A battery abnormality warning system based on intelligent battery gateway

Abstract

The present invention discloses a battery abnormality warning system based on an intelligent battery gateway, which relates to the field of battery monitoring technology. The system includes: a sensor module for real-time acquisition of the voltage, internal resistance and temperature of a single battery; an intelligent battery gateway, including a data processing module and an early warning module, the data processing module performs fusion analysis on the sensor data through an intelligent algorithm, calculates the battery's remaining capacity SOC, health status SOH and executes an intelligent balancing strategy; the early warning module generates a multi-level early warning signal according to the parameter threshold and the anomaly detection model. The present invention uses multi-dimensional fusion perception technology and dual-drive intelligent algorithms to accurately capture signs of aging of the battery's internal microstructure, and achieves high-precision estimation of the battery's remaining capacity and health status; in terms of early warning performance, a multi-dimensional linkage detection model and an intelligent early warning mechanism are established to effectively improve the accuracy and timeliness of fault early warnings, and significantly advance the warning time.

Description

Technical Field

[0001] The present invention relates to the technical field of battery monitoring, and in particular to a battery abnormality early warning system based on an intelligent battery gateway. Background Art

[0002] Traditional monitoring solutions face a core challenge in the full lifecycle management of batteries: Due to their use of single-parameter monitoring and fixed-threshold alarm mechanisms, they struggle to accurately capture the aging of the battery's internal microstructure, leading to large deviations in remaining capacity (SOC) and state of health (SOH) estimations, and an inability to provide effective early warnings of failures. Specifically, relying solely on surface parameters such as voltage and temperature, they are unable to detect internal aging, such as SEI film growth and plate sulfation. Warning methods based on single-point thresholds (such as an alarm for temperatures > 60°C) are unable to identify parameter coupling anomalies that precede thermal runaway, resulting in insufficient lead time for warnings. Traditional algorithms struggle to adapt to varying operating conditions, such as high and low temperatures and varying charge and discharge rates. Long-term operation leads to significant degradation in estimation accuracy, resulting in delayed maintenance and wasted resources. Summary of the Invention

[0003] The purpose of the present invention is to provide a battery abnormality warning system based on an intelligent battery gateway to solve the problems raised in the above background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a battery abnormality warning system based on an intelligent battery gateway, comprising the following steps:

[0005] The sensor module is used to collect the voltage, internal resistance, temperature, thermal runaway status and float charge voltage parameters of the single battery in real time;

[0006] The intelligent battery gateway includes a data processing module and an early warning module. The data processing module uses intelligent algorithms to integrate and analyze sensor data, calculate the remaining battery capacity (SOC) and health status (SOH), and implement intelligent balancing strategies. The early warning module generates multi-level early warning signals based on parameter thresholds and anomaly detection models.

[0007] Communication module, used for local data interaction and remote platform data upload, supporting Bluetooth, serial port and network communication;

[0008] Human-computer interaction module, used for visual early warning and operation and maintenance management of abnormal conditions.

[0009] Preferably, the sensor module includes:

[0010] Composite parameter measurement unit, used to synchronously collect the voltage response curve and AC internal resistance of the single battery using the pulse current injection method, and obtain the charge transfer resistance , electric double layer capacitors Electrochemical impedance spectroscopy (EIS) data;

[0011] The thermal runaway detection unit is used to identify thermal runaway precursors through a three-dimensional linkage criterion of temperature gradient, internal resistance mutation rate, and voltage second-order derivative. The criterion formula is:

[0012] ;

[0013] Where, is the temperature change rate per unit time, is the rate of change of internal resistance relative to the initial value, is the curvature of the voltage curve.

[0014] Preferably, the sensor module further includes a self-calibration unit, and the self-calibration unit includes:

[0015] The MEMS temperature and pressure compensation module is used to monitor the ambient air pressure and humidity in real time, and dynamically correct the thermal resistance coefficient of the temperature sensor through a lookup table method. The temperature measurement accuracy after correction is ±0.5°C.

[0016] The standard voltage source module is used to automatically calibrate the voltage acquisition channel every 24 hours. After calibration, the voltage drift error is less than 0.05%.

[0017] Preferably, the SOC calculation algorithm of the data processing module is an improved volumetric Kalman filter algorithm, and the calculation of the battery remaining capacity SOC includes:

[0018] Set up a third-order RC equivalent circuit model to identify ohmic internal resistance in real time , charge transfer resistance , electric double layer capacitors Etc. model parameters;

[0019] Construct multi-input measurement equations to integrate voltage , internal resistance, temperature T Construct the state-space equations:

[0020] ;

[0021] in, is the open circuit voltage, is the ohmic voltage drop, is the current flowing through the battery, is the polarization voltage, is the rate of change of polarization voltage with time, is the change in polarization voltage caused by the current charging the double-layer capacitance, is the polarization voltage relaxation time constant;

[0022] An anti-outlier adaptive mechanism is set up, and the measurement residual is judged by Mahalanobis distance. When the limit is exceeded, it switches to square root cubature Kalman filtering. The calculation accuracy of SOC is ±3%.

[0023] Preferably, the SOH prediction algorithm of the data processing module is a gated recurrent unit algorithm based on the attention mechanism, including:

[0024] Construct a multi-feature input layer to extract 12-dimensional aging features such as internal resistance growth rate and capacity decay rate, and build a time window sliding matrix;

[0025] Construct an attention mechanism layer to calculate the contribution of historical data to the current SOH by weight, focusing on the characteristics of the accelerated aging stage. The weight formula is:

[0026] ;

[0027] in, for Attention weight at all times, is the hidden layer state, is the hidden layer state at the previous moment, Is a scoring function used to measure the hidden layer state at the previous moment and the current hidden layer state the degree of correlation between them;

[0028] Construct a physical constraint correction layer and correct the prediction results through the capacity-internal resistance coupling equation:

[0029] ;

[0030] in, It represents the change ratio of the measured internal resistance relative to the initial internal resistance and the failure internal resistance. The coefficient of 0.8 indicates that the internal resistance growth has a dominant weight in the SOH assessment. is the capacity attenuation ratio. The coefficient of 0.2 indicates that the capacity attenuation has a relatively small impact. The prediction accuracy of the corrected SOH is ±5%.

[0031] Preferably, the intelligent balancing strategy is dynamic group balancing based on impedance spectrum clustering, including:

[0032] Construct health status stratification and cluster the principal components of the impedance spectrum using the K-means algorithm to divide the batteries into healthy group, degradation group, and fault warning group;

[0033] Set up a time-varying threshold trigger mechanism and set differentiated balancing thresholds based on charging and discharging conditions:

[0034] ；

[0035] Monitor the internal resistance change rate during the balancing process. Automatically terminated when

[0036] Preferably, the early warning module includes:

[0037] The five-dimensional feature anomaly detection unit constructs a feature space including voltage, internal resistance, temperature, SOC, and SOH, and uses the isolation forest algorithm to detect the isolation of data points;

[0038] The Bayesian network inference engine establishes a causal relationship knowledge base containing 56 failure modes, derives the root cause of the failure through posterior probability calculation, and outputs maintenance recommendations.

[0039] Preferably, the smart battery gateway adopts an edge computing architecture, and the smart battery gateway further includes:

[0040] Data hierarchical processing module, used to upload abnormal data fragments and 15-dimensional feature vectors, and process regular data locally;

[0041] The federated learning update module has a built-in lightweight AM-GRU model and is used to synchronize model parameter gradients through federated learning.

[0042] Preferably, the system further comprises:

[0043] A digital twin mapping module includes a built-in virtual image of the battery pack in the smart battery gateway, synchronizes single-cell battery parameters in real time, simulates the evolution of fault scenario parameters through finite element simulation, predicts potential faults, and generates personalized maintenance work orders.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] Through multi-dimensional fusion sensing technology and dual-drive intelligent algorithms, it accurately captures signs of aging in the battery's internal microstructure, achieving high-precision estimation of the battery's remaining power and health status; in terms of early warning performance, it establishes a multi-dimensional linkage detection model and intelligent early warning mechanism to effectively improve the accuracy and timeliness of fault warnings, significantly advancing the warning time; at the management and operation level, the dynamic grouping intelligent balancing strategy and fault inference engine not only improve the efficiency of battery pack management, but also realize the root cause diagnosis of faults and provide maintenance suggestions. At the same time, the edge computing architecture optimizes data processing and protects user privacy. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of the structure of a battery abnormality warning system based on an intelligent battery gateway provided in an embodiment of the present invention;

[0047] Figure 2 A diagram illustrating the implementation steps of a battery abnormality warning system based on an intelligent battery gateway provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] See also Figure 1 The present invention provides a battery abnormality early warning system based on an intelligent battery gateway, comprising:

[0050] The sensor module 11 is used to collect the voltage, internal resistance, temperature, thermal runaway state and float charge voltage parameters of the single battery in real time;

[0051] The smart battery gateway 12 includes a data processing module and an early warning module. The data processing module uses intelligent algorithms to integrate and analyze sensor data, calculate the remaining battery capacity (SOC) and health status (SOH), and implement intelligent balancing strategies. The early warning module generates multi-level early warning signals based on parameter thresholds and anomaly detection models.

[0052] Communication module 13, used for local data interaction and remote platform data upload, supporting Bluetooth, serial port and network communication;

[0053] The human-computer interaction module 14 is used for visual warning and operation and maintenance management of abnormal conditions.

[0054] In an optional embodiment, the sensor module 11 further includes:

[0055] Composite parameter measurement unit, used to synchronously collect the voltage response curve and AC internal resistance of the single battery using the pulse current injection method, and obtain the charge transfer resistance , electric double layer capacitors Electrochemical impedance spectroscopy (EIS) data.

[0056] Specifically, a pulse current with a pulse width of 10ms and an amplitude of 500mA is injected into the single battery. During this transient process, the voltage response curve of the battery is synchronously collected at a high sampling rate of 10kHz. At the same time, the AC internal resistance is accurately measured using a 10mHz-1kHz sweep frequency technology. In this way, the charge transfer resistance ( ), double layer capacitance ( ) and other battery microstructural characteristics. Traditional DC internal resistance measurement methods have limitations and struggle to capture subtle changes within the battery. However, this invention, through its high sensitivity, can accurately identify internal resistance changes as low as 0.1 mΩ. This allows for the effective detection of the earliest signs of battery aging, providing critical data support for subsequent maintenance and abnormality warnings.

[0057] Thermal runaway detection unit, used to detect the temperature gradient ( , unit ℃ / min), internal resistance mutation rate ( , relative rate of change of initial value), second-order derivative of voltage ( , reflecting the curvature of the curve) is used to identify thermal runaway precursors. The criterion formula is:

[0058] ;

[0059] Where, is the temperature change rate per unit time, is the rate of change of internal resistance relative to the initial value, is the curvature of the voltage curve.

[0060] Previous single-temperature threshold detection methods often suffered from delayed warnings or high false alarm rates. In contrast, the three-dimensional linkage criteria in this embodiment significantly improves warning time, with an advance warning time of up to 30 minutes, while reducing the false alarm rate by 60%. This significantly improves the accuracy and reliability of thermal runaway warnings, laying a solid foundation for ensuring the safe and stable operation of battery systems.

[0061] In an optional embodiment, the sensor module 11 further includes:

[0062] The MEMS temperature and pressure compensation module is used to monitor the ambient air pressure and humidity in real time, and dynamically correct the thermal resistance coefficient of the temperature sensor through a lookup table method. The temperature measurement accuracy after correction is ±0.5°C.

[0063] The standard voltage source module is used to automatically calibrate the voltage acquisition channel every 24 hours. After calibration, the voltage drift error is less than 0.05%.

[0064] Optionally, this embodiment provides a self-calibrating sensor array with integrated MEMS temperature and pressure compensation units. This array enables real-time, high-precision monitoring of ambient air pressure, with an accuracy of ±0.1 kPa, and relative humidity, with an accuracy of ±2% RH. Using a lookup table, the thermal resistance coefficient of the temperature sensor is dynamically corrected based on the real-time monitored environmental parameters. After correction, the temperature measurement accuracy can be stably controlled within ±0.5°C, effectively eliminating interference from environmental factors on temperature measurement.

[0065] In addition, the sensor array has a built-in 16-bit ADC and a standard voltage source with an accuracy of up to ±0.02% FS. Every 24 hours, the system automatically starts the calibration procedure to calibrate the voltage acquisition channel, which can effectively suppress the drift error generated by the sensor during long-term operation and ensure that it always maintains high-precision operation. The drift error is controlled at <0.05%, providing reliable guarantee for the accurate collection of battery status parameters.

[0066] In an optional embodiment, the SOC calculation algorithm of the data processing module is an improved volumetric Kalman filter algorithm, and the calculation of the battery remaining capacity SOC includes:

[0067] Set up a third-order RC equivalent circuit model to identify ohmic internal resistance in real time , charge transfer resistance , electric double layer capacitors Etc. model parameters;

[0068] Construct multi-input measurement equations to integrate voltage , internal resistance, temperature T Construct the state-space equations:

[0069] ;

[0070] in, is the open circuit voltage, is the ohmic voltage drop, is the current flowing through the battery, is the polarization voltage, is the rate of change of polarization voltage with time, is the change in polarization voltage caused by the current charging the double-layer capacitance, is the polarization voltage relaxation time constant;

[0071] An anti-outlier adaptive mechanism is set up, and the measurement residual is judged by Mahalanobis distance. When the limit is exceeded, it switches to square root cubature Kalman filtering. The calculation accuracy of SOC is ±3%.

[0072] Understandably, in order to achieve a more accurate dynamic estimation of battery SOC, we use the improved volumetric Kalman filter (CV-KF) algorithm, the core of which is to build a third-order RC equivalent circuit model. In this model, nonlinear polarization capacitors are introduced to fully reflect the complex electrochemical characteristics of the battery. In actual operation, model parameters (such as ohmic internal resistance , charge transfer resistance , electric double layer capacitors ) is not fixed; rather, through real-time identification, the model ensures that it can sensitively capture the dynamic changes in polarization characteristics across different SOC ranges. For example, in the initial stages of charging, polarization is relatively weak. As charging progresses, the polarization effect gradually intensifies in the mid-to-high SOC range. Real-time adjustment of model parameters ensures that the model consistently reflects the actual battery state.

[0073] Among them, the open circuit voltage Closely related to the SOC of the battery, in order to accurately obtain the relationship between the two, a large amount of 300℃ high temperature accelerated aging test data was obtained. Through in-depth analysis and processing of the test data, the open circuit voltage was finally The fitting is a fifth-order polynomial, and the fitting degree is This high-precision fitting result provides a solid data basis for subsequent SOC estimation.

[0074] In complex and ever-changing real-world application environments, batteries are inevitably affected by various interference factors, such as pulse current interference, which can cause deviations or even divergence in the SOC estimation results. To effectively address this issue, this embodiment is equipped with an anti-interference adaptive mechanism. This mechanism dynamically evaluates the reliability of the estimation results by calculating the Mahalanobis distance of the measurement residual in real time. Once the measurement residual exceeds 3σ, the system automatically triggers a switching mechanism, rapidly switching from the original cubature Kalman filter to the square root cubature Kalman filter (SCKF). This adaptive switching can greatly suppress the adverse effects of pulse current interference on the estimation results, significantly improving the accuracy of SOC estimation.

[0075] In an optional embodiment, the SOH prediction algorithm of the data processing module is a gated recurrent unit algorithm based on an attention mechanism, including constructing a multi-feature input layer, constructing an attention mechanism layer, and constructing a physical constraint correction layer.

[0076] Construct a multi-feature input layer to extract 12-dimensional aging features such as internal resistance growth rate and capacity decay rate, and build a time window sliding matrix;

[0077] During the data collection phase, key features of the battery aging process are deeply explored. By accurately extracting 12 aging features, including internal resistance growth rate, capacity decay rate, and self-discharge rate, the aging status of the battery is comprehensively reflected. To effectively utilize this time series data, a time window sliding matrix is ​​constructed. The window length is set based on the battery's design life and is 1 / 100 of the battery's design life. For example, for a battery with a design life of 10 years, the time window is selected to be 36 days. This sliding window mechanism can dynamically capture the changes in battery aging characteristics at different stages, providing a rich data foundation for subsequent analysis.

[0078] Construct an attention mechanism layer, and focus on the characteristics of the accelerated aging stage by weighted calculation of the contribution of historical data to the current SOH. In traditional time series analysis, all historical data are often treated equally, and the differences in the degree of influence of data in different time periods on the current state are ignored. This embodiment uses the attention mechanism for optimization. By weighted calculation of the key time periods in historical data that have the greatest impact on the current SOH, effective screening and focusing of data can be achieved. Factors that have a significant impact on battery aging, such as the number of deep discharges and the duration of high-temperature operation, can be given special attention under this mechanism. The weight formula is:

[0079] ;

[0080] in, for t Attention weight at all times, is the hidden layer state, is the hidden layer state at the previous moment, Is a scoring function used to measure the hidden layer state at the previous moment and the current hidden layer state The degree of correlation between The scores at all moments are normalized so that the sum of all weights is 1. In actual operation, this mechanism can keenly focus on the feature contributions of the accelerated battery aging stage, avoiding the gradient vanishing problem that is prone to occur in traditional recurrent neural networks when processing long sequence data, thereby enabling the model to more accurately capture the key factors affecting SOH.

[0081] A physical constraint correction layer is constructed. Considering that battery aging is not only a data-driven process but also follows certain electrochemical aging mechanisms, this embodiment uses a capacity-internal resistance coupling constraint equation to combine the physical model with the data-driven model and correct the prediction results using the capacity-internal resistance coupling equation:

[0082] ;

[0083] in, It represents the change ratio of the measured internal resistance relative to the initial internal resistance and the failure internal resistance. The coefficient of 0.8 indicates that the internal resistance growth has a dominant weight in the SOH assessment. is the capacity decay ratio. The coefficient of 0.2 indicates that the capacity decay effect is relatively small. The accuracy of the corrected SOH prediction is ±5%. In this equation, by calculating the battery internal resistance ( ) and capacity ( ) by comparing real-time monitoring data with initial and failure values, comprehensively considering the impact of capacity fade and internal resistance growth on SOH. This fusion approach integrates electrochemical aging mechanisms into data-driven results, significantly improving the accuracy of SOH prediction. Field testing has verified that this solution can improve SOH prediction accuracy to ±5%, a qualitative leap compared to the ±8% of traditional solutions, providing a more reliable basis for battery anomaly warnings.

[0084] In an optional embodiment, the intelligent balancing strategy is dynamic group balancing based on impedance spectrum clustering, including building health status stratification and setting a time-varying threshold trigger mechanism.

[0085] Construct health status stratification, cluster the principal components of the impedance spectrum using the K-means algorithm, and divide the batteries into healthy group, degradation group, and fault warning group.

[0086] Specifically, to achieve more precise battery management, this embodiment uses the K-means algorithm to perform cluster analysis on the principal components (PC1-PC3) of the impedance spectra of individual cells. Through extensive experimental data and in-depth analysis of battery characteristics, batteries are clearly divided into different health levels. When PC1 < 0.3, the battery is classified as the "healthy group," indicating good performance and no need for excessive balancing intervention. When 0.3 ≤ PC1 < 0.7, the battery enters the "degradation group," indicating a certain degree of performance decline and requiring appropriate attention and adjustment. When PC1 ≥ 0.7, the battery is placed in the "fault warning group," indicating an imminent failure requiring urgent attention. This stratified approach effectively avoids unnecessary and ineffective balancing operations on healthy batteries, greatly improving the utilization efficiency of balancing resources.

[0087] A time-varying threshold trigger mechanism is set. Considering the significant differences in battery operating conditions during the charge and discharge process, this embodiment provides a flexible time-varying threshold trigger mechanism. Specifically, the equalization threshold is dynamically adjusted according to the charge and discharge conditions as follows:

[0088] ；

[0089] Monitor the internal resistance change rate during the balancing process. Automatically terminated when

[0090] Specifically, in the above formula, Represents the change in the battery's state of charge (State of Charge), It indicates the change in the battery's state of health. More than 8% and When the load exceeds 10%, the system determines that the balance condition is triggered; when discharging, the triggering condition is more stringent. Must be greater than 12% and Greater than 15%. In addition, during the balancing process, the system will monitor the battery internal resistance change rate in real time. When the battery is fully charged, the system will automatically terminate the balancing operation. Actual tests have shown that this strategy significantly improves balancing efficiency by 40% while reducing energy consumption by 30%, effectively saving energy while improving battery performance.

[0091] In an optional embodiment, the early warning module includes:

[0092] The five-dimensional feature anomaly detection unit constructs a feature space including voltage, internal resistance, temperature, SOC, and SOH, and uses the isolation forest algorithm to detect the isolation of data points.

[0093] Specifically, to accurately detect early signs of battery failure during operation, this system innovatively constructs a five-dimensional feature vector space. This vector space encompasses parameters in five key dimensions: voltage (V), internal resistance (R), temperature (T), state of charge (SOC), and state of health (SOH). These five parameters act as a barometer of battery health; their changes can intuitively reflect the physical and chemical changes within the battery.

[0094] To effectively mine the information contained in these five-dimensional feature vectors, this embodiment employs the isolation forest algorithm. The core principle of this algorithm lies in calculating the "isolation degree" of a data point. Simply put, if a data point is relatively far away from other data points in the data space, that is, relatively isolated, its "isolation degree" is high. In battery fault detection scenarios, early-stage failures often cause certain data points to deviate from the data distribution under normal operating conditions. The isolation forest algorithm can keenly identify these anomalies.

[0095] For example, self-discharge anomaly ( / 24h>5%), the change of SOC parameters will form a point that deviates from the normal trajectory in the five-dimensional feature space. The isolation forest algorithm can quickly capture this change. ), abnormal changes in internal resistance R and state of health (SOH) are also accurately identified by the algorithm. Compared to traditional methods that rely solely on threshold judgment, this unsupervised detection method based on a five-dimensional feature space and the isolation forest algorithm has significantly increased the warning accuracy from the previous low level to 98%, providing strong technical support for the timely detection of early-stage battery failures.

[0096] The Bayesian network inference engine establishes a causal relationship knowledge base containing 56 failure modes, derives the root cause of the failure through posterior probability calculation, and outputs maintenance recommendations.

[0097] Specifically, for example, positive plate corrosion failures can be triggered when the system detects the simultaneous fulfillment of the following three conditions: "float charge voltage > 2.45V, internal resistance growth > 20% / month, and temperature > 50°C." Excessively high float charge voltage accelerates the oxidation reaction in the positive plate, while a rapid internal resistance growth rate indicates possible changes in the plate's internal structure. High temperatures further increase the rate of chemical reactions. These three factors interact to significantly increase the risk of positive plate corrosion.

[0098] Regarding electrolyte depletion, its triggering conditions are "voltage fluctuation > 0.1V + internal resistance mutation rate > 30% + ambient humidity < 20% RH." Abnormal voltage fluctuation reflects unstable electrochemical reactions within the battery. Sudden internal resistance changes may be due to reduced electrolyte, hindering ion conduction. Low ambient humidity accelerates electrolyte evaporation. These factors, combined, point to the possibility of electrolyte depletion.

[0099] When the system detects an anomaly, it quickly invokes the Bayesian network fault inference engine. Based on the Bayesian principle, this engine uses complex and precise posterior probability calculations to deduce the most likely root cause of the current anomaly from a multitude of possible causes, with a confidence level exceeding 90%. Once the cause of the fault is determined, the system immediately outputs detailed and targeted maintenance recommendations, such as "It is recommended that the electrolyte specific gravity of battery No. 15 be tested." This transition from a simple "threshold alarm" to a more in-depth "root cause diagnosis" greatly improves the efficiency and accuracy of fault handling, providing reliable protection for the stable operation of the battery system.

[0100] In an optional embodiment, the smart battery gateway 12 adopts an edge computing architecture, and the smart battery gateway 12 further includes:

[0101] Data hierarchical processing module, used to upload abnormal data fragments and 15-dimensional feature vectors, and process regular data locally;

[0102] The federated learning update module has a built-in lightweight AM-GRU model and is used to synchronize model parameter gradients through federated learning.

[0103] Optionally, up to 95% of routine data can be processed in real time at the local edge layer. Through precise data screening algorithms, data in normal operating conditions can be quickly identified and processed locally. Only when an abnormality is detected is specific data uploaded to the cloud, namely the abnormal data fragment. This fragment contains detailed waveforms 10 minutes before and after the abnormality occurs, so that subsequent in-depth analysis of the abnormality's development trend and change characteristics can be carried out. At the same time, 15-dimensional feature vectors that have been calculated and refined are also uploaded. These feature vectors condense key information about the battery's operating status. After this optimization, the data transmission volume is reduced by 70% compared to the traditional mode, greatly alleviating the pressure on network transmission. In addition, the local edge layer is equipped with a network disconnection and resumption function. Through the built-in 16MB cache, the cache can store up to 7 days of abnormal data, ensuring that after the network is restored, the abnormal data that has not been uploaded can be transmitted to the cloud completely and accurately, ensuring the continuity and integrity of the data.

[0104] In addition, each gateway is equipped with a lightweight AM-GRU model, specifically optimized for battery data processing and capable of efficient and accurate feature extraction and state prediction. Through federated learning technology, gateways can periodically synchronize model parameter gradients, rather than raw data. This approach plays a key role in protecting user privacy, avoiding the privacy risks associated with the aggregation of raw data from different sites. Furthermore, the synchronization of model parameter gradients across sites enables the model to comprehensively learn the battery aging characteristics of different sites, achieving global optimization of these characteristics. Compared to traditional centralized models, this effectively addresses the overfitting problem caused by data concentration in a single center, significantly improving the model's generalization and prediction accuracy in complex and changing real-world scenarios.

[0105] In an optional embodiment, the system further comprises:

[0106] A digital twin mapping module has a built-in virtual image of the battery pack in the smart battery gateway 12, synchronizes single battery parameters in real time, simulates the evolution of fault scenario parameters through finite element simulation, predicts potential faults and generates personalized maintenance work orders.

[0107] Specifically, an advanced digital twin model of the battery pack is deployed within the gateway, creating a precise virtual image for each individual battery cell and enabling real-time, simultaneous mapping of the battery's electrical and thermal parameters. This model efficiently transforms all types of collected data into intuitive representations in a virtual space, providing operations and maintenance personnel with a clear understanding of the battery's operating status. Furthermore, leveraging finite element simulation technology, the system deeply simulates the evolution of battery parameters under common fault scenarios, such as loose terminals and perforated separators. Through data training and validation, the system can keenly detect potential fault signs 72 hours in advance and generate highly personalized maintenance work orders based on the predicted results. For example, when the system detects an impending abnormality in battery cell No. 8 of battery pack No. 3, it accurately generates a work order, clearly stating, "Replacement of battery cell No. 8 of battery pack No. 3 is recommended; expected failure date is July 15th." This provides a strong basis for operations and maintenance, greatly improving the targeted and timely nature of maintenance work.

[0108] In this embodiment, through multi-dimensional fusion perception technology and dual-drive intelligent algorithms, the signs of aging of the battery's internal microstructure are accurately captured, achieving high-precision estimation of the battery's remaining power and health status; in terms of early warning performance, a multi-dimensional linkage detection model and intelligent early warning mechanism are established to effectively improve the accuracy and timeliness of fault warnings, significantly advancing the warning time; at the management and operation level, the dynamic grouping intelligent balancing strategy and fault inference engine not only improve the efficiency of battery pack management, but also realize the root cause diagnosis of faults and provide maintenance suggestions. At the same time, the edge computing architecture optimizes data processing and protects user privacy.

[0109] Based on the above embodiments, Figure 2 As shown, the present invention also provides an exemplary implementation step of a battery abnormality warning system based on an intelligent battery gateway, and the exemplary implementation steps include:

[0110] Step 100, hardware deployment and connection: Install sensors, deploy smart gateways, connect Bluetooth / network communication modules, and ensure stable hardware operation.

[0111] Specifically, step 100 includes:

[0112] Step 110, sensor installation: The voltage probe, NTC temperature sensor, impedance excitation coil, and other integrated composite parameter measurement units are tightly attached to the surface of the single battery and the electrode connection through a flexible printed circuit board (FPCB), ensuring that the thickness does not exceed 0.5mm to accommodate the battery gap installation. The thermal runaway detection unit is deployed in key locations such as the battery pack heat dissipation channel. After completion, the sensor module 11 is connected to the smart battery gateway 12 via a cable.

[0113] Step 120, gateway and communication equipment configuration: connect the smart battery gateway 12 to the on-site network, configure Bluetooth, serial port or network communication parameters to support data interaction with the local mobile APP and remote monitoring platform; connect the local alarm indicator light to ensure the normal operation of the human-computer interaction module 14.

[0114] Step 200, software configuration and algorithm initialization: load the SOC / SOH algorithm, set the warning threshold and intelligent balancing strategy, and complete parameter initialization.

[0115] Specifically, step 200 includes:

[0116] Step 210, data processing module algorithm loading: In the data processing module of the intelligent battery gateway 12, the improved volumetric Kalman filter (CV-KF) algorithm for SOC estimation and the attention-based gated recurrent unit (AM-GRU) algorithm for SOH prediction are loaded. Based on the type and specifications of the battery, the relevant parameters in the algorithm are initialized, such as the initial parameters of the third-order RC equivalent circuit model and the eigenvector dimension of the AM-GRU algorithm. The algorithm is preliminarily debugged and optimized using battery data collected on-site to ensure that the algorithm can accurately calculate SOC and SOH.

[0117] Step 220, early warning module parameter setting: In the early warning module, set safety thresholds for various parameters, such as the upper and lower limits of cell voltage (e.g., <2.0V or >2.4V), internal resistance increase threshold (e.g., >30%), and temperature threshold (e.g., >60°C). Configure the triggering conditions and methods for multi-level early warnings, determine the color and flashing frequency of the local alarm indicator, the content and method of push notifications on the mobile app, and the alarm prompts and processing procedures on the remote monitoring platform. Conduct simulation tests on the early warning module to verify the accuracy and timeliness of early warnings under different abnormal conditions.

[0118] Step 230, intelligent balancing strategy configuration: cluster the main components (PC1-PC3) of the impedance spectrum of the battery pack based on the K-means algorithm, and divide the batteries into healthy group, decay group, and fault warning group; set differentiated intelligent balancing thresholds according to the charging and discharging conditions, such as Trigger balance, discharge state and Trigger balancing; During the balancing process, monitor the internal resistance change rate in real time. The balancing operation is automatically terminated when the balancing operation is completed. After the configuration is completed, perform the actual balancing test to observe the balancing effect.

[0119] Step 300, system debugging and performance optimization: testing data collection, state estimation, early warning and balancing functions, and optimizing algorithms and communication configurations.

[0120] Specifically, step 300 includes:

[0121] Step 310, functional test: Start the system and check whether the sensor module 11 can accurately collect various parameters of the single battery in real time and transmit the data to the intelligent battery gateway 12; verify the calculation results of the data processing module, compare the actual measurement data with the SOC and SOH values ​​estimated by the algorithm, and evaluate the calculation accuracy; test the response of the early warning module in different abnormal situations to ensure that it can issue multi-level early warning signals in a timely and accurate manner; check the execution of the intelligent balancing strategy and observe the changes in battery pack parameters during the balancing process;

[0122] Step 320, Performance Optimization: Based on the functional test results, optimize system performance. If the SOC or SOH estimation accuracy does not meet the requirements, adjust the algorithm parameters or improve the algorithm model. If data loss or delay occurs during communication, optimize the communication protocol and network configuration. For the intelligent balancing strategy, further optimize the clustering algorithm and threshold settings to improve balancing efficiency and battery pack consistency. Conduct long-term system operation tests, collect actual operating data, and continuously optimize system performance.

[0123] Step 400, system application and operation and maintenance management: access the actual scene for real-time monitoring, generate maintenance work orders based on digital twins, and perform regular maintenance and upgrades.

[0124] Specifically, step 400 includes:

[0125] Step 410, System Application: The debugged and optimized system is formally applied to battery pack monitoring and management scenarios, such as communication base stations and energy storage power stations. Through the mobile app and remote monitoring platform, the battery pack's operating status, SOC, SOH, and other information can be viewed in real time, and abnormality warning notifications can be received. Utilizing the digital twin mapping system, a virtual image of the battery pack is built into the gateway, and single cell parameters are synchronized in real time. Finite element simulation is used to simulate fault scenarios, predicting potential faults 72 hours in advance and generating personalized maintenance work orders.

[0126] Step 420, daily operation and maintenance: regularly inspect and maintain the system's hardware equipment, including sensor calibration, cleaning and performance testing of the smart battery gateway 12; update and upgrade the system software, promptly fix vulnerabilities and optimize functions; analyze the data accumulated during system operation, summarize the operating rules and failure modes of the battery pack, and provide a basis for subsequent operation and maintenance decisions; when the system issues an abnormal warning, promptly troubleshoot and handle the fault according to the warning information and maintenance suggestions.

[0127] In this embodiment, through multi-dimensional fusion perception technology and dual-drive intelligent algorithms, the signs of aging of the battery's internal microstructure are accurately captured, achieving high-precision estimation of the battery's remaining power and health status; in terms of early warning performance, a multi-dimensional linkage detection model and intelligent early warning mechanism are established to effectively improve the accuracy and timeliness of fault warnings, significantly advancing the warning time; at the management and operation level, the dynamic grouping intelligent balancing strategy and fault inference engine not only improve the efficiency of battery pack management, but also realize the root cause diagnosis of faults and provide maintenance suggestions. At the same time, the edge computing architecture optimizes data processing and protects user privacy.

[0128] In addition, it should be noted that the combination of the various technical features in this case is not limited to the combination described in the claims of this case or the combination described in the specific embodiments. All technical features recorded in this case can be freely combined or combined in any way unless there is a contradiction between them.

[0129] It should be noted that the above examples are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples, and many similar variations are possible. All variations directly derived from or associating with the present invention by those skilled in the art are intended to fall within the scope of protection of the present invention.

[0130] The above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, 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. A battery abnormality warning system based on an intelligent battery gateway, characterized in that: include: The sensor module is used to collect the voltage, internal resistance, temperature, thermal runaway status and float charge voltage parameters of the single battery in real time; The intelligent battery gateway includes a data processing module and an early warning module. The data processing module uses an intelligent algorithm to integrate and analyze voltage, internal resistance, and temperature data, calculates the remaining battery capacity (SOC) and health status (SOH), and implements an intelligent balancing strategy. The early warning module generates a multi-level early warning signal based on parameter thresholds and anomaly detection models; Communication module, used for local data interaction and remote platform data upload, supporting Bluetooth, serial port and network communication; Human-computer interaction module, used for visual warning and operation and maintenance management of abnormal conditions; Wherein, the sensor module includes: Composite parameter measurement unit, used to synchronously collect the voltage response curve and AC internal resistance of the single battery using the pulse current injection method, and obtain the charge transfer resistance , electric double layer capacitors Electrochemical impedance spectroscopy (EIS) data; The thermal runaway detection unit is used to identify thermal runaway precursors through a three-dimensional linkage criterion of temperature gradient, internal resistance mutation rate, and voltage second-order derivative. The criterion formula is: ； Where, is the temperature change rate per unit time, is the rate of change of internal resistance relative to the initial value, is the curvature of the voltage curve; The early warning module includes: The five-dimensional feature anomaly detection unit constructs a feature space including voltage, internal resistance, temperature, SOC, and SOH, and uses the isolation forest algorithm to detect the isolation of data points; The Bayesian network inference engine establishes a causal relationship knowledge base containing 56 failure modes, involving float charge voltage parameters, derives the root cause of the failure through posterior probability calculation, and outputs maintenance recommendations.

2. The battery abnormality warning system based on the intelligent battery gateway according to claim 1 is characterized in that: The sensor module further includes a self-calibration unit, which includes: The MEMS temperature and pressure compensation module is used to monitor the ambient air pressure and humidity in real time, and dynamically correct the thermal resistance coefficient of the temperature sensor through a lookup table method. The temperature measurement accuracy after correction is ±0.5°C. The standard voltage source module is used to automatically calibrate the voltage acquisition channel every 24 hours. After calibration, the voltage drift error is less than 0.05%.

3. The battery abnormality warning system based on the intelligent battery gateway according to claim 1 is characterized in that: The SOC calculation algorithm of the data processing module is an improved volumetric Kalman filter algorithm, and the calculation of the battery remaining capacity SOC includes: Set up a third-order RC equivalent circuit model to identify ohmic internal resistance in real time , charge transfer resistance , electric double layer capacitors Etc. model parameters; Construct multi-input measurement equations to integrate voltage , internal resistance, temperature T Construct the state-space equations: ； in, is the open circuit voltage, is the ohmic voltage drop, is the current flowing through the battery, is the polarization voltage, is the rate of change of polarization voltage with time, is the change in polarization voltage caused by the current charging the double-layer capacitance, is the polarization voltage relaxation time constant; An anti-outlier adaptive mechanism is set up, and the measurement residual is judged by Mahalanobis distance. When the limit is exceeded, it switches to square root cubature Kalman filtering. The calculation accuracy of SOC is ±3%.

4. The battery abnormality warning system based on the intelligent battery gateway according to claim 1 is characterized in that: The SOH prediction algorithm of the data processing module is a gated recurrent unit algorithm based on the attention mechanism, including: Construct 12-dimensional aging features and build a time window sliding matrix; Construct an attention mechanism layer to calculate the contribution of historical data to the current SOH by weight, focusing on the characteristics of the accelerated aging stage. The weight formula is: ； in, for t Attention weight at all times, is the hidden layer state, is the hidden layer state at the previous moment, Is a scoring function used to measure the hidden layer state at the previous moment and the current hidden layer state the degree of correlation between them; Construct a physical constraint correction layer and correct the prediction results through the capacity-internal resistance coupling equation: ； in, It represents the change ratio of the measured internal resistance relative to the initial internal resistance and the failure internal resistance. The coefficient of 0.8 indicates that the internal resistance growth has a dominant weight in the SOH assessment. is the capacity attenuation ratio. The coefficient of 0.2 indicates that the capacity attenuation has a relatively small impact. The prediction accuracy of the corrected SOH is ±5%.

5. The battery abnormality warning system based on the intelligent battery gateway according to claim 1 is characterized in that: The intelligent balancing strategy is a dynamic group balancing based on impedance spectrum clustering, including: Construct health status stratification and cluster the principal components of the impedance spectrum using the K-means algorithm to divide the batteries into healthy group, degradation group, and fault warning group; Set up a time-varying threshold trigger mechanism and set differentiated balancing thresholds based on charging and discharging conditions: ； 。 6. The battery abnormality warning system based on the intelligent battery gateway according to claim 1 is characterized in that: The smart battery gateway adopts an edge computing architecture, and the smart battery gateway also includes: Data hierarchical processing module, used to upload abnormal data fragments and 15-dimensional feature vectors, and process regular data locally; The federated learning update module has a built-in lightweight AM-GRU model and is used to synchronize model parameter gradients through federated learning.

7. The battery abnormality warning system based on the intelligent battery gateway according to claim 1 is characterized in that: The system further comprises: A digital twin mapping module includes a built-in virtual image of the battery pack in the smart battery gateway, synchronizes single-cell battery parameters in real time, simulates the evolution of fault scenario parameters through finite element simulation, predicts potential faults, and generates personalized maintenance work orders.

Images