is used to extract temperature parameters and hotspot pixel location information from thermal imaging data. The thermal imaging hierarchical early warning unit is used to generate multi-level heat source early warnings, and the early warning information carries the coordinates of hotspot pixels. The visible light anomaly identification unit is used to detect local brightness anomalies in the monitoring area through pixel-by-pixel brightness time-series analysis. The wireless data receiving and packetizing unit is used to retain wireless node identifiers and gateway traceability information and encapsulate multi-parameter data. The control execution unit is used to receive and execute on-site control commands.

[0014] The background analysis and management module includes a regional node attribution unit, a sensor instance identification unit, a multi-parameter aggregation unit, a hotspot caching and pairing unit, a spatial inverse calculation model management unit, a fire source location management unit, and a control command queue unit. The regional node attribution unit is used to determine the monitoring area node to which the reported data belongs, based on the source IP of the edge industrial control computer as the priority matching basis. The sensor instance identification unit is used to distinguish independent sensor instances in the area based on the wireless node identifier after the regional node is determined. The hotspot caching and pairing unit is used to pair the hotspot coordinates of different cameras within a preset time window in the same monitoring area group. The spatial inverse calculation model management unit and the fire source location management unit are used to call the fully connected neural network model calibrated by the area to calculate the three-dimensional coordinates of the fire source from the dual-camera hotspot coordinates and generate the location confidence. The control command queue unit is used to store control commands and distribute the commands to be executed according to the regional node isolation.

[0015] The display and control module includes a comprehensive situation display unit, a three-dimensional area display unit, a multimodal monitoring unit, and a control authorization and remote control unit, which are used to display the situation of the monitoring area, the spatial positioning results of the fire source, and multi-parameter monitoring data, and generate control commands after authorization verification.

[0016] As a further preferred embodiment of this technical solution: the thermal imaging graded early warning unit is equipped with a four-level early warning mechanism:

[0017] The Level 1 warning is for abnormal battery temperature rise. The triggering condition is that the highest temperature in the monitored area reaches the abnormal heat source threshold, or the area of ​​the hot spot reaches the abnormal area threshold.

[0018] The Level II warning is a heat spread warning, which is triggered when the monitored temperature or hotspot area increases by a preset amount relative to the historical average within a sliding historical window.

[0019] Level 3 warning is a fire warning, triggered when the area of ​​pixels in the thermal imaging temperature matrix that reach the fire temperature threshold exceeds the minimum fire area threshold.

[0020] Level 4 warning is for fire expansion, triggered when the area of ​​the fire zone increases by a preset amount within a sliding historical window.

[0021] As a further preferred embodiment of this technical solution: the hotspot cache and pairing unit update the hotspot cache with the area node and camera number as joint conditions, retain only the latest hotspot record of the same camera, and clear the old cache whose record time exceeds the pairing time window;

[0022] During pairing, within the same monitoring area group, select another area node hotspot records with complementary camera numbers and a recording time difference not exceeding the pairing time window to form a dual-camera hotspot pair;

[0023] After successful pairing, a four-dimensional input vector is constructed according to the fixed camera order, and the two hotspot caches used in this pairing are deleted.

[0024] As a further preferred embodiment of this technical solution, the solution process of the spatial inverse calculation model management unit includes:

[0025] A1. Standardize the four-dimensional input vector composed of hotspot coordinates of the dual cameras by using the mean and standard deviation of each component input saved during the training phase.

[0026] A2. Input the standardized input vector into a multi-layer fully connected neural network, and output a standardized three-dimensional coordinate vector.

[0027] A3. Using the mean and standard deviation of each component output saved during the training phase, the standardized three-dimensional coordinate vector is de-standardized to obtain the actual three-dimensional coordinates of the fire source.

[0028] A4. Calculate the location reliability based on the average root mean square error of the X, Y, and Z axes obtained from model training. The smaller the average error, the higher the location reliability.

[0029] As a further preferred embodiment of this technical solution: the regional node attribution unit matches the pre-bound regional node records sequentially according to the priority order of the HTTP request source IP, request body IP address, and node number, and returns the first valid matching result;

[0030] If none of the three types of identifiers can be matched, refuse to save the business data and return an attribution failure message;

[0031] Among them, the control command polling request executes the same affiliation matching rule, and only returns the corresponding instruction to be executed to the matched region node.

[0032] As a further preferred embodiment of this technical solution: the sensor instance identification unit performs character normalization processing on the wireless node identifier to obtain the instance key, and generates an independent indicator instance identifier by combining the regional node number and the gas indicator type;

[0033] The multi-parameter aggregation unit uses the regional node number and instance key as the aggregation key to aggregate multiple gas indicators of the same wireless sensor node into the same device status.

[0034] The wireless gateway address is stored only as information for tracing the transmission link and is not used to determine the ownership of regional nodes.

[0035] As a further preferred embodiment of this technical solution, the detection process of the visible light anomaly recognition unit includes:

[0036] B1. Convert continuous visible light frames into grayscale frames and write them into a circular history buffer. Calculate the global brightness baseline for each frame and subtract it from the grayscale value of each pixel to suppress global brightness synchronization disturbance.

[0037] B2. Calculate the dynamic range, number of state transitions and observation frequency for the historical local brightness signal of each pixel, and filter candidate pixels that meet the amplitude, number of transitions and preset frequency range;

[0038] B3. Perform morphological expansion and connected component analysis on candidate pixels to filter out noise regions with too few pixels and global perturbation regions that occupy too large a proportion of the whole frame.

[0039] B4. Extract the brightness time sequence of representative pixels from the preserved connected regions. Calculate the flicker confidence level based on the half-cycle regularity, brightness change amplitude, number of effective transition edges, and duty cycle weighting. When the confidence level meets the standard, output the coordinates of the abnormal region and map them to the preset grid area.

[0040] As a further preferred embodiment of this technical solution: the fire source location management unit calculates the three-dimensional Euclidean distance between the newly generated fire source location and the existing active locations of the same area and detection type;

[0041] When the 3D Euclidean distance is less than the preset merging distance, update the coordinates, temperature, and confidence parameters of the original active positioning.

[0042] If no active location meets the criteria, create a new fire source location record.

[0043] Active locations that have not been updated beyond the preset expiration time are marked as expired.

[0044] As a further preferred embodiment of this technical solution: the control command queue unit initializes control requests generated by automatic linkage or manual authorization into a pending execution state;

[0045] When the edge industrial control computer polls for instructions, the regional node attribution unit re-determines the regional node based on the request source IP, and the control instruction queue unit only returns the pending instructions of the regional node and updates its status to executing.

[0046] After the edge industrial control computer executes the control action, it sends back the execution result. The background will associate and save the execution result with the corresponding regional node and the triggered alarm to form a closed loop of handling.

[0047] A multi-parameter coupled localization and hierarchical early warning method for thermal runaway in new energy batteries includes the following steps:

[0048] S1. Collect thermal imaging temperature matrix, visible light video data and multiple types of gas sensing data of the monitoring area, assign wireless node identifiers to the gas sensing data and aggregate and transmit them through the wireless gateway.

[0049] S2. Analyze the thermal imaging temperature matrix at the edge, and generate multi-level heat source early warning based on temperature threshold, hotspot area and historical growth pattern, carrying hotspot pixel coordinates;

[0050] Simultaneously, pixel-by-pixel brightness time-series analysis is performed on visible light video to generate early warning of local brightness anomalies;

[0051] S3. After receiving the reported data, the background determines the monitoring area node to which the data belongs based on the IP address of the source request from the edge industrial control computer.

[0052] Once the regional nodes are determined, the sensor instances within the region are distinguished based on the wireless node identifiers, and multiple gas parameters from the same wireless sensor node are aggregated.

[0053] S4. Within the same monitoring area group, pair dual-camera hotspot coordinates based on complementary camera numbers and preset time windows;

[0054] The paired hotspot coordinates are combined into a four-dimensional input vector, which is then standardized and input into a fully connected neural network with region calibration to calculate the three-dimensional coordinates of the fire source and generate location confidence.

[0055] S5. Perform spatial merging and expiration management on fire source location records, associate and manage multi-source alarms according to a unified structure, and establish a record association between spatial location results and visual alarms that trigger them;

[0056] S6. Based on the alarm linkage strategy or with manual authorization, generate control commands, distribute commands to be executed in isolation according to regional nodes, receive on-site execution results and save them in association, forming a complete closed loop from detection to handling.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] 1. This invention constructs a four-level progressive thermal imaging early warning system, combining multi-parameter cross-judgment to achieve extremely early identification of thermal runaway, reducing the risk of false alarms and missed alarms. Addressing the issues of delayed warnings from single-parameter thresholds, inability to match the progression of thermal runaway, and susceptibility to environmental interference, it sets up a four-level early warning mechanism for abnormal battery temperature rise, thermal diffusion, flame warning, and fire expansion. This accurately matches the full-stage development characteristics of battery thermal runaway from temperature rise initiation and heat diffusion to open flame generation and fire spread. Simultaneously, it integrates thermal imaging temperature characteristics, multiple types of characteristic gas parameters of battery thermal runaway, and local anomaly signals from visible light to form a multi-dimensional cross-validation system. This overcomes the limitations of single-parameter methods being susceptible to interference from environmental residual heat, ventilation conditions, and local obstruction. It can achieve extremely early warning during the abnormal temperature rise stage and provide a clear basis for graded response, effectively reducing the risk of false alarms and missed alarms.

[0059] 2. This invention employs a visible light anomaly detection scheme based on pixel-by-pixel brightness temporal analysis, significantly improving anti-interference capabilities in complex lighting scenarios and accurately capturing weak early signals. Addressing the issues of existing visible light flame identification being susceptible to interference from complex lighting, equipment reflections, and indicator light flicker, and the ease with which early, small-sized, low-amplitude anomalies are drowned out by environmental noise, this invention suppresses overall ambient lighting synchronization disturbances through global brightness baseline subtraction. Based on pixel-level historical brightness data, it calculates the dynamic range, state transition count, and observation frequency to filter candidate pixels. Morphological expansion and connected component analysis are combined to filter out minute noise and global disturbance regions. Finally, flicker confidence is calculated using a multi-dimensional weighted approach based on half-cycle regularity, brightness amplitude, edge count, and duty cycle. This technical solution effectively resists complex lighting interference in industrial environments, accurately identifies small-sized, low-amplitude local brightness anomalies, and significantly improves the detection rate and stability of early, weak anomaly signals.

[0060] 3. This invention employs a two-layer management mechanism that decouples regional affiliation from sensor instance management. It adapts to multi-level wireless transmission architectures, supports large-scale deployment and maintenance, and addresses the issues of inconsistent regional affiliation and chaotic sensor instance management in a large-scale architecture of "wireless sensor node—wireless gateway—edge industrial control computer—backend." It decouples the determination of monitoring region affiliation from the differentiation of sensor instances within the region. The source IP address of the edge industrial control computer is prioritized as the core basis for region affiliation determination. After the region is determined, independent sensor instances within the region are distinguished by the wireless node identifier, and multiple gas parameters from the same node are aggregated. The wireless gateway address is only used for transmission link tracing information and does not participate in affiliation determination. This mechanism avoids the problem of inconsistent regional affiliation when the same edge industrial control computer connects to multiple gateways and nodes. It can stably adapt to large-scale deployment scenarios with multiple gateways, nodes, and industrial control computers coexisting, significantly improving on-site deployment flexibility and maintenance efficiency.

[0061] 4. This invention presents a fire source localization scheme based on dual-camera hotspot pairing and regional calibration neural networks. It balances positioning accuracy with on-site adaptability, enabling quantifiable evaluation of positioning results. Addressing the issues of traditional binocular geometric positioning relying on fixed camera parameters, difficulty in correcting after on-site adjustments, high maintenance costs, and lack of confidence assessment, this invention pairs the hotspot coordinates of complementary cameras within the same monitoring area group through a preset time window. A fully connected neural network model with regional calibration is used to map pixel coordinates to three-dimensional spatial coordinates, and quantified location confidence is generated based on the root mean square error of the X / Y / Z axes. Simultaneously, it supports model version management, allowing for rapid updates to the calibration model based on on-site deployment adjustments. This technical solution eliminates the need for complex camera intrinsic and extrinsic parameter calibration. After adjustments to the monitoring area size and camera installation position, only re-collection of calibration data is required to update the model. It boasts low maintenance costs, strong on-site adaptability, and location confidence that directly reflects the reliability of the positioning results, achieving a balance between positioning accuracy and engineering adaptability.

[0062] 5. This invention constructs a node-based unified alarm management and regional isolation linkage control closed loop, meeting the standardized requirements of industrial-grade safety management. Addressing the issues of independent management of multi-source alarm data, disconnect between alarms and control, the risk of cross-regional erroneous execution, and the lack of a complete handling loop, this invention uniformly associates and manages thermal imaging warnings, gas warnings, visible light anomalies, and fire source location results by regional nodes. Control commands are stored and distributed in isolation by regional nodes. During command polling, the regional attribution is re-verified, and the execution results are associated with and retained with the corresponding alarm and regional node after being returned. This technical solution forms a complete management closed loop of anomaly detection, hierarchical early warning, precise location, linkage handling, and historical auditing, effectively avoiding the safety hazards of cross-regional command erroneous execution. Furthermore, the entire process is traceable and auditable, fully meeting the standardized and reliable requirements of industrial-grade battery safety management. Attached Figure Description

[0063] Figure 1This is a schematic diagram of the architecture of the new energy battery thermal runaway multi-parameter coupled positioning and hierarchical early warning system of the present invention;

[0064] Figure 2 This is a flowchart illustrating the operation of the multi-parameter coupled localization and hierarchical early warning method for thermal runaway in new energy batteries according to the present invention.

[0065] Figure 3 This is a flowchart of the solution process for the spatial inverse calculation model management unit in the multi-parameter coupled positioning and hierarchical early warning system for thermal runaway of new energy batteries of the present invention.

[0066] Figure 4 This is a flowchart of the visible light anomaly identification unit in the multi-parameter coupled positioning and hierarchical early warning system for thermal runaway of new energy batteries of the present invention. Detailed Implementation

[0067] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0068] Example 1

[0069] Please see Figures 1-4 The present invention provides a technical solution: a multi-parameter coupled positioning and hierarchical early warning system for thermal runaway of new energy batteries, including a front-end multi-parameter sensing module, an edge computing and transmission module, a back-end analysis and management module and a display and control module, wherein each module interacts with data through a wired or wireless network;

[0070] The front-end multi-parameter sensing module is used to collect thermal imaging data, visible light video data, and gas sensing data of the new energy battery monitoring area. The edge computing and transmission module is used to generate hierarchical visual warnings based on the thermal imaging temperature matrix, detect local brightness anomalies in visible light, receive wireless gas sensing data, and execute control commands. The back-end analysis and management module is used to determine the area nodes based on the source IP of the edge industrial control computer, distinguish sensor instances within the area nodes based on the wireless node identifier, pair dual-camera hotspot coordinates, perform three-dimensional spatial inverse calculations, and manage alarm, positioning, and control records. The display and control module is used to display node status, sensor data, spatial positioning results, and alarm status, and create control commands after automatic linkage or manual authorization.

[0071] The edge computing and transmission module carries the highest temperature pixel coordinates in the heat source early warning, and the background analysis and management module pairs the hotspot coordinates of two cameras in the same monitoring area group and within a preset time window, and obtains the three-dimensional coordinates of the fire source through the spatial inverse calculation model corresponding to the area node.

[0072] The front-end multi-parameter sensing module includes a dual-light image acquisition unit, a thermal imaging temperature measurement unit, a visible light video acquisition unit, a gas sensing unit, a wireless acquisition unit, and a wireless gateway unit. The dual-light image acquisition unit is used to simultaneously acquire thermal imaging data and visible light video data of the same monitoring area. The gas sensing unit is used to acquire multiple types (TVOC, CH2O, CO2) characteristic gas parameters of battery thermal runaway. The wireless acquisition unit is used to assign wireless node identifiers to the gas sensing unit and encapsulate the acquired data. The wireless gateway unit is used to aggregate data from at least one wireless acquisition unit and forward it to the edge computing and transmission module.

[0073] The thermal imaging temperature measurement unit is used to output the temperature matrix and the corresponding pixel positions.

[0074] The edge computing and transmission module includes a temperature matrix analysis unit, a thermal imaging hierarchical early warning unit, a visible light anomaly identification unit, a wireless data receiving and packetizing unit, a data reporting unit, and a control execution unit. The temperature matrix analysis unit is used to extract temperature parameters (maximum temperature, minimum temperature, and average temperature), hotspot pixel location information, and area from thermal imaging data. The thermal imaging hierarchical early warning unit is used to generate multi-level heat source early warnings, and the early warning information carries the coordinates of hotspot pixels. The visible light anomaly identification unit is used to detect local brightness anomalies in the monitoring area through pixel-by-pixel brightness time-series analysis. The wireless data receiving and packetizing unit is used to retain wireless node identifiers and gateway traceability information and encapsulate multi-parameter data. The control execution unit is used to receive and execute on-site control commands.

[0075] The edge computing and transmission module also includes a video streaming unit, which is used to transmit thermal imaging frames and visible light frames, and a data reporting unit is used to transmit sensor data, visual warnings and heartbeat information to the background.

[0076] The background analysis and management module includes a regional node attribution unit, a sensor instance identification unit, a multi-parameter aggregation unit, a hotspot caching and pairing unit, a spatial inverse calculation model management unit, a fire source location management unit, and a control command queue unit. The regional node attribution unit is used to determine the monitoring area node to which the reported data belongs, based on the source IP of the edge industrial control computer as the priority matching basis. The sensor instance identification unit is used to distinguish independent sensor instances in the area based on the wireless node identifier after the regional node is determined. The hotspot caching and pairing unit is used to pair the hotspot coordinates of different cameras within a preset time window in the same monitoring area group. The spatial inverse calculation model management unit and the fire source location management unit are used to call the fully connected neural network model calibrated by the area, calculate the three-dimensional coordinates of the fire source from the dual-camera hotspot coordinates, and generate the location confidence. The control command queue unit is used to store control commands and distribute the commands to be executed according to the regional node isolation.

[0077] The background analysis and management module also includes an alarm receiving and association unit and a historical analysis unit, and the multi-parameter aggregation unit is used to organize different gas indicators of the same wireless sensor node into the same device status.

[0078] The display and control module includes a comprehensive situation display unit, a three-dimensional area display unit, a multi-modal monitoring unit, and a control authorization and remote control unit, which are used to display the situation of the monitoring area, the spatial positioning results of the fire source, and multi-parameter monitoring data, and generate control commands after authorization verification;

[0079] The integrated situation display unit is used to display the overall status of regional nodes, sensors and alarms; the three-dimensional regional display unit is used to display the fire source location coordinates; the multimodal monitoring unit is used to display infrared images, visible light images and gas parameters in conjunction; the control authorization unit is used to verify manual control permissions; and the remote control unit is used to create switch control commands and display the execution results.

[0080] In this embodiment, specifically: the thermal imaging graded early warning unit is equipped with a four-level early warning mechanism:

[0081] The Level 1 warning is for abnormal battery temperature rise. The triggering condition is that the highest temperature in the monitored area reaches the abnormal heat source threshold, or the area of ​​the hot spot reaches the abnormal area threshold.

[0082] The Level II warning is a heat spread warning, which is triggered when the monitored temperature or hotspot area increases by a preset amount relative to the historical average within a sliding historical window.

[0083] Level 3 warning is a fire warning, triggered when the area of ​​pixels in the thermal imaging temperature matrix that reach the fire temperature threshold exceeds the minimum fire area threshold.

[0084] Level 4 warning is for fire expansion, triggered when the area of ​​the fire zone increases by a preset amount within a sliding historical window.

[0085] In this embodiment, specifically: the hotspot cache and pairing unit updates the hotspot cache with the area node and camera number as joint conditions, retains only the latest hotspot record of the same camera, and clears the old cache whose record time exceeds the pairing time window;

[0086] During pairing, within the same monitoring area group, select another area node hotspot records with complementary camera numbers and a recording time difference not exceeding the pairing time window to form a dual-camera hotspot pair;

[0087] After successful pairing, a four-dimensional input vector is constructed according to the fixed camera order, and the two hotspot caches used in this pairing are deleted.

[0088] In this embodiment, the specific solution process of the spatial inverse calculation model management unit includes:

[0089] A1. Standardize the four-dimensional input vector composed of hotspot coordinates of the dual cameras by using the mean and standard deviation of each component input saved during the training phase.

[0090] A2. Input the standardized input vector into a multi-layer fully connected neural network, and output a standardized three-dimensional coordinate vector.

[0091] A3. Using the mean and standard deviation of each component output saved during the training phase, the standardized three-dimensional coordinate vector is de-standardized to obtain the actual three-dimensional coordinates of the fire source.

[0092] A4. Calculate the location reliability based on the average root mean square error of the X, Y, and Z axes obtained from model training. The smaller the average error, the higher the location reliability.

[0093] In this embodiment, specifically: the regional node attribution unit matches the pre-bound regional node records in the order of priority of HTTP request source IP, request body IP address, and node number, and returns the first valid matching result;

[0094] If none of the three types of identifiers can be matched, refuse to save the business data and return an attribution failure message;

[0095] Among them, the control command polling request executes the same affiliation matching rule, and only returns the corresponding instruction to be executed to the matched region node.

[0096] In this embodiment, specifically: the sensor instance identification unit performs character normalization processing on the wireless node identifier to obtain the instance key, and generates an independent indicator instance identifier by combining the regional node number and the gas indicator type;

[0097] The multi-parameter aggregation unit uses the region node number and instance key as the aggregation key to aggregate multiple gas indicators of the same wireless sensor node into the same device status.

[0098] The wireless gateway address is stored only as information for tracing the transmission link and is not used to determine the ownership of regional nodes.

[0099] In this embodiment, the specific detection process of the visible light anomaly recognition unit includes:

[0100] B1. Convert continuous visible light frames into grayscale frames and write them into a circular history buffer. Calculate the global brightness baseline for each frame and subtract it from the grayscale value of each pixel to suppress global brightness synchronization disturbance.

[0101] B2. Calculate the dynamic range, number of state transitions and observation frequency for the historical local brightness signal of each pixel, and filter candidate pixels that meet the amplitude, number of transitions and preset frequency range;

[0102] B3. Perform morphological expansion and connected component analysis on candidate pixels to filter out noise regions with too few pixels and global perturbation regions that occupy too large a proportion of the whole frame.

[0103] B4. Extract the brightness time sequence of representative pixels from the preserved connected regions. Calculate the flicker confidence level based on the half-cycle regularity, brightness change amplitude, number of effective transition edges, and duty cycle weighting. When the confidence level meets the standard, output the coordinates of the abnormal region and map them to the preset grid area.

[0104] In this embodiment, specifically: the fire source location management unit calculates the three-dimensional Euclidean distance between the newly generated fire source location and the existing active locations of the same area and detection type;

[0105] When the 3D Euclidean distance is less than the preset merging distance, update the coordinates, temperature, and confidence parameters of the original active positioning.

[0106] If no active location meets the criteria, create a new fire source location record.

[0107] Active locations that have not been updated beyond the preset expiration time are marked as expired.

[0108] In this embodiment, specifically: the control command queue unit initializes the control requests generated by automatic linkage or manual authorization to a pending execution state;

[0109] When the edge industrial control computer polls for instructions, the regional node attribution unit re-determines the regional node based on the IP address of the request source, and the control instruction queue unit only returns the instructions to be executed for that regional node and updates its status to "in execution".

[0110] After the edge industrial control computer executes the control action, it sends back the execution result. The background will associate and save the execution result with the corresponding regional node and the triggered alarm to form a closed loop of handling.

[0111] A multi-parameter coupled localization and hierarchical early warning method for thermal runaway in new energy batteries includes the following steps:

[0112] S1. Collect thermal imaging temperature matrix, visible light video data and multiple types of gas sensing data of the monitoring area, assign wireless node identifiers to the gas sensing data and aggregate and transmit them through the wireless gateway.

[0113] S2. Analyze the thermal imaging temperature matrix at the edge, and generate multi-level heat source early warning based on temperature threshold, hotspot area and historical growth pattern, carrying hotspot pixel coordinates;

[0114] Simultaneously, pixel-by-pixel brightness time-series analysis is performed on visible light video to generate early warning of local brightness anomalies;

[0115] S3. After receiving the reported data, the background determines the monitoring area node to which the data belongs based on the IP address of the source request from the edge industrial control computer.

[0116] Once the regional nodes are determined, the sensor instances within the region are distinguished based on the wireless node identifiers, and multiple gas parameters from the same wireless sensor node are aggregated.

[0117] S4. Within the same monitoring area group, pair dual-camera hotspot coordinates based on complementary camera numbers and preset time windows;

[0118] The paired hotspot coordinates are combined into a four-dimensional input vector, which is then standardized and input into a fully connected neural network with region calibration to calculate the three-dimensional coordinates of the fire source and generate location confidence.

[0119] S5. Perform spatial merging and expiration management on fire source location records, associate and manage multi-source alarms according to a unified structure, and establish a record association between spatial location results and visual alarms that trigger them;

[0120] S6. Based on the alarm linkage strategy or with manual authorization, generate control commands, distribute commands to be executed in isolation according to regional nodes, receive on-site execution results and save them in association, forming a complete closed loop from detection to handling.

[0121] Example 2

[0122] The new energy battery monitoring area is equipped with a front-end multi-parameter sensing module, which includes two dual-light cameras, each of which has a visible light channel and an infrared thermal imaging channel.

[0123] The first camera corresponds to the first back-end area node, and the second camera corresponds to the second back-end area node. The two area nodes are set to the same monitoring area group.

[0124] The edge industrial control computer performs video streaming, temperature matrix analysis, visible light anomaly identification, wireless data reception and packetization, data reporting, and control program execution.

[0125] The edge industrial control computer traverses the effective temperature points in the thermal imaging temperature matrix and counts the highest temperature, lowest temperature, average temperature, the pixel coordinates of the highest temperature, and the pixel areas that exceed the hot spot threshold and the flame threshold.

[0126] When the highest temperature or the area of ​​a hot spot reaches the first-level condition, a first-level visual warning shall be reported.

[0127] When the current temperature or area reaches a multiple of the historical average value of the sliding window, a level-two visual warning is reported.

[0128] When the area of ​​a pixel exceeding the flame temperature threshold reaches the minimum flame area, a level three warning is reported.

[0129] When the flame area increases by a multiple relative to the historical average, a Level IV warning should be issued.

[0130] The Level 1 and Level 2 warnings include the horizontal and vertical percentage coordinates of the highest temperature pixel.

[0131] After receiving a Level 1 or Level 2 alert from any camera, the backend writes or updates the hotspot cache based on the area node and camera number, and deletes the old cache that has been paired for more than 8 seconds.

[0132] The backend searches for the latest hotspot records of another node and another camera number within the same monitoring area group;

[0133] When the time difference between two hotspot records does not exceed 8 seconds, the input vector is formed according to the fixed order of the first camera and the second camera. And delete the two caches used after successful pairing.

[0134] The spatial inverse calculation model is trained using calibration data containing the fields cam1_x, cam1_y, ​​cam2_x, cam2_y, X, Y, and Z.

[0135] During training, the four-dimensional input and three-dimensional output are standardized separately, and the mean squared error is used as the loss function to train fully connected networks with layer widths of 4, 64, 128, 256, 128, 64 and 3.

[0136] During inference, the input is first standardized according to p_tilde_j=(p_j-mu_pj) / sigma_pj, and then r_tilde is obtained through multiple mappings. Finally, the three-dimensional coordinates are obtained according to r_k=sigma_rk*r_tilde_k+mu_rk. .

[0137] After training, the backend calculates RMSE_X, RMSE_Y, and RMSE_Z based on the predicted and true coordinates of each calibrated sample, and calculates the location reliability using C=clip(1 / (1+10*RMSE_bar),0,1), where RMSE_bar=(RMSE_X+RMSE_Y+RMSE_Z) / 3.

[0138] The backend calculates the Euclidean distance between the three-dimensional coordinates and the active localization of the same node and the same detection type.

[0139] Update the existing record if the distance is less than 2 meters; otherwise, create a new record.

[0140] Active locations are marked as expired if they have not been updated for more than 30 seconds.

[0141] The above time and distance parameters can be adjusted according to the site conditions.

[0142] Multiple gas sensing nodes are connected to an edge industrial control computer via one or more wireless gateways.

[0143] The edge industrial control computer reports TVOC, CH2O, CO2, wireless node identifier detector_meta, and gateway traceability information to the backend.

[0144] The backend first extracts the source IP from the request connection information and then queries it in the pre-bound region node;

[0145] The ip_address and node_id in the request body are used sequentially only if the source IP cannot be matched.

[0146] If none of the three identifiers match, the backend will refuse to save the data.

[0147] After the regional nodes are determined, the backend normalizes detector_meta to obtain the instance key K, and creates N.node_id||TVOC||K, N.node_id||CH2O||K and N.node_id||CO2||K index instances respectively.

[0148] When demonstrating, (N.node_id,K) is used as the aggregation key to combine the three types of indicators into a wireless gas sensing node.

[0149] The wireless gateway IP is only stored as transmission path tracing information and does not participate in regional node matching.

[0150] The edge industrial control computer converts visible light frames into grayscale frames and writes them into a 90-frame circular buffer. It calculates the global grayscale median for each frame and subtracts the median from the history of each pixel.

[0151] The system calculates the dynamic range, number of state transitions, and observation frequency for each pixel, and forms candidate images from pixels that meet the amplitude, number of transitions, and frequency range.

[0152] After morphological expansion and eight-neighbor connectivity analysis, candidate images are filtered out for small noise regions and global perturbation regions that occupy too large a proportion of the whole frame.

[0153] For each preserved region, the system selects the representative pixel with the largest dynamic range, constructs a threshold using the 20% and 80% quantile values ​​of the brightness history, extracts the brightness transition edge, and calculates f=1 / (2*T_half) based on the median of the edge interval.

[0154] The system further calculates the confidence level according to C_f=0.45R+0.30B+0.15E_s+0.10D;

[0155] Once the confidence level reaches the threshold, the weighted center of the candidate region is mapped to a preset grid. The grid is used only for displaying classification information, and pixel coordinates, bounding box, frequency, peak brightness, number of jumps, and confidence level are reported to the backend.

[0156] Visual alarms, gas alarms, and visible light anomalies are stored using a unified field structure that includes area nodes, event types, alarm levels, handling status, and occurrence time. Spatial positioning results are linked with the visual alarms that triggered the positioning, and can be queried jointly by area nodes and time ranges.

[0157] When the linkage strategy is met, the background automatically creates a switch control command;

[0158] When operating manually, the display and control module first verifies the control password and then creates a manual control command.

[0159] The control commands are initially in a pending state.

[0160] When the edge industrial control computer polls, the background system redetermines the regional node based on the IP address of the request source, returns only the pending instructions for that node, and sets the instructions that have already been returned to be executed.

[0161] After the edge industrial control computer executes the field switch action, it sends back the instruction number, success or failure status and result information. The background saves the execution time and associates the result with the node and alarm history.

[0162] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-parameter coupled positioning and hierarchical early warning system for thermal runaway in new energy batteries, characterized in that, It includes a front-end multi-parameter perception module, an edge computing and transmission module, a back-end analysis and management module, and a display and control module. Each module interacts with data through wired or wireless networks. The front-end multi-parameter sensing module includes a dual-light image acquisition unit, a thermal imaging temperature measurement unit, a visible light video acquisition unit, a gas sensing unit, a wireless acquisition unit, and a wireless gateway unit. The dual-light image acquisition unit is used to simultaneously acquire thermal imaging data and visible light video data of the same monitoring area. The gas sensing unit is used to acquire multiple types of characteristic gas parameters of battery thermal runaway. The wireless acquisition unit is used to assign wireless node identifiers to the gas sensing units and encapsulate the acquired data. The wireless gateway unit is used to aggregate data from at least one wireless acquisition unit and forward it to the edge computing and transmission module. The edge computing and transmission module includes a temperature matrix analysis unit, a thermal imaging hierarchical early warning unit, a visible light anomaly identification unit, a wireless data receiving and packetizing unit, a data reporting unit, and a control execution unit. The temperature matrix analysis unit is used to extract temperature parameters and hotspot pixel location information from thermal imaging data. The thermal imaging hierarchical early warning unit is used to generate multi-level heat source early warnings, and the early warning information carries the coordinates of hotspot pixels. The visible light anomaly identification unit is used to detect local brightness anomalies in the monitoring area through pixel-by-pixel brightness time-series analysis. The wireless data receiving and packetizing unit is used to retain wireless node identifiers and gateway traceability information and encapsulate multi-parameter data. The control execution unit is used to receive and execute on-site control commands. The background analysis and management module includes a regional node attribution unit, a sensor instance identification unit, a multi-parameter aggregation unit, a hotspot caching and pairing unit, a spatial inverse calculation model management unit, a fire source location management unit, and a control command queue unit. The regional node attribution unit is used to determine the monitoring area node to which the reported data belongs, based on the source IP of the edge industrial control computer as the priority matching basis. The sensor instance identification unit is used to distinguish independent sensor instances in the area based on the wireless node identifier after the regional node is determined. The hotspot caching and pairing unit is used to pair the hotspot coordinates of different cameras within a preset time window in the same monitoring area group. The spatial inverse calculation model management unit and the fire source location management unit are used to call the fully connected neural network model calibrated by the area to calculate the three-dimensional coordinates of the fire source from the dual-camera hotspot coordinates and generate the location confidence. The control command queue unit is used to store control commands and distribute the commands to be executed according to the regional node isolation. The display and control module includes a comprehensive situation display unit, a three-dimensional area display unit, a multimodal monitoring unit, and a control authorization and remote control unit, which are used to display the situation of the monitoring area, the spatial positioning results of the fire source, and multi-parameter monitoring data, and generate control commands after authorization verification.

2. The new energy battery thermal runaway multi-parameter coupled positioning and hierarchical early warning system according to claim 1, characterized in that: The thermal imaging hierarchical early warning unit is equipped with a four-level early warning mechanism: The Level 1 warning is for abnormal battery temperature rise. The triggering condition is that the highest temperature in the monitored area reaches the abnormal heat source threshold, or the area of ​​the hot spot reaches the abnormal area threshold. The Level II warning is a heat spread warning, which is triggered when the monitored temperature or hotspot area increases by a preset amount relative to the historical average within a sliding historical window. Level 3 warning is a fire warning, triggered when the area of ​​pixels in the thermal imaging temperature matrix that reach the fire temperature threshold exceeds the minimum fire area threshold. Level 4 warning is for fire expansion, triggered when the area of ​​the fire zone increases by a preset amount within a sliding historical window.

3. The new energy battery thermal runaway multi-parameter coupled positioning and hierarchical early warning system according to claim 1, characterized in that: The hotspot cache and pairing unit update the hotspot cache using the area node and camera number as joint conditions, retaining only the latest hotspot record of the same camera and clearing the old cache whose record time exceeds the pairing time window; During pairing, within the same monitoring area group, select another area node hotspot records with complementary camera numbers and a recording time difference not exceeding the pairing time window to form a dual-camera hotspot pair; After successful pairing, a four-dimensional input vector is constructed according to the fixed camera order, and the two hotspot caches used in this pairing are deleted.

4. The new energy battery thermal runaway multi-parameter coupled positioning and hierarchical early warning system according to claim 1, characterized in that: The solution process of the spatial inverse calculation model management unit includes: A1. Standardize the four-dimensional input vector composed of hotspot coordinates of the dual cameras by using the mean and standard deviation of each component input saved during the training phase. A2. Input the standardized input vector into a multi-layer fully connected neural network, and output a standardized three-dimensional coordinate vector. A3. Using the mean and standard deviation of each component output saved during the training phase, the standardized three-dimensional coordinate vector is de-standardized to obtain the actual three-dimensional coordinates of the fire source. A4. Calculate the location reliability based on the average root mean square error of the X, Y, and Z axes obtained from model training. The smaller the average error, the higher the location reliability.

5. The new energy battery thermal runaway multi-parameter coupled positioning and hierarchical early warning system according to claim 1, characterized in that: The regional node attribution unit matches the pre-bound regional node records sequentially according to the priority order of HTTP request source IP, request body IP address, and node number, and returns the first valid matching result; If none of the three types of identifiers can be matched, refuse to save the business data and return an attribution failure message; Among them, the control command polling request executes the same affiliation matching rule, and only returns the corresponding instruction to be executed to the matched region node.

6. The new energy battery thermal runaway multi-parameter coupled positioning and hierarchical early warning system according to claim 1, characterized in that: The sensor instance identification unit performs character normalization processing on the wireless node identifier to obtain the instance key, and generates an independent indicator instance identifier by combining the area node number and the gas indicator type. The multi-parameter aggregation unit uses the regional node number and instance key as the aggregation key to aggregate multiple gas indicators of the same wireless sensor node into the same device status. The wireless gateway address is stored only as information for tracing the transmission link and is not used to determine the ownership of regional nodes.

7. The new energy battery thermal runaway multi-parameter coupled positioning and hierarchical early warning system according to claim 1, characterized in that: The detection process of the visible light anomaly detection unit includes: B1. Convert continuous visible light frames into grayscale frames and write them into a circular history buffer. Calculate the global brightness baseline for each frame and subtract it from the grayscale value of each pixel to suppress global brightness synchronization disturbance. B2. Calculate the dynamic range, number of state transitions and observation frequency for the historical local brightness signal of each pixel, and filter candidate pixels that meet the amplitude, number of jumps and preset frequency range; B3. Perform morphological expansion and connected component analysis on candidate pixels to filter out noise regions with too few pixels and global perturbation regions that occupy too large a proportion of the whole frame. B4. Extract the brightness time sequence of representative pixels from the preserved connected regions. Calculate the flicker confidence level based on the half-cycle regularity, brightness change amplitude, number of effective transition edges, and duty cycle weighting. When the confidence level meets the standard, output the coordinates of the abnormal region and map them to the preset grid area.

8. The new energy battery thermal runaway multi-parameter coupled positioning and hierarchical early warning system according to claim 1, characterized in that: The fire source location management unit calculates the three-dimensional Euclidean distance between the newly generated fire source location and the existing active locations of the same area and detection type. When the 3D Euclidean distance is less than the preset merging distance, update the coordinates, temperature, and confidence parameters of the original active positioning. If no active location meets the criteria, create a new fire source location record. Active locations that have not been updated beyond the preset expiration time are marked as expired.

9. The multi-parameter coupled positioning and hierarchical early warning system for thermal runaway of new energy batteries according to claim 1, characterized in that: The control command queue unit initializes control requests generated automatically or manually as pending execution. When the edge industrial control computer polls for instructions, the regional node attribution unit re-determines the regional node based on the request source IP, and the control instruction queue unit only returns the pending instructions of the regional node and updates its status to executing. After the edge industrial control computer executes the control action, it sends back the execution result. The background will associate and save the execution result with the corresponding regional node and the triggered alarm to form a closed loop of handling.

10. A multi-parameter coupled localization and hierarchical early warning method for thermal runaway in new energy batteries, characterized in that, Includes the following steps: S1. Collect thermal imaging temperature matrix, visible light video data and multiple types of gas sensing data of the monitoring area, assign wireless node identifiers to the gas sensing data and aggregate and transmit them through the wireless gateway. S2. Analyze the thermal imaging temperature matrix at the edge, and generate multi-level heat source early warning based on temperature threshold, hotspot area and historical growth pattern, carrying hotspot pixel coordinates; Simultaneously, pixel-by-pixel brightness time-series analysis is performed on visible light video to generate early warning of local brightness anomalies; S3. After receiving the reported data, the background determines the monitoring area node to which the data belongs based on the IP address of the source request from the edge industrial control computer. Once the regional nodes are determined, the sensor instances within the region are distinguished based on the wireless node identifiers, and multiple gas parameters from the same wireless sensor node are aggregated. S4. Within the same monitoring area group, pair dual-camera hotspot coordinates based on complementary camera numbers and preset time windows; The paired hotspot coordinates are combined into a four-dimensional input vector, which is then standardized and input into a fully connected neural network with region calibration to calculate the three-dimensional coordinates of the fire source and generate location confidence. S5. Perform spatial merging and expiration management on fire source location records, associate and manage multi-source alarms according to a unified structure, and establish a record association between spatial location results and visual alarms that trigger them; S6. Based on the alarm linkage strategy or with manual authorization, generate control commands, distribute commands to be executed in isolation according to regional nodes, receive on-site execution results and save them in association, forming a complete closed loop from detection to handling.

Images