Ternary IoT Battery Management System

Abstract

The present invention discloses a ternary Internet of Things battery management system, which specifically relates to the technical field of lithium battery management. It realizes closed-loop thermal safety management of lithium batteries through multi-sensor fusion, including: deploying a digital sensor array to obtain spatiotemporal correlation data; constructing a spatiotemporal feature vector including the fusion of local temperature gradient, neighborhood temperature gradient and stress coupling ratio; using a density clustering algorithm to dynamically divide self-heating, radiant heat and stress coupling clusters, and updating cluster boundaries in real time; calculating dual thermal coupling coefficients based on clustering results, and eliminating interference from temperature readings through a dual temperature compensation formula; establishing a three-level alarm mechanism for radiant heat proportion, stress coupling degree and compensation error, and linking the side gateway to realize parameter adaptive optimization, thereby improving the thermal runaway warning accuracy and structural failure detection rate; and realizing strain monitoring resource optimization through physical-driven key point screening and dynamic graph neural network prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery management, and more specifically, to a ternary IoT battery management system. Background Art

[0002] In high-rate charge and discharge scenarios, not only does the battery module experience complex temperature coupling due to electrochemical reactions and internal resistance heat generation, but it also experiences mechanical stress due to thermal expansion of the outer shell and adhesive, leading to voltage measurement drift. Accurately separating and compensating for these dual coupling effects can not only eliminate misjudgments of temperature and voltage readings, but also improve the accuracy of health diagnosis and safety warnings, thereby significantly enhancing the reliability of ternary IoT lithium battery management and the accuracy of life prediction.

[0003] Existing technologies usually partially alleviate temperature drift or voltage drift under static conditions through periodic calibration or single-source filtering, or separate single coupling effects on a simulation platform, but it is difficult to meet the needs of accurate decoupling and compensation of online real-time, dual-source coupling errors. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a ternary IoT battery management system, which jointly analyzes temperature gradient and stress data, synchronously separates thermal coupling and mechanical coupling components, and realizes real-time correction and accurate early warning of temperature and voltage readings under high-rate working conditions.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: a ternary IoT battery management system, comprising:

[0006] The sensor data acquisition module numbers the lithium batteries by battery block and deploys digital temperature sensors and digital strain gauge sensors respectively, outputting temperature and stress readings with timestamps and coordinates. The stress readings reflect the changes in thermal expansion stress between the battery casing and the adhesive.

[0007] The feature construction module calculates the local temperature gradient, neighborhood temperature gradient, spatial gradient ratio, and stress coupling ratio characteristics of the battery block based on temperature and stress readings, and constructs a spatiotemporal feature vector by combining spatial weights and gradient variance;

[0008] The cluster analysis module performs cluster analysis on spatiotemporal feature vectors based on clustering parameters, including a cluster radius threshold and a minimum neighborhood point threshold. The minimum neighborhood point threshold is used to define the density standard for cluster formation. The module divides the spatiotemporal feature vectors into self-heating clusters, radiant heat clusters, and stress coupling clusters, and updates cluster centers and cluster boundaries in real time.

[0009] The temperature compensation calculation module calculates the characteristic gradient means of the self-heating cluster, the radiant heat cluster, and the stress coupling cluster based on the clustering results, and calculates the first thermal coupling coefficient and the second thermal coupling coefficient. The first thermal coupling coefficient and the second thermal coupling coefficient are filtered using the exponential moving average method and used to make real-time corrections to the temperature reading of battery block i.

[0010] The alarm management module triggers a local alarm and sends temperature compensation parameters and clustering parameters to the edge gateway to achieve remote dynamic adaptive closed loop when the proportion of radiation heat clusters, stress coupling cluster proportions or temperature compensation error exceeds the threshold.

[0011] Preferably, the feature construction module also includes a dynamic aging factor calculation unit, which generates an aging coefficient based on the number of battery cycles and the capacity attenuation rate, applies aging weight compensation to the neighborhood temperature gradient, and the aging coefficient is calibrated online through electrochemical impedance spectroscopy.

[0012] Preferably, the cluster radius threshold is obtained as follows:

[0013] Through online or periodic offline thermal imaging, the steady-state thermal conductivity of any two adjacent battery blocks is calculated. The steady-state thermal conductivity represents the ratio of the heat flow per unit time from battery block i to battery block j to the temperature difference between the two under steady-state conditions. The effective thermal coupling distance between battery blocks is defined as the inverse of the steady-state thermal conductivity. When clustering, the effective thermal coupling distance is used instead of the geometric distance to calculate the neighborhood range of each pair of nodes, so that the cluster radius threshold can be adaptively expanded and contracted according to the thermal coupling strength, and the cluster center and cluster boundaries are updated. The cluster center refers to the mean vector of all spatiotemporal eigenvectors in the cluster, and the cluster boundary is defined by the cluster radius threshold.

[0014] Preferably, the digital twin model is used to simulate the multi-physics field thermal coupling matrix under the current working condition in real time, and the simulated steady-state thermal conductivity is dynamically output; the cluster radius threshold is adjusted based on the steady-state thermal conductivity average value and standard deviation; a group of steady-state thermal conductivity samples are measured by thermal imaging several times in each group of adjacent battery blocks; the average value is calculated and standard deviation ; Adjust the cluster radius threshold using the following formula :

[0015]

[0016] in, is the experience adjustment coefficient, The value is 0.001 to prevent the denominator from being 0.

[0017] Preferably, the system includes a clustering enhancement module for forcing the clustering results to conform to Fourier's heat conduction law, suppressing abnormal cluster splitting caused by measurement noise, and improving clustering robustness under extreme working conditions by defining a physical enhancement distance function in the DBSCAN algorithm; the clustering enhancement module includes a residual calculation unit, a distance fusion unit, and a clustering determination optimization unit; the operation process of the clustering enhancement module includes the following steps:

[0018] The residual calculation unit performs discrete Laplace operator calculation on the temperature field data in the candidate neighborhood set, which refers to the battery blocks within the cluster radius threshold, to generate the thermal conduction residual. ;in, represents the temperature change rate of the i-th battery block, Indicates the thermal diffusivity of battery materials, calibrated by differential scanning calorimetry, represents the temperature of the battery block adjacent to battery block i, represents the geometric center distance between battery block k and battery block i;

[0019] The distance fusion unit transforms the spatiotemporal feature vector The Euclidean distance is weightedly fused with the heat conduction residual to generate the physical enhancement distance ;in, represents the spatial distance between battery block i and battery j; Represents an adjustable weight coefficient; represents the thermal conduction residual of battery block j;

[0020] The cluster determination optimization unit uses the physical enhancement distance as the updated cluster radius threshold.

[0021] Preferably, the system also includes: a strain sensing management module, which obtains the position weight of each battery block through principal component analysis of the thermal coupling matrix, screens and outputs key battery blocks based on the position weight, and matches real strain sensors for key battery blocks; matches stress prediction models for non-critical battery blocks, outputs the stress of non-critical battery blocks through the stress prediction model, and activates real strain gauge calibration when the prediction confidence is lower than the preset requirement.

[0022] Preferably, the key battery block screening process includes a vibration sensitivity correction unit, which corrects the position weight based on the vibration energy integral obtained by the acceleration sensor, and the vibration energy integral is calibrated by a material fatigue test, including:

[0023] In order to deal with the material fatigue risk caused by vehicle vibration, the triaxial acceleration sensor data is introduced , after retaining the energy of the resonance frequency band, calculate the vibration energy integral ;

[0024] Based on vibration energy integration and vibration coupling coefficient Corrected position weight ; Represents the corrected position weight, according to Select key battery blocks in descending order.

[0025] Preferably, the screening process of the key battery blocks includes the following steps:

[0026] Obtain the thermal coupling matrix, perform singular value decomposition, extract the first k principal component directions based on Pareto, and generate a projection vector level. Use m to represent the index of the principal component direction. Calculate the position weight of the battery block using the following formula:

[0027]

[0028] in, Indicates the dominant direction of heat flow transfer in the mth principal direction, Represents the geometric center coordinates of battery block i;

[0029] The key battery blocks are screened based on the position weight, which reflects the projection intensity of the battery block in the main direction of heat conduction.

[0030] Preferably, the stress prediction model is constructed in the following manner:

[0031] Step 1: Obtain the real stress time series data, temperature spatiotemporal series data, battery block position weights, and battery operating status parameters of key battery blocks; use a dynamic sliding window to extract spatiotemporal coupling features and output a multidimensional feature vector;

[0032] Step 2: Using the measured stress of key battery blocks as labels, a training set is constructed based on multidimensional feature vectors. A time-series graph convolutional network is constructed using position weights as edge weights and physical connections as topology. Dynamic characteristics are captured using gated recurrent units. A two-layer loss function is defined during the training process: the main loss is the Huber loss, and the constraint term is the physical consistency penalty term.

[0033] Step 3: Physical consistency verification and edge adaptive deployment: Use the confidence of the stress prediction model and the material yield strength violation rate to quantify the prediction reliability. When the prediction reliability meets the requirements, the trained stress prediction model is deployed and applied.

[0034] Preferably, the stress prediction model includes a real-time graph topology generation engine, which dynamically reconstructs the graph neural network connection edges according to the temperature gradient vector field, and the heat flux intensity threshold is adaptively adjusted according to the local temperature change rate, including:

[0035] At each sampling moment, based on the battery block Calculate the three-dimensional temperature gradient vector based on the spatial temperature distribution , forming a full-envelope temperature gradient vector field;

[0036] Define the local temperature change rate As a heat flow intensity adjustment factor, dynamically adjust the connection edge judgment threshold ;in Indicates the basic heat flux intensity threshold, It represents the sensitivity coefficient of temperature change rate;

[0037] Traverse all battery block pairs , calculate the gradient direction similarity , Indicates battery pack The spatial temperature distribution of the three-dimensional temperature gradient vector is calculated. When establishing connection edges in graph neural networks ; When the gradient is greater than 0 and in the same direction, it indicates that the heat flow is enhanced and a connection is established; when the gradient is less than 0 and in the opposite direction, no connection is established;

[0038] The topology is updated every frame and input into the temporal graph convolutional network to reduce the prediction error.

[0039] Technical effects and advantages of the present invention:

[0040] (1) The embodiment of the present invention replaces the geometric topology with thermodynamic topology to solve the problem of missed detection of cross-module thermal coupling caused by the limitation of geometric layout in traditional clustering; the pulse excitation method is used to obtain thermal conductivity online, and the neighborhood relationship is reconstructed based on the definition of the effective distance of thermal coupling based on the inverse of thermal conductivity, so that the neighborhood radius of the high thermal conductivity area is significantly expanded; the clustering parameters are dynamically generated by combining the main direction analysis of heat conduction to capture the long-range thermal radiation path that cannot be identified by traditional methods; the thermodynamic consistency within the cluster is verified by physically enhancing the distance, eliminating the abnormal cluster splitting caused by noise, and solving the problem of delayed thermal runaway warning under high-rate working conditions.

[0041] (2) The present invention solves the contradiction between the high cost of fully deployed strain gauges and the low reliability of local monitoring through the architecture of sparse sensing of key nodes and virtual prediction closed loop; based on the key battery blocks screened by principal component analysis of thermal coupling matrix, the stress of non-key blocks is predicted using graph timing network, and the thermal expansion coefficient constraint loss is introduced to ensure physical rationality; the weight is dynamically corrected by combining vibration energy integration, and the prediction reliability is maintained through the confidence trigger calibration mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a structural block diagram of the ternary IoT battery management system of the present invention.

[0043] Figure 2 This is a structural block diagram of the clustering enhancement module of the present invention. DETAILED DESCRIPTION

[0044] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0045] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0046] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0047] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0048] Example 1, see Figure 1 The present invention provides a structural block diagram of a ternary IoT battery management system. Figure 1 The ternary IoT battery management system shown,

[0049] The sensor data acquisition module numbers the lithium batteries by battery block and deploys digital temperature sensors and digital strain gauge sensors respectively, outputting temperature and stress readings with timestamps and coordinates. The stress readings reflect the changes in thermal expansion stress between the battery casing and the adhesive.

[0050] The feature construction module calculates the local temperature gradient, neighborhood temperature gradient, spatial gradient ratio, and stress coupling ratio features of the battery block based on temperature and stress readings, and constructs a spatiotemporal feature vector by combining spatial weights and gradient variance. The stress coupling ratio feature is used to quantify the coupling effect of thermal expansion stress on temperature readings, quantified by the ratio of the stress change rate to the local temperature change rate. The local temperature gradient refers to the temperature change of the same battery block between the current sampling time and the previous sampling time. The neighborhood temperature gradient refers to the difference between the temperature of battery block i and the average temperature of the battery blocks physically connected to the battery block at the same time. The spatial gradient ratio refers to the ratio of the neighborhood temperature gradient to the local temperature gradient.

[0051] A clustering analysis module performs cluster analysis on spatiotemporal feature vectors based on clustering parameters, including a cluster radius threshold and a minimum neighborhood point threshold. The minimum neighborhood point threshold is used to define the density standard for cluster formation; the spatiotemporal feature vectors are divided into spontaneous heating clusters, radiant heat clusters, and stress coupling clusters, and cluster centers and cluster boundaries are updated in real time. In one possible embodiment, based on a baseline threshold, adjustments are made based on thermal conductivity confidence and temperature gradient variance. When the thermal conductivity confidence is lower than the threshold, the minimum neighborhood point threshold is increased to suppress pseudo-clusters caused by measurement drift. Under highly dynamic conditions, the minimum neighborhood point threshold is increased to improve robustness to temperature changes.

[0052] Explanation: Self-heating clusters refer to local temperature gradients that are significantly high (for example, greater than 2.0°C / min) and have extremely low spatial gradient ratios (for example, less than 0.3). Essentially, this is because instantaneous heat generated by electrochemical reactions or internal resistance within the battery block plays a dominant role, while the influence of external heat transfer can be ignored. Self-heating clusters reflect the active heat generation of the battery itself.

[0053] Radiative heat clusters refer to extremely low local temperature gradients (e.g., less than 0.5°C / min) with significantly high spatial gradient ratios (e.g., greater than 3.0). Essentially, temperature changes are driven primarily by heat transferred from adjacent battery blocks through radiation or conduction, with minimal internal heat generation. Radiative heat clusters indicate external heat source threats that could trigger cascading thermal runaway.

[0054] Stress coupling clusters refer to clusters with abnormally high stress coupling ratios (e.g., greater than 2.0) and drastic fluctuations in the temperature gradient in the neighborhood (e.g., gradient variance greater than 1.5°C). 2 / min 2 ); The essence is that the mechanical stress generated by the thermal expansion of the battery casing or adhesive significantly interferes with the temperature sensor reading, and is accompanied by unstable heat flow; the stress coupling cluster indicates a high-risk state where the risk of mechanical structure failure and measurement distortion coexist.

[0055] Background information shows that the self-heating mode and the radiation heat mode are naturally separated in the three-dimensional feature space, where the local temperature gradient reflects the instantaneous temperature change caused by the electrochemical reaction and internal resistance heating within the block, the neighborhood temperature gradient reflects the heat change transferred by the surrounding blocks through conduction or radiation, and the spatial gradient ratio quantifies the contribution of external heat sources relative to self-heating; based on these designed spatiotemporal feature vectors, two clusters of high-density distribution are presented under the Euclidean distance metric - one cluster is concentrated with high local gradient and low spatial gradient ratio, representing the dominance of self-heating; the other cluster is concentrated with low local gradient and high spatial gradient ratio, representing the dominance of neighboring radiation heat; the incremental DBSCAN algorithm uses adjustable ε and MinPts to accurately define the cluster center and boundary, and updates the cluster structure in real time to adapt to temperature mutations and measurement noise, ensuring that the heat source type can be robustly distinguished under any operating conditions, thereby providing a reliable basis for subsequent compensation and alarm.

[0056] The temperature compensation calculation module calculates the characteristic gradient means of the self-heating cluster, the radiant heat cluster, and the stress coupling cluster based on the clustering results to obtain the first thermal coupling coefficient and the second thermal coupling coefficient. These coefficients are filtered using the exponential moving average method (smoothing factor β≈0.1) and then used to make real-time corrections to the temperature reading of battery block i.

[0057] In a possible embodiment, the temperature reading of battery block i is corrected in real time by the following formula:

[0058]

[0059] in, Indicates the Battery pack at time The raw temperature reading; Indicates the Battery pack at time Corrected temperature readings after double compensation; Represents the first coupling coefficient (thermal coupling coefficient), at time It is obtained by filtering the ratio of the average neighborhood gradient of the radiant heat cluster to the average local gradient of the spontaneous heat cluster through an exponential moving average. It is dimensionless. Represents the second coupling coefficient (stress coupling coefficient), at time It is obtained by filtering the ratio of the average stress coupling ratio of the stress coupling cluster to the average local gradient of the spontaneous heating cluster through exponential moving average. It is dimensionless. Indicates the Battery pack at time The neighborhood temperature gradient is the difference from the average temperature of all directly connected battery blocks, dimensionless; Indicates the Battery pack at time The stress coupling ratio characteristic is the ratio of the difference between the bulk stress and the neighborhood average stress to the local temperature gradient, which is dimensionless;

[0060] The alarm management module triggers a local alarm and sends temperature compensation parameters and clustering parameters to the edge gateway to achieve remote dynamic adaptive closed-loop when the proportion of radiant heat clusters, stress coupling cluster proportions, or temperature compensation error exceeds the threshold.

[0061] For example, when the proportion of radiation heat clusters and stress coupling clusters in the last N frames (for example, 100 frames) exceeds the preset threshold or the compensation error exceeds the limit, a local alarm is triggered, and the current first thermal coupling coefficient, second thermal coupling coefficient, and clustering algorithm parameters are sent to the edge gateway to achieve remote dynamic update and adaptive closed loop of clustering parameters (clustering radius threshold) and temperature compensation parameters (first thermal coupling coefficient, second thermal coupling coefficient); at the same time, the temperature compensation error is monitored, and a local alarm is triggered when it exceeds the threshold.

[0062] Explanation: The alarm for excessive radiation heat cluster ratio is aimed at the risk of heat diffusion chain reaction - when the radiation heat transfer of adjacent battery blocks dominates the local temperature rise (manifested as a high spatial gradient ratio feature), it may trigger systemic thermal runaway, and the radiation heat source needs to be located immediately and the clustering sensitivity needs to be enhanced; the alarm for excessive stress coupling cluster ratio / compensation error responds to the risk of structural damage and measurement distortion - abnormal thermal expansion stress will distort temperature readings and threaten mechanical integrity, and the structural status needs to be urgently checked and the stress coupling effect needs to be dynamically corrected; the dual-path early warning mechanism monitors the clustering results and compensation accuracy in real time, combined with the remote closed-loop update of the clustering parameters, to ensure that the system maintains robustness in multi-dimensional fault scenarios such as heat propagation runaway and structural failure, forming an in-depth defense system for battery safety management.

[0063] In a possible embodiment, the feature construction module further includes a dynamic aging factor calculation unit, which generates an aging coefficient based on the number of battery cycles and the capacity attenuation rate, applies aging weight compensation to the neighborhood temperature gradient, and the aging coefficient is calibrated online through electrochemical impedance spectroscopy;

[0064] Specific implementation method: During the long-term operation of lithium battery packs, the phase change of electrode active materials and the thickening of SEI film lead to thermal conductivity attenuation. The traditional neighborhood temperature gradient calculation does not take the aging effect into account, resulting in systematic deviations. To eliminate this effect, the dynamic aging factor calculation unit collects the number of battery cycles in real time. With current capacity , combined with the initial capacity Calculating capacity fade rate ;Determine the cumulative number of battery cycles based on the capacity decay rate;

[0065] Acquire electrochemical impedance spectroscopy online simultaneously through AC impedance meter , and compare it with the initial impedance reference Comparison generates impedance attenuation terms; by the circulation attenuation term and the impedance attenuation term The product constitutes the aging coefficient ,in is the reference threshold value under 2000 cycles, The attenuation slope coefficient calibrated for material fatigue testing;

[0066] Aging coefficient Acting on the neighborhood temperature gradient , through the formula Calculate the corrected neighborhood temperature gradient , ensuring that the stress coupling ratio characteristics after battery aging can accurately reflect the thermal-mechanical coupling phenomenon.

[0067] It needs to be further explained in the embodiment of the present invention that the cluster radius threshold is obtained in the following manner:

[0068] Through online or periodic offline thermal imaging, the steady-state thermal conductivity of any two adjacent battery blocks is calculated. The steady-state thermal conductivity represents the ratio of the heat flow per unit time from battery block i to battery block j to the temperature difference between the two under steady-state conditions. The effective thermal coupling distance between battery blocks is defined as the inverse of the steady-state thermal conductivity. When clustering, the effective thermal coupling distance is used instead of the geometric distance to calculate the neighborhood range of each pair of nodes, so that the cluster radius threshold can be adaptively expanded and contracted according to the thermal coupling strength, and the cluster center and cluster boundaries are updated. The cluster center refers to the mean vector of all spatiotemporal eigenvectors in the cluster, and the cluster boundary is defined by the cluster radius threshold.

[0069] Explanation: The steady-state thermal conductivity is embedded in the density clustering, and the physical thermal parameters are deeply integrated with the machine learning clustering threshold. With slight changes in the environment or structure, the steady-state thermal conductivity and thermal coupling effective distance of two adjacent battery blocks can be updated in real time. The cluster radius threshold truly reflects the thermal coupling strength between battery blocks, and the cluster boundary and cluster center are no longer restricted by the geometric topology, which significantly improves the partitioning accuracy under different heat dissipation paths and different structural layouts.

[0070] In one possible embodiment, a digital twin model is used to simulate the multi-physics field thermal coupling matrix under the current working conditions in real time, and the simulated steady-state thermal conductivity is dynamically output; the cluster radius threshold is adjusted based on the steady-state thermal conductivity average value and standard deviation; a group of steady-state thermal conductivity samples are measured by thermal imaging several times in each group of adjacent battery blocks; and the average value is calculated. and standard deviation ; Adjust the cluster radius threshold using the following formula:

[0071]

[0072] in, is the experience adjustment coefficient, The value is 0.001 to prevent the denominator from being 0 and to perform dimensionless processing on each variable in the calculation.

[0073] In one possible embodiment, see Figure 2The structural block diagram of the clustering enhancement module of the system includes a clustering enhancement module for forcing the clustering results to conform to Fourier's heat conduction law, suppressing abnormal cluster splitting caused by measurement noise, and improving the clustering robustness under extreme working conditions by defining a physical enhancement distance function in the DBSCAN algorithm; the clustering enhancement module includes a residual calculation unit, a distance fusion unit, and a clustering determination optimization unit; the operation process of the clustering enhancement module includes the following steps:

[0074] The residual calculation unit performs discrete Laplace operator calculation on the temperature field data in the candidate neighborhood set. Refers to the battery blocks within the cluster radius threshold, generating heat conduction residuals ;in, represents the temperature change rate of the i-th battery block, Indicates the thermal diffusivity of battery materials, calibrated by differential scanning calorimetry, represents the temperature of the battery block adjacent to battery block i, represents the geometric center distance between battery block k and battery block i;

[0075] The distance fusion unit transforms the spatiotemporal feature vector The Euclidean distance is weightedly fused with the heat conduction residual to generate the physical enhancement distance ;in, represents the spatial distance between battery block i and battery j; Represents an adjustable weight coefficient; represents the thermal conduction residual of battery block j;

[0076] The cluster determination optimization unit uses the physical enhancement distance as the updated cluster radius threshold; in a possible embodiment, To perform refined cluster division under the conditions, generate cluster membership relationships that conform to the laws of thermodynamics; when the maximum residual within the cluster When the cluster splitting engine performs a physical consistency check on the cluster, it generates a sub-cluster structure that isolates the abnormal points.

[0077] Summary: The embodiment of the present invention deploys temperature / stress sensors with spatiotemporal coordinates to construct a spatiotemporal feature vector that integrates the local temperature gradient, neighborhood temperature gradient, and stress coupling ratio. The clustering radius threshold is dynamically set based on the inverse of the thermal conductivity, and the feature vector is divided into physically interpretable clusters using an incremental clustering algorithm. The interference of temperature readings is further eliminated through a dual temperature compensation formula. When the proportion of radiation heat clusters / stress coupling clusters exceeds the limit or the compensation error exceeds the standard, the clustering parameters and compensation coefficients are dynamically sent to the edge gateway to achieve closed-loop updates, significantly improving the thermal runaway warning accuracy and structural failure detection rate.

[0078] Example 2 differs from Example 1 in that the system further includes: a strain sensing management module, which obtains the position weight of each battery block through principal component analysis of a thermal coupling matrix, screens and outputs key battery blocks based on the position weights, and matches real strain sensors for the key battery blocks; matches stress prediction models for non-critical battery blocks, outputs the stress of non-critical battery blocks through the stress prediction model, and activates real strain gauge calibration when the prediction confidence is lower than the preset requirement; thereby achieving the purpose of saving resources and resolving the contradiction between the high cost of fully deployed strain gauges and the low reliability of local monitoring.

[0079] In a possible embodiment, the key battery block screening process includes a vibration sensitivity correction unit that corrects the position weight based on the vibration energy integral obtained by the acceleration sensor, and the vibration energy integral is calibrated through a material fatigue test.

[0080] Explanation: To address the contradiction between the high cost of deploying strain gauges across the entire battery pack and the insufficient reliability of local monitoring in thermal stress monitoring, this solution uses physical drive to select key monitoring points; and constructs a thermal coupling effective distance matrix based on thermal imaging data. ,right Before performing singular value decomposition extraction principal component directions (Cumulative variance contribution rate z85%); calculate each battery block In order to deal with the material fatigue risk caused by vehicle vibration, the three-axis acceleration sensor data is introduced. , after retaining the energy of the resonance frequency band, calculate the vibration energy integral ; Perform non-quantization processing in calculation;

[0081] Based on vibration energy integration and vibration coupling coefficient (A dimensionless parameter that quantifies the fatigue damage strength of the battery structure due to mechanical vibration. Calibrated through accelerated aging tests on a vibration table. Its value reflects the increase in thermal stress sensitivity of the battery case and adhesive per unit vibration energy.) Corrected position weight ; Represents the corrected position weight, according to The top 20% of battery blocks are selected in descending order to deploy real strain sensors, covering more than 95% of high stress risk areas.

[0082] It should be explained in the embodiment of the present invention that the screening process of the key battery blocks includes the following steps:

[0083] Obtain the thermal coupling matrix, perform singular value decomposition, extract the first k principal component directions based on Pareto, and generate a projection vector level. Use m to represent the index of the principal component direction. Calculate the position weight of the battery block using the following formula:

[0084]

[0085] in, Indicates the dominant direction of heat flow transfer in the mth principal direction, Represents the geometric center coordinates of battery block i;

[0086] The key battery blocks are screened based on the position weight, which reflects the projection intensity of the battery block in the main direction of heat conduction.

[0087] It should be explained in the embodiment of the present invention that the stress prediction model construction method includes the following steps:

[0088] Step 1: Obtain the real stress time series data, temperature spatiotemporal series data, battery block position weights, and battery operating status parameters of key battery blocks; use a dynamic sliding window to extract spatiotemporal coupling features and output a multidimensional feature vector;

[0089] Explanation: The multidimensional feature vector includes statistical characteristics of temperature / current (mean, zero-crossing rate, autocorrelation), stress-temperature hysteresis cross-correlation, and integrates spatial attributes to generate neighborhood weighted temperature (key block weight dominance)

[0090] The specific implementation method is as follows: input data preparation and multi-dimensional feature extraction: the input data covers four types of source information: the real stress time series data of key battery blocks (from the deployed strain sensors), the temperature time series data of all battery blocks (with timestamps and coordinates), the battery block position weight (derived from the spatial projection intensity of the principal component analysis of the thermal coupling matrix), and the battery operating status parameters (including charge and discharge current, voltage and SOC); in the feature extraction stage, a sliding window (for example, a 60-second window) is used to calculate statistical features for time series variables such as temperature and current: first-order features such as mean, variance and linear trend slope; high-order features such as zero-crossing rate (the number of times the mean is crossed per unit time) and autocorrelation coefficient (lagged by 5 seconds); cross-features such as temperature-current covariance and stress-temperature lagged cross-correlation (time shifted by 3 seconds); the extracted features are fused with spatial attributes, and the neighborhood weighted average temperature is calculated based on the position weight (the weight of physically connected key battery blocks accounts for 70%), and physical constraint features such as the product of thermal expansion coefficient and temperature gradient are introduced to embed the thermodynamic properties of the material;

[0091] Step 2: Using the measured stress of key battery blocks as labels, a training set is constructed in combination with multidimensional feature vectors. A time-series graph convolutional network is constructed using position weights as edge weights and physical connections as topology. Dynamic characteristics are captured through gated recurrent units. A two-layer loss function is defined during the training process: the main loss is the Huber loss (to suppress outlier interference), and the constraint term is a physical consistency penalty term (which activates a gradient penalty when the rate of change of the predicted value exceeds the material yield strength threshold).

[0092] The specific implementation method is as follows: training data set construction and constraint model training: the training data set uses the actual stress data measured by the sensor of the key battery block as the label and the multi-dimensional feature vector as the input; the data is enhanced by adding Gaussian noise with a standard deviation of 0.05 to simulate the actual measurement disturbance; the stress prediction model adopts a lightweight time-series graph convolutional network, whose graph structure uses battery blocks as nodes, physical connections as edges, and position weights as edge weights. The time-series unit uses a gated recurrent unit to process the multi-dimensional feature vector; the training process defines a double-layer loss function: the main loss is the Huber loss (δ=0.1, balancing the influence of outliers), and the constraint term is the physical consistency penalty term (λ=0.3, a linear penalty is imposed when the stress change rate exceeds the material threshold τ=2MPa); the optimizer uses Nesterov accelerated gradient descent, the learning rate periodically decays from 0.01 according to the cosine annealing strategy, the batch size is 128, and the early stopping patience value is set to 15 rounds;

[0093] Step 3: Physical consistency verification and edge adaptive deployment: Use the confidence of the stress prediction model and the material yield strength violation rate to quantify the prediction reliability. When the prediction reliability meets the requirements, the trained stress prediction model is deployed and applied.

[0094] The specific implementation method is as follows: in the verification phase, the prediction confidence C of the core indicator is defined, and the physical violation rate (the ratio of the predicted value to the yield strength of the metamaterial) and delay sensitivity (output fluctuation of ±5% disturbance of the input) are monitored at the same time; the test scenarios cover extreme working conditions (such as 5C discharge superimposed on the temperature change of 25℃→45℃) and failure modes (such as 0.3MPa sensor bias drift); during deployment, the model is compressed to the edge device through 8-bit integer quantization, and the inference is performed at a synchronized temperature sampling cycle of 1Hz; the dynamic calibration mechanism calculates the confidence C in real time. If it is lower than the threshold of 0.85 for 5 consecutive frames, the strain gauge calibration of the target non-critical battery block is activated, and new data is injected into the online learning pipeline to start 10 minutes of incremental training, forming a resource adaptive closed loop.

[0095] In one possible embodiment, the stress prediction model includes a real-time graph topology generation engine, which dynamically reconstructs graph neural network connection edges according to the temperature gradient vector field, and the heat flux intensity threshold is adaptively adjusted according to the local temperature change rate;

[0096] Explanation: To adapt to the transient changes in heat flow direction under dynamic conditions such as fast charging, the stress prediction model integrates a real-time graph topology generation engine; at each sampling moment, based on the battery block Calculate the three-dimensional temperature gradient vector based on the spatial temperature distribution , forming a full-coverage temperature gradient vector field; defining the local temperature change rate As a heat flow intensity adjustment factor, dynamically adjust the connection edge judgment threshold ( K / mm is the basic threshold, is the sensitivity coefficient); Indicates the basic heat flux intensity threshold, that is, the minimum strength requirement for determining the heat flux connection between battery blocks under static conditions, calibrated by the material thermal conductivity coefficient; It represents the temperature change rate sensitivity coefficient, which is used to quantify the modulation intensity of local temperature change on the heat flux intensity threshold and is determined by thermal shock test;

[0097] Traverse all battery block pairs , calculate the gradient direction similarity , Indicates battery pack The spatial temperature distribution of the three-dimensional temperature gradient vector is calculated. When establishing connection edges in graph neural networks ; When the gradient is greater than 0 and in the same direction, it indicates that the heat flow is enhanced and a connection is established; when the gradient is less than 0 and in the opposite direction, no connection is established;

[0098] The topology is updated every frame and input into the time-series graph convolutional network to ensure that thermal flow mutations under fast charging conditions can be captured in real time, reducing prediction errors.

[0099] Summary: The embodiment of the present invention realizes strain monitoring resource optimization through physical-driven key point screening and dynamic graph neural network prediction: first, the main direction of heat flow is extracted based on the singular value decomposition of the thermal coupling matrix, and the battery block position weight is calculated; the vibration energy integral is introduced to correct the battery block position weight; a physical constraint type timing graph convolutional network is constructed for non-critical blocks - the position weight is used as the edge weight and the physical connection is used as the topology, and the multi-dimensional features of the neighborhood weighted temperature and thermal expansion coefficient constraints are input and fused. The Huber loss master function and the material yield strength penalty term are used in training to improve the reliability of stress monitoring of the battery system.

[0100] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. Ternary IoT battery management system, characterized by: include: The sensor data acquisition module numbers the lithium batteries by battery block and deploys digital temperature sensors and digital strain gauge sensors respectively, outputting temperature and stress readings with timestamps and coordinates. The stress readings reflect the changes in thermal expansion stress between the battery casing and the adhesive. The feature construction module calculates the local temperature gradient, neighborhood temperature gradient, spatial gradient ratio, and stress coupling ratio characteristics of the battery block based on temperature and stress readings, and constructs a spatiotemporal feature vector by combining spatial weights and gradient variance; The cluster analysis module performs cluster analysis on spatiotemporal feature vectors based on clustering parameters, including a cluster radius threshold and a minimum neighborhood point threshold. The minimum neighborhood point threshold is used to define the density standard for cluster formation. The module divides the spatiotemporal feature vectors into self-heating clusters, radiant heat clusters, and stress coupling clusters, and updates cluster centers and cluster boundaries in real time. The temperature compensation calculation module calculates the characteristic gradient means of the self-heating cluster, the radiation heat cluster, and the stress coupling cluster according to the clustering results, and obtains the first thermal coupling coefficient and the second thermal coupling coefficient, which are then used after being filtered by the exponential moving average method; Correcting the temperature reading of the battery block i in real time based on the calculated first thermal coupling coefficient and the second thermal coupling coefficient; The alarm management module triggers a local alarm and sends temperature compensation parameters and clustering parameters to the edge gateway to achieve remote dynamic adaptive closed-loop when the proportion of radiant heat clusters, stress coupling cluster proportions, or temperature compensation error exceeds the threshold. The strain sensor management module obtains the position weight of each battery block through principal component analysis of the thermal coupling matrix, screens and outputs key battery blocks based on the position weight, and matches real strain sensors for the key battery blocks; matches stress prediction models for non-critical battery blocks, outputs the stress of the non-critical battery blocks through the stress prediction model, and activates real strain gauge calibration when the prediction confidence level is lower than the preset requirement. The screening process of the key battery blocks includes the following steps: Obtain the thermal coupling matrix, perform singular value decomposition, extract the first k principal component directions based on Pareto, and generate a projection vector level. Use m to represent the index of the principal component direction. Calculate the position weight of the battery block using the following formula: Among them, u m Indicates the dominant direction of heat flow transfer in the mth principal direction, v i Represents the geometric center coordinates of battery block i; The key battery blocks are screened based on the position weight, which reflects the projection strength of the battery block in the main direction of heat conduction. The stress prediction model is constructed as follows: Step 1: Obtain the real stress time series data, temperature spatiotemporal series data, battery block position weights, and battery operating status parameters of key battery blocks; use a dynamic sliding window to extract spatiotemporal coupling features and output a multidimensional feature vector; Step 2: Using the measured stress of key battery blocks as labels, a training set is constructed based on multidimensional feature vectors. A time-series graph convolutional network is constructed using position weights as edge weights and physical connections as topology. Dynamic characteristics are captured using gated recurrent units. A two-layer loss function is defined during the training process: the main loss is the Huber loss, and the constraint term is the physical consistency penalty term. Step 3: Physical consistency verification and edge adaptive deployment: Use the confidence of the stress prediction model and the material yield strength violation rate to quantify the prediction reliability. When the prediction reliability meets the requirements, the trained stress prediction model is deployed and applied.

2. The ternary IoT battery management system according to claim 1, characterized in that: The feature construction module also includes a dynamic aging factor calculation unit, which generates an aging coefficient based on the number of battery cycles and the capacity attenuation rate, applies aging weight compensation to the neighborhood temperature gradient, and the aging coefficient is calibrated online through electrochemical impedance spectroscopy.

3. The ternary IoT battery management system according to claim 1, characterized in that: The cluster radius threshold is obtained as follows: Through online or periodic offline thermal imaging, the steady-state thermal conductivity of any two adjacent battery blocks is calculated. The steady-state thermal conductivity represents the ratio of the heat flow per unit time from battery block i to battery block j to the temperature difference between the two under steady-state conditions. The effective thermal coupling distance between battery blocks is defined as the inverse of the steady-state thermal conductivity. When clustering, the effective thermal coupling distance is used instead of the geometric distance to calculate the neighborhood range of each pair of nodes, so that the cluster radius threshold can be adaptively expanded and contracted according to the thermal coupling strength, and the cluster center and cluster boundaries are updated. The cluster center refers to the mean vector of all spatiotemporal eigenvectors in the cluster, and the cluster boundary is defined by the cluster radius threshold.

4. The ternary IoT battery management system according to claim 3, characterized in that: The digital twin model is used to simulate the multi-physics field thermal coupling matrix under the current working conditions in real time, and the simulated steady-state thermal conductivity is dynamically output; the cluster radius threshold is adjusted based on the steady-state thermal conductivity average value and standard deviation; a group of steady-state thermal conductivity samples are measured by thermal imaging several times in each group of adjacent battery blocks; the average value E[K ij ] and standard deviation σ ij ; Adjust the cluster radius threshold ε by the following formula ij : Among them, ρ is the empirical adjustment coefficient, and τ is set to 0.001 to prevent the denominator from being zero.

5. The ternary IoT battery management system according to claim 1, characterized in that: The system includes a clustering enhancement module for forcing clustering results to conform to Fourier's heat conduction law, suppressing abnormal cluster splitting caused by measurement noise, and improving clustering robustness under extreme working conditions by defining a physical enhancement distance function in the DBSCAN algorithm. The clustering enhancement module includes a residual calculation unit, a distance fusion unit, and a clustering determination optimization unit. The operation process of the clustering enhancement module includes the following steps: The residual calculation unit performs discrete Laplace operator calculation on the temperature field data in the candidate neighborhood set, which refers to the battery blocks within the cluster radius threshold, to generate the thermal conduction residual. in, represents the temperature change rate of the i-th battery block, K α Indicates the thermal diffusivity of battery materials, calibrated by differential scanning calorimetry, T k represents the temperature of the battery block adjacent to battery block i, represents the geometric center distance between battery block k and battery block i; The distance fusion unit transforms the spatiotemporal feature vector The Euclidean distance is weightedly fused with the heat conduction residual to generate the physical enhancement distance in, represents the spatial distance between battery block i and battery j; λ represents the adjustable weight coefficient; R phy,j represents the thermal conduction residual of battery block j; The cluster determination optimization unit uses the physical enhancement distance as the updated cluster radius threshold.

6. The ternary IoT battery management system according to claim 1, characterized in that: The key battery block screening process includes a vibration sensitivity correction unit that corrects the position weight based on the vibration energy integral obtained by the acceleration sensor. The vibration energy integral is calibrated through material fatigue testing, including: In order to deal with the material fatigue risk caused by vehicle vibration, the triaxial acceleration sensor data a(t) is introduced, and after retaining the energy in the resonance frequency band, the vibration energy integral is calculated. Based on the vibration energy integral and vibration coupling coefficient Z γ Corrected position weight Represents the corrected position weight, according to Select key battery blocks in descending order.

7. The ternary IoT battery management system according to claim 1, characterized in that: The stress prediction model includes a real-time graph topology generation engine, which dynamically reconstructs the graph neural network connection edges based on the temperature gradient vector field, and adaptively adjusts the heat flux intensity threshold according to the local temperature change rate, including: At each sampling moment, the three-dimensional temperature gradient vector is calculated based on the spatial temperature distribution of battery block i. Forming an all-encompassing temperature gradient vector field; Define the local temperature change rate As a heat flow intensity adjustment factor, dynamically adjust the connection edge judgment threshold where θ base represents the basic heat flux intensity threshold, κ represents the temperature change rate sensitivity coefficient; Traverse all battery block pairs (i, j) and calculate the gradient direction similarity s ij =g i ·g j , g j Represents the spatial temperature distribution of battery block j and calculates the three-dimensional temperature gradient vector. When s ij ≥θ threshold When establishing a connection edge e in the graph neural network ij ;s ij When the gradient is greater than 0 and in the same direction, it indicates that the heat flow is enhanced and a connection is established; when the gradient is less than 0 and in the opposite direction, no connection is established; The topology is updated every frame and input into the temporal graph convolutional network to reduce the prediction error.

Images