Battery management system based on AI

Abstract

The invention relates to the technical field of battery management, in particular to an AI-based battery management system, which comprises an acquisition module, an identification module, a judgment module, a screening module, an early warning module and a correction module. According to the method, a plurality of core parameters of the battery compartment which may have structural, thermal, electrical and chemical changes in a plateau environment are collected, voltage deviation is used as a primary screening basis, and combined characteristics of structural stress and shell deformation are introduced, so that the risk battery compartment in a critical or early instability state is further accurately identified; furthermore, the number and the spatial position density of the risk battery cabins in the battery cluster are combined, so that the judgment on the local aggregation trend is realized, and the regional risk of possible spreading of thermal instability or structural abnormality is effectively revealed; the method effectively solves the problem that the existing AI algorithm is only based on a fixed electrical model and a static threshold value, so that the thermal instability and early microscopic fault recognition of the battery compartment structure in the plateau area are insufficient, and the potential safety hazard of the battery pack is caused.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of battery management, and in particular to an AI-based battery management system. BACKGROUND

[0002] With the accelerated promotion of new energy applications, the safety and management efficiency of batteries as core energy storage units have attracted widespread attention. In particular, in complex environmental conditions such as highlands and cold regions, the battery is prone to induce systemic failure in the running process due to problems such as severe voltage fluctuation, great difficulty in thermal management, and rapid attenuation of insulation performance. At the same time, the battery cluster structure is complex, the monitoring data dimension is high, and the traditional management method has a lagging response and limited accuracy in identifying potential abnormalities and predicting evolution trends, which is difficult to meet the actual operation requirements, and the battery operation risk assessment and precise intervention are thus facing greater challenges.

[0003] Chinese patent application publication No. CN116667491A discloses a battery cluster management system and method based on AI edge computing. The system includes a dedicated isolated power supply DCDC module, a power management module, a bus current sampling circuit, a total voltage and insulation resistance sampling circuit, a wired transmission module, a wireless communication module, a TF card storage module, an HDMI screen display module, an MCU main control module, and an AI chip module. The dedicated isolated power supply DCDC module is used to access the positive and negative ends of a certain battery cluster, convert the kilovolt direct current voltage of the battery cluster into the voltage required for system operation, and provide power for other modules while implementing the power management strategy provided by the MCU main control module. The total voltage and insulation resistance sampling circuit samples the battery cluster leakage current, processes the analog signals obtained after sampling the end voltage, and performs analog-to-digital conversion with the MCU main control module to complete the total voltage and insulation resistance sampling of the battery cluster. The bus current sampling circuit processes signals with the MCU main control module to complete the battery cluster charging and discharging current sampling work. The MCU main control module provides the power management module with a set power management strategy, controls the bus current sampling circuit to sample the bus current, controls the total voltage and insulation resistance sampling circuit to sample the total voltage and insulation resistance, transmits the bus current parameters and total voltage and insulation resistance parameters to the AI chip module, and receives the SOC and SOH parameters of the AI chip module. The AI chip module receives the bus current parameters and total voltage and insulation resistance parameters from the MCU main control module, runs the battery cell SOC and SOH parameter estimation algorithm, and controls the wired transmission module and the wireless communication module to transmit the bus current parameters, total voltage and insulation resistance parameters, and SOC and SOH parameters to the battery management system. The wired transmission module / wireless communication module transmits the bus current parameters, total voltage and insulation resistance parameters, and SOC and SOH parameters to the third-party battery management system. The TF card storage module stores the bus current parameter records obtained through the bus current sampling circuit and the MCU main control module, the battery cluster total voltage and insulation resistance parameter records obtained through the total voltage and insulation resistance sampling circuit and the MCU main control module, the SOC and SOH parameter records obtained by running the battery parameter estimation algorithm through the AI chip module, and the MCU main control module and AI chip module operation control command records. The HDMI screen display module displays the historical data of the battery cluster, the estimated parameters, and the emergency situation alarm.

[0004] Therefore, the battery cluster management system and method based on AI edge computing have the following problems: the centralized MCU and AI chip of the system coexist, which may cause single point failure risk; the TF card storage and HDMI display interface of the system are prone to data loss or display abnormalities due to physical wear during on-site maintenance; the calibration and synchronization of the multi-channel sampling circuit of the system increase the system debugging difficulty; and the wired and wireless communication of the system in parallel may cause transmission delay and data consistency problems. SUMMARY

[0005] To this end, the present application provides an AI-based battery management system to overcome the problem of insufficient identification of battery cabin structure thermal instability and early micro-failure in highland areas in the prior art, which causes safety hazards of battery packs, by AI-driven multi-modal parameter coupling analysis and dynamic threshold adaptive adjustment.

[0006] To achieve the above-mentioned purpose, the present application provides an AI-based battery management system, comprising: A collection module is configured to collect the shell shrinkage, support stress, evaporation rate of electrolyte, battery temperature rise rate, and battery voltage of each battery cabin to be processed in the distributed battery cabin cluster in the wind-solar-storage combined power station in the highland area in real time. An identification module is connected with the collection module and configured to identify a plurality of concerned battery cabins according to the battery voltage and a preset voltage threshold, and identify a plurality of risk battery cabins according to the battery voltage, the shell shrinkage, and the support stress of each concerned battery cabin. A determination module is connected with the identification module and the collection module, respectively, and configured to determine the position of each risk battery cabin according to the number and position of the risk battery cabin in the battery cluster to which the distributed battery cabin cluster belongs, to obtain a first determination result and a second determination result. A screening module is connected with the determination module and the collection module, respectively, and configured to screen a plurality of first abnormal battery cabins according to the first determination result and the evaporation rate of each risk battery cabin, and screen a plurality of second abnormal battery cabins according to the second determination result, the battery temperature rise rate, the battery voltage of all the battery cabins to be processed, and a preset artificial intelligence model. An early warning module is connected with the screening module and configured to issue a warning for the battery cabin to be processed that overlaps in the first abnormal battery cabin and the second abnormal battery cabin. A correction module is connected with the screening module and the identification module, respectively, and configured to correct the preset voltage threshold according to the change characteristics of the overlap rate of the first abnormal battery cabin and the second abnormal battery cabin within a preset correction period.

[0007] Further, the identification module comprises: A concerned identification unit is configured to identify a plurality of concerned battery cabins from all the battery cabins to be processed according to the comparison result of the battery voltage and the preset voltage threshold. A risk identification unit is connected with the concerned identification unit and configured to identify a plurality of risk battery cabins from all the concerned battery cabins according to the battery voltage, the shell shrinkage, and the support stress.

[0008] Further, the risk identification unit comprises: a first similarity calculation sub-unit, configured to calculate a first change similarity according to the battery voltage and the shell contraction amount within a preset risk identification duration; a second similarity calculation sub-unit, configured to calculate a second change similarity according to the battery voltage and the support stress within the risk identification duration; a risk identification sub-unit, connected with the first similarity calculation sub-unit and the second similarity calculation sub-unit respectively, configured to identify a plurality of risk battery bays from all the concerned battery bays according to the first change similarity and the second change similarity.

[0009] Further, the risk identification sub-unit is configured to calculate a risk representation coefficient according to the relative deviation of the first change similarity and a preset first change threshold value, and the relative deviation of the second change similarity and a preset second change threshold value, and identify a plurality of risk battery bays from all the concerned battery bays according to a comparison result of the risk representation coefficient and a preset standard representation coefficient.

[0010] Further, the determination module comprises: an acquisition unit, configured to acquire the number of risk battery bays in each battery cluster to obtain a risk number, and acquire the position coordinates of risk battery bays in each battery cluster to obtain a plurality of risk coordinates; a coefficient calculation unit, connected with the acquisition unit, configured to calculate the aggregation representation coefficient of each risk battery bay according to a comparison result of the risk number and a preset number threshold value and the risk coordinates; a position determination unit, connected with the coefficient calculation unit, configured to determine whether each risk battery bay is in an aggregation area according to a comparison result of the aggregation representation coefficient and a preset density threshold value, to obtain a plurality of first determination results and a plurality of second determination results.

[0011] Further, the coefficient calculation unit comprises: a coefficient first calculation sub-unit, configured to calculate the aggregation representation coefficient according to all the risk coordinates and a preset reference coordinate when the risk number is less than the preset number threshold value; a coefficient second calculation sub-unit, configured to calculate the aggregation representation coefficient according to the distance between each risk coordinate and a preset coefficient when the risk number is greater than or equal to the preset number threshold value.

[0012] Further, the position determination unit comprises: a first determining subunit configured to determine that the risk battery compartment is in an aggregation area when the aggregation representation coefficient of each of the risk battery compartments is greater than a preset aggregation coefficient threshold, to obtain a plurality of first determining results; a second determining subunit configured to determine that the risk battery compartment is not in an aggregation area when the aggregation representation coefficient of each of the risk battery compartments is less than or equal to the preset aggregation coefficient threshold, to obtain a plurality of second determining results.

[0013] Further, the screening module comprises: a first screening unit configured to screen a plurality of first abnormal battery compartments from all the risk battery compartments according to a comparison result of the evaporation rate and a preset rate threshold based on the first determining results; a second screening unit configured to input all the battery temperature rise rates and the battery voltages into the preset artificial intelligence model based on the second determining results, to screen a plurality of second abnormal battery compartments.

[0014] Further, the warning module comprises: a selection unit configured to select the to-be-handled battery compartments that overlap in the first abnormal battery compartments and the second abnormal battery compartments, to obtain a plurality of warning battery compartments; a recording unit connected with the selection unit and configured to record when the warning battery compartments are obtained, to obtain a duration; a warning module connected with the recording unit and configured to issue a warning to the warning battery compartments when the duration is greater than a preset duration threshold.

[0015] Further, the correction module comprises: an overlap rate calculation unit configured to calculate an overlap rate of the first abnormal battery compartments and the second abnormal battery compartments at each time in the preset correction period; an overlap fluctuation calculation unit configured to calculate a standard deviation of all the overlap rates, to obtain an overlap fluctuation value; a correction unit connected with the overlap fluctuation calculation unit and configured to correct the preset voltage threshold according to a relative deviation of the overlap fluctuation value and a preset overlap fluctuation threshold when the overlap fluctuation value is greater than a preset overlap fluctuation threshold.

[0016] Compared with the prior art, the beneficial effects of the present application are that by collecting multiple core parameters of the battery cabin that may occur structural, thermal, electrical and chemical changes in the plateau environment, the real-time quantification of the multi-physical field state of the battery cabin is realized, the system uses voltage deviation as the basis for preliminary screening, introduces the combination of structural stress and shell deformation, further accurately identifies the risk battery cabin in the critical or early instability state, and then combines the number and spatial position density of the risk battery cabin in the battery cluster to realize the judgment of the local aggregation trend, effectively revealing the regional risk of possible spread of thermal instability or structural abnormalities. At the same time, the system models and identifies the abnormality dominated by the structural characteristics and the thermal evolution trend dominated by the temperature rise and voltage change respectively, screens the potential abnormal battery cabin in two different dimensions, and ensures that the early warning has higher credibility through overlapping analysis. Finally, the system also introduces a dynamic correction mechanism based on the time sequence fluctuation of the abnormal overlap rate, so that the early warning threshold can adapt to the automatic optimization of the battery state evolution characteristics, improve the response ability and intelligent judgment level of the management system to the nonlinear degradation behavior, significantly enhance the accuracy of capturing early hidden dangers and the stability of early warning, and effectively solve the problem that the existing AI algorithm only based on fixed electrical model and static threshold leads to insufficient identification of battery cabin structural thermal instability and early micro-failure in plateau areas, thereby causing safety hazards of the battery pack.

[0017] Further, first, the voltage threshold is used as the threshold to quickly exclude most normal battery cabins, and the cells outside the voltage deviation range are locked as the focus; further, the voltage change of the focus cabin is combined with the micron-level contraction of the shell and the stress fluctuation of the support, the thermal expansion and contraction characteristics caused by temperature-stress coupling, and the voltage attenuation characteristics caused by electrochemical degradation are mutually verified to accurately select the battery cabin that truly exists structural or thermal instability risk, which can avoid false positives caused by single voltage judgment, and capture micro-damage or internal short circuit trend caused by thermal cycling or load impact in advance by means of real-time feedback of mechanical deformation and stress response, greatly improving the sensitivity and reliability of fault identification.

[0018] Further, first, the time series of the battery voltage and the shell contraction amount are normalized, and the synchronous change of the two in the deformation process induced by thermal cycling or internal short circuit is revealed by the Pearson correlation coefficient; then, the standard deviation of the voltage and the stress fluctuation of the support are normalized and the correlation coefficient is calculated to capture the coupling characteristics between the running load fluctuation and the structural stress response. The comprehensive judgment of the two similarity indexes can accurately distinguish those battery cabins that simultaneously appear abnormal in the electrical and mechanical dimensions, which can avoid false positives caused by isolated mutation of a single signal, and can identify potential instability units caused by thermal-structural interaction earlier, thereby greatly improving the sensitivity and reliability of fault warning.

[0019] Furthermore, the risk characterization coefficient is calculated by weighting and accumulating the relative deviations of the two types of synchronously changing indicators. This considers both the degree of deviation of voltage and deformation synchronicity from the threshold and the intensity of the linkage anomaly between voltage and stress fluctuations. Therefore, it can accurately quantify battery compartments that exhibit anomalies in both electrical and structural responses as high-risk units. Through the calculation of the absolute value of parameter deviations and weight allocation, the system can automatically amplify small but continuous abnormal signals while suppressing false alarms caused by occasional fluctuations. When the risk characterization coefficient exceeds the preset standard characterization coefficient, it indicates that the thermodynamic state of the compartment has entered the danger zone, thereby achieving early and reliable location of potential instability points and significantly improving the accuracy and timeliness of overall fault warnings.

[0020] Furthermore, by statistically analyzing the number of risky battery compartments and combining their spatial coordinates within the battery cluster, a clustering characterization coefficient reflecting the relationship between the distance and number of compartments is constructed. This coefficient, compared with a preset density threshold, accurately distinguishes between "isolated anomalies" and "regional clustering." When the number of battery compartments is small and dispersed, the system will not misjudge it as a large-scale failure. However, when multiple high-risk compartments are densely distributed and close together, the judgment module can promptly identify potential areas of mutual reinforcement of heat conduction, stress superposition, or electromagnetic interference. This supports subsequent refined intervention measures such as local isolation or enhanced cooling, significantly improving the ability to prevent the spread of group failures and overall operational efficiency.

[0021] Furthermore, the coefficient calculation unit employs two adaptive clustering measurement methods for different scales of risk distribution: when there are few risk points, the inverse of the standard deviation of the distance between each risk coordinate and the reference point is calculated, resulting in a larger clustering coefficient as the coordinates are closer to the reference point; when there are many risk points, the inverse of the standard deviation of the distance distribution of the coordinates in the same cluster around each risk coordinate is calculated, amplifying the clustering characteristics of the densest local areas. These two methods complement each other, accurately reflecting the clustering trend from individuals to groups regardless of the number of risk battery compartments. This allows for a smooth transition between early isolated anomalies and large-area cluster risks, providing a reliable quantitative basis for subsequent refined isolation, targeted cooling, or localized maintenance.

[0022] Furthermore, by comparing the aggregation characterization coefficient with the preset aggregation threshold, the risky battery compartments are automatically divided into two categories: "aggregated areas" and "non-aggregated areas." When the aggregation coefficient exceeds the threshold, it indicates that multiple units with abnormal heating or stress are concentrated, which may lead to increased heat diffusion in the area or a chain reaction of fatigue in the overall structure. When the aggregation coefficient is lower than or equal to the threshold, it indicates that the abnormal battery compartments are relatively dispersed and are not likely to form a large-scale risk spread. By combining the electro-thermal-mechanical coupling characteristics, it can accurately identify high-risk blocks that require group collaborative intervention and avoid over-handling isolated anomalies. This improves the accuracy of early warning while optimizing the target allocation of operation and maintenance resources.

[0023] Furthermore, in the first screening unit, a battery compartment identified as a cluster is marked as a first abnormal battery compartment only when the electrolyte evaporation rate exceeds a preset rate threshold, thereby accurately distinguishing the risk of chemical leakage caused by encapsulation failure or internal overheating. In the second screening unit, the temperature rise rate and voltage data of all battery compartments to be treated are input into an artificial intelligence model. By utilizing the model's deep learning capabilities on thermo-electric behavior, second abnormal battery compartments that do not show obvious clustering characteristics but still pose an abnormal risk in dynamic thermal response and voltage change patterns are identified. This approach can capture high-risk compartments with regional chemical leakage and also detect early signs of thermal runaway in dispersed compartments, ensuring comprehensive early warning coverage and the lowest false alarm rate.

[0024] Furthermore, by setting up a cross-screening mechanism and a time-dimensional cumulative criterion, the stability and reliability of anomaly identification are enhanced. Specifically, the module prioritizes battery compartments that simultaneously exhibit both the first abnormality (affected by abnormal evaporation rate and aggregation distribution) and the second abnormality (identified by an artificial intelligence model as exhibiting abnormal thermoelectric trends) as key monitoring targets. This avoids the risk of false alarms caused by fluctuations in a single indicator. At the same time, the recording unit quantifies and tracks the duration of anomalies in these warning battery compartments. Combined with a preset duration threshold, short-term disturbances or occasional fluctuations can be effectively filtered out, thereby improving the effectiveness and accuracy of the warnings. Through a three-dimensional cross-judgment method of "spatial aggregation + trend identification + time continuity," the module further ensures the early detection capability of potential thermal runaway or structural anomalies in battery compartments, providing a reliable basis for scheduling and intervention.

[0025] Furthermore, by calculating the overlap rate of the first and second abnormal battery compartments at each moment within a preset correction period, a temporal overlap rate sequence is formed. The standard deviation of this sequence is then calculated to obtain an overlap fluctuation value, which measures the intensity of fluctuation between the two identification results over time. When this fluctuation value exceeds a set overlap fluctuation threshold, it indicates significant inconsistency or instability in the identification of abnormal battery compartments. Based on this, the system introduces a preset threshold adjustment coefficient, dynamically lowering the preset voltage threshold according to the relative deviation between the overlap fluctuation value and the threshold. This allows for flexible responses to environmental changes or system identification biases, improving identification sensitivity when volatility increases, and promoting earlier exposure of risky battery compartments. This effectively enhances the diagnostic accuracy and stability of the system in distributed scenarios, avoiding missed or delayed detections due to identification bias. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the AI-based battery management system in this embodiment; Figure 2 This is a logic diagram of the attention identification unit determining the battery compartment in this embodiment; Figure 3 The decision logic diagram for calculating the aggregation characterization coefficients in the coefficient calculation unit of this embodiment; Figure 4 This is a logic diagram for the filtering module in this embodiment to filter the first abnormal battery compartment. Detailed Implementation

[0027] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0028] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0029] Please see Figure 1 As shown, this is a schematic diagram of the AI-based battery management system of this embodiment. This embodiment provides an AI-based battery management system, including: The data acquisition module is used to collect data in real time on the shell shrinkage, support stress, electrolyte evaporation rate, battery temperature rise rate, and battery voltage of each battery compartment in the distributed battery compartment cluster of the wind-solar-storage combined power station in the plateau region. An identification module, connected to the acquisition module, is used to identify several battery compartments of interest based on the battery voltage and a preset voltage threshold, and to identify several risk battery compartments based on the battery voltage of each battery compartment of interest, the amount of shell shrinkage, and the stress of the support component. The determination module is connected to the identification module and the acquisition module respectively, and is used to determine the location of each risk battery compartment according to the number and location of the risk battery compartments in the battery cluster to which the distributed battery compartment cluster belongs, and to obtain a first determination result and a second determination result. A screening module, which is connected to the judgment module and the acquisition module respectively, is used to screen out a number of first abnormal battery compartments based on the first judgment result and the evaporation rate of each of the risk battery compartments, and to screen out a number of second abnormal battery compartments based on the second judgment result, the battery temperature rise rate of all the battery compartments to be processed, the battery voltage and a preset artificial intelligence model. An early warning module, which is connected to the screening module, is used to issue an early warning for the battery compartments to be processed that overlap in the first abnormal battery compartment and the second abnormal battery compartment; A correction module, which is connected to the filtering module and the identification module respectively, is used to correct the preset voltage threshold according to the change characteristics of the overlap rate of the first abnormal battery compartment and the second abnormal battery compartment within a preset correction period.

[0030] In this embodiment, the data acquisition module is deployed in a distributed battery compartment cluster of a wind-solar-storage combined power station in a plateau region. Each battery compartment consists of several modular battery packs encapsulated in containers or metal cabinets with heat insulation, dustproof, and wind erosion resistance functions, distributed at multiple points along the wind turbines, photovoltaic arrays, and PCS inverter units. Several battery compartments constitute a battery cluster, and each battery cluster, as a unit with regional control and collaborative management capabilities, undertakes specific loads or energy storage tasks. The operating states of the battery compartments within a cluster are coupled, while the clusters are hierarchically controlled through wired or wireless communication. This cluster-level organizational structure facilitates local autonomy, state clustering, and risk isolation in the complex environment of plateau regions. To address shell shrinkage, high-precision laser displacement sensors or fiber Bragg grating sensors are used to continuously measure micron-level displacement caused by thermal expansion and contraction of the battery compartment shell. For support component stress, thin-film strain gauges or MEMS stress sensors are attached to key structural locations inside the battery compartment to monitor load changes caused by thermal cycling and vibration in real time. For electrolyte evaporation rate, a built-in miniature differential pressure sensor detects minute pressure fluctuations within the shell and estimates the evaporation rate using the mass loss method. For battery temperature rise rate, a distributed thermocouple array or infrared thermal imager records individual cell and local temperature changes, then calculates the temperature rise rate using time-series differential analysis. For battery voltage, a high-precision voltage divider and ADC module are used at the bus port of each battery compartment to acquire the total voltage of the cell cluster and upload it in real time via CAN / LoRa buses. Through the synergy of these multiple sensing and communication technologies, the acquisition module can perform comprehensive, real-time quantitative monitoring of the structure, thermal, electrical, and chemical states of the distributed battery cluster system under conditions of thin-climate high-altitude air pressure, severe diurnal temperature variations, and complex wind and solar loads.

[0031] In this embodiment, the preset artificial intelligence model is a two-layer Long Short-Term Memory (LSTM) binary classification model, specifically designed to identify latent fault trends in battery compartments caused by internal thermoelectric coupling anomalies in non-aggregated states. The model takes the normalized historical sequences of battery temperature rise rate and voltage as input, sequentially passing through two LSTM layers with 64 and 32 units (each containing dropout layers to prevent overfitting), followed by a fully connected layer with 16 neurons to extract deep temporal features, and finally outputting anomaly probability values ​​for risk classification. The training data comes from approximately 5000 sample windows of data from 50 battery compartments in a high-altitude solar-energy storage scenario (including approximately 800 known anomalies annotated by experts). The training, validation, and test sets are divided in an 8:1:1 ratio. During training, the Adam optimizer (learning rate 0.001) is used, with a batch size of 64, a maximum of 50 iterations, and an early stopping strategy for the validation set. After training, the model has approximately 75,000 parameters, and the test set accuracy is 95%, recall is 92%, and precision is 94%. It can effectively combine the dynamic correlation between temperature rise rate and voltage to help the system efficiently identify latent thermal runaway precursors under complex operating conditions.

[0032] The preset voltage threshold refers to the lower limit of the reference voltage used for initial screening of battery compartments of interest. Its value depends on the battery type (e.g., lithium iron phosphate, ternary lithium), operating temperature range, battery state of charge (SOC) characteristic curve, and the overall power station operation strategy. It is typically set between 2.8V and 3.2V; in this embodiment, it is set to 2.95V. This effectively eliminates slight voltage deviations caused by natural fluctuations, ensuring that only battery compartments exhibiting significant voltage anomalies are further identified, thereby improving the targeting and accuracy of subsequent risk identification.

[0033] The preset correction period refers to the time window used to statistically and dynamically adjust the preset voltage threshold. Its value depends on the thermal inertia of the battery system, the data acquisition frequency, and the fluctuation cycle of the photovoltaic and wind power load in the plateau region. It is usually set between 10 minutes and 2 hours. In this embodiment, it is set to 30 minutes. This allows for timely response to identification offsets caused by changes in battery status without interfering with the judgment of normal voltage fluctuation trends. By analyzing the fluctuation trend of the temporal overlap rate of abnormal battery compartments, the voltage judgment benchmark is continuously optimized, enabling the system to maintain a balance between adaptive sensitivity and robustness during long-term operation.

[0034] By collecting multiple core parameters of the battery compartment that may undergo structural, thermal, electrical, and chemical changes in the high-altitude environment, the system achieves real-time quantification of the multi-physics state of the battery compartment. Using voltage deviation as an initial screening criterion, the system introduces the combined characteristics of structural stress and shell deformation to further accurately identify risky battery compartments in critical or early instability states. Furthermore, by combining the number and spatial density of risky battery compartments within the battery cluster, the system assesses local aggregation trends, effectively revealing regional risks that may spread due to thermal instability or structural anomalies. Simultaneously, the system models and identifies anomalies dominated by structural features and thermal evolution trends dominated by temperature rise and voltage changes separately, screening potential abnormal battery compartments in two different dimensions. Overlap analysis ensures higher reliability of the early warning system. Finally, the system introduces a dynamic correction mechanism based on the temporal fluctuations of the anomaly overlap rate, enabling the early warning threshold to automatically optimize according to the battery state evolution characteristics. This improves the management system's responsiveness and intelligent judgment level to nonlinear degradation behavior, significantly enhancing the accuracy of early hazard detection and the stability of early warnings. This effectively solves the problem of insufficient identification of battery compartment structural thermal instability and early micro-faults in high-altitude areas due to existing AI algorithms relying solely on fixed electrical models and static thresholds, thus causing safety hazards to the battery pack.

[0035] Please see Figure 2 As shown, this is the logic diagram for the attention identification unit to determine the battery compartment in this embodiment. In this embodiment, the identification module includes: The attention identification unit is used to determine that the battery compartment to be processed is a battery compartment of interest when the battery voltage is greater than the preset voltage threshold, so as to identify a number of battery compartments of interest. A risk identification unit, connected to the concern identification unit, is used to identify several of the risk battery compartments from all the concern battery compartments based on the battery voltage, the housing shrinkage, and the support stress.

[0036] First, using the voltage threshold as a checkpoint, most normally functioning battery compartments are quickly eliminated, and units with voltage deviations outside the range are identified as the focus. Then, the voltage changes of the focus compartments are combined with the micron-level shrinkage of the shell and the stress fluctuations of the supporting components. By utilizing the material's thermal expansion and contraction characteristics caused by temperature-stress coupling, and the voltage decay characteristics caused by electrochemical degradation, the battery compartments with genuine structural or thermal instability risks are accurately screened. This avoids false alarms caused by relying solely on voltage judgment and, with the help of real-time feedback from mechanical deformation and stress response, can detect microscopic damage or internal short-circuit trends caused by thermal cycling or load impact in advance, significantly improving the sensitivity and reliability of fault identification.

[0037] Specifically, the risk identification unit includes: The first similarity calculation subunit is used to perform maximum and minimum value normalization processing on the battery voltage and shell shrinkage within the preset risk identification time to obtain the voltage normalization sequence and the shrinkage normalization sequence, and to calculate the Pearson correlation coefficient of the voltage normalization sequence and the shrinkage normalization sequence to obtain the first change similarity. The second similarity calculation subunit is used to calculate the standard deviation of all battery voltages from the initial time to each time within the risk identification period, to obtain several voltage fluctuation values; and to calculate the standard deviation of all support stresses from the initial time to each time within the risk identification period, to obtain several stress fluctuation values; and to perform maximum and minimum value normalization processing on all voltage fluctuation values ​​and all stress fluctuation values ​​respectively, to obtain voltage fluctuation normalization sequence and stress fluctuation normalization sequence; and to calculate the Pearson correlation coefficient of voltage fluctuation normalization sequence and stress fluctuation normalization sequence, to obtain the second change similarity. A risk identification subunit, which is connected to the first similarity calculation subunit and the second similarity calculation subunit respectively, is used to identify a number of the risk battery compartments from all the battery compartments of concern based on the first change similarity and the second change similarity.

[0038] The preset risk identification time is a time window used to calculate the similarity of voltage, shrinkage, and stress fluctuation. It depends on the battery thermal response characteristics, data acquisition frequency, and the load fluctuation cycle of plateau wind and solar power generation. It is usually set between 5 minutes and 30 minutes. In this embodiment, it is set to 15 minutes, which can effectively filter out short-term random disturbances while fully capturing the abnormal evolution process of thermal-structural coupling, and achieve high sensitivity identification and low false alarm rate warning of potential instability trends.

[0039] First, the time series of battery voltage and casing shrinkage are normalized, and then the synchronous changes of the two during deformation induced by thermal cycling or internal short circuits are revealed using the Pearson correlation coefficient. Subsequently, the standard deviations of voltage and support stress fluctuations are normalized, and correlation coefficients are calculated to capture the coupling characteristics between operating load fluctuations and structural stress response. The combined judgment of these two similarity indices can accurately distinguish battery compartments exhibiting anomalies in both electrical and mechanical dimensions. This avoids false alarms caused by isolated abrupt changes in a single signal and allows for earlier identification of potential unstable units caused by thermo-structural interactions, thereby significantly improving the sensitivity and reliability of fault warnings.

[0040] Specifically, the risk identification subunit is used to calculate a risk characterization coefficient based on the relative deviation between the first change similarity and a preset first change threshold, and the relative deviation between the second change similarity and a preset second change threshold, Q=w1×︱(Y-Y0) / Y︱+w2×︱(U-U0) / U︱, where Q is the risk characterization coefficient, w1 is the preset first change weight, Y is the first change similarity, Y0 is the preset first change threshold, w2 is the preset second change weight, U is the second change similarity, and U0 is the preset second change threshold. Furthermore, when the risk characterization coefficient is greater than a preset standard characterization coefficient, the battery compartment of concern is determined to be a risk battery compartment, thereby identifying several risk battery compartments.

[0041] The preset first change weight is used to balance the contribution of voltage-contraction synchronization deviation in the risk characterization coefficient. It depends on the primary and secondary position of electro-thermal coupling anomaly in the overall fault evolution and the discrimination effect of the first change similarity in historical fault data. It is usually set between 0.4 and 0.7. In this embodiment, it is set to 0.6, which can enhance the impact of the first change similarity on risk assessment when cell casing contraction anomaly and voltage deviation occur simultaneously, and improve the sensitivity to early thermal expansion and contraction instability trend.

[0042] The preset second variation weight is used to adjust the weight contribution of voltage-stress fluctuation coupling to the comprehensive risk characterization coefficient. It depends on the degree of connection between structural fatigue failure and subsequent thermal runaway or electrical faults, as well as the predictive effect of the second variation similarity in historical samples on fault hits. It is usually set between 0.3 and 0.6. In this embodiment, it is set to 0.4, which can balance the performance of structural stress response in risk assessment, neither over-amplifying accidental impacts nor over-capturing the continuous stress risk caused by load fluctuations.

[0043] The preset first change threshold is a reference boundary used to distinguish between normal operation fluctuations and abnormal structural thermal deformation. Its value depends on the normalized similarity statistical distribution of healthy batteries under typical plateau diurnal temperature difference conditions. It is usually set between 0.85 and 0.95. In this embodiment, it is set to 0.90, which can filter out most small cyclic fluctuations. Only when the synchronous change of voltage and shrinkage exceeds this level is it included in the risk judgment, ensuring a low false alarm rate and accurately capturing actual instability signals.

[0044] The preset second change threshold is a benchmark level used to define the synchronous abnormality of voltage fluctuation and support stress fluctuation. Its determination depends on the normal fluctuation range of electrochemical-mechanical feedback and the system reliability requirements under the high-altitude wind-solar-storage operation environment. It is usually set between 0.80 and 0.90. In this embodiment, it is set to 0.85, which can eliminate short-term small load fluctuations. Risk identification is only triggered when the fluctuations of the two are highly consistent, further reducing the impact of occasional vibration interference on the early warning results, thereby ensuring the accuracy and stability of the risk battery compartment identification.

[0045] The risk characterization coefficient is calculated by weighting and accumulating the relative deviations of two types of synchronously changing indicators. This considers both the degree of deviation of voltage and deformation synchronicity from the threshold and the intensity of the linkage anomaly between voltage and stress fluctuations. Therefore, it can accurately quantify battery compartments that exhibit anomalies in both electrical and structural responses as high-risk units. Through the calculation of the absolute value of parameter deviations and weight allocation, the system can automatically amplify small but continuous abnormal signals while suppressing false alarms caused by occasional fluctuations. When the risk characterization coefficient exceeds the preset standard characterization coefficient, it indicates that the thermodynamic state of the compartment has entered the danger zone, thereby achieving early and reliable location of potential instability points and significantly improving the accuracy and timeliness of overall fault warnings.

[0046] Specifically, the determination module includes: The acquisition unit is used to acquire the number of risk battery compartments in each battery cluster, thereby obtaining the number of risks, and to acquire the two-dimensional rectangular coordinates of the location of the risk battery compartments in each battery cluster, thereby obtaining several risk coordinates. A coefficient calculation unit, connected to the acquisition unit, is used to calculate the aggregation characterization coefficient of each of the risk battery compartments based on the comparison result between the number of risks and the preset number threshold, and the risk coordinates. A location determination unit, connected to the coefficient calculation unit, is used to determine whether each of the risk battery compartments is in an aggregation zone based on the comparison result of the aggregation characterization coefficient and the preset density threshold, so as to obtain several first determination results and several second determination results.

[0047] By statistically analyzing the number of risky battery compartments and combining their spatial coordinates within the battery cluster, a clustering characterization coefficient reflecting the relationship between the distance and number of compartments is constructed. This coefficient is then compared with a preset density threshold to accurately distinguish between "isolated anomalies" and "regional clustering." When the number of battery compartments is small and dispersed, the system will not misjudge it as a large-scale failure. However, when multiple high-risk compartments are densely distributed and close together, the judgment module can promptly identify potential areas of mutual reinforcement of heat conduction, stress superposition, or electromagnetic interference. This supports subsequent refined intervention measures such as local isolation or enhanced cooling, significantly improving the ability to prevent the spread of group failures and overall operational efficiency.

[0048] Please see Figure 3 As shown, this is a logic diagram for the coefficient calculation unit to calculate the aggregation characterization coefficients in this embodiment. In this embodiment, the coefficient calculation unit includes: The first coefficient calculation subunit is used to calculate the reciprocal of the standard deviation of the Euclidean distance from all the risk coordinates to the preset reference coordinates when the number of risks is less than a preset number threshold, so as to obtain the aggregation characterization coefficient. The second coefficient calculation subunit is used to calculate the reciprocal of the standard deviation of the Euclidean distance to the center of all other risk coordinates within a circular range centered on each risk coordinate and with the distance calculated by the preset coefficient as the radius when the number of risks is greater than or equal to the preset number threshold, so as to obtain the aggregation characterization coefficient.

[0049] The preset number threshold is the dividing line between the two judgment methods of "a small number of risk points" and "large-scale clusters". It depends on the total number of modules in the battery cluster, the individual failure tolerance, and the operation and maintenance resource allocation strategy. It is usually set between 3 and 8. In this embodiment, it is set to 5, which can switch to refined isolation analysis when small-scale anomalies are captured in the early stage, and start the group cluster judgment in time when a large number of risk chambers appear, thus balancing detection sensitivity and operation and maintenance costs.

[0050] The preset reference coordinates are the benchmark points for measuring the aggregation of "few risk points". They depend on the geometric layout of the battery cluster, the arrangement of the compartments, and the location of the typical heat distribution center. Usually, the coordinates of the central compartment or the geometric centroid of the battery cluster are selected. In this embodiment, the reference coordinates are set as the geometric centroid coordinates of the center of the battery cluster. This ensures that when the compartments are concentrated around the central area, the aggregation characterization coefficient can reflect the concentration of the structure and heat load to the greatest extent.

[0051] The coefficient calculation unit employs two adaptive clustering measurement methods to address risk distributions of varying scales: when there are few risk points, the inverse of the standard deviation of the distance between each risk coordinate and a reference point is calculated, resulting in a larger clustering coefficient as the coordinates approach the reference point; when there are many risk points, the inverse of the standard deviation of the distance distribution between each risk coordinate and its surrounding cluster coordinates is calculated, amplifying the clustering characteristics of the densest local areas. These two methods complement each other, accurately reflecting the clustering trend from individual to group regardless of the number of risk battery compartments. This allows for a smooth transition between early isolated anomalies and large-scale cluster risks, providing a reliable quantitative basis for subsequent refined isolation, targeted cooling, or localized maintenance.

[0052] Specifically, the position determination unit includes: The first determination subunit is used to determine that the risk battery compartment is in a clustering area when the clustering characterization coefficient of each of the risk battery compartments is greater than a preset clustering coefficient threshold, so as to obtain several first determination results. The second determination subunit is used to determine that the risk battery compartment is not in the clustering area when the clustering characterization coefficient of each risk battery compartment is less than or equal to the preset clustering coefficient threshold, so as to obtain several second determination results.

[0053] The preset clustering coefficient threshold is the boundary used to distinguish whether risky battery compartments exhibit a significant spatial concentration distribution. Its value depends on the typical spacing between battery compartments, the density of arrangement within the cluster, and the tolerance for operation and maintenance intervention. It is usually set between 0.3 and 0.7. In this embodiment, it is set to 0.5, which can promptly identify clustered blocks and trigger group warnings when multiple risky battery compartments are close to each other and the regional risk intensifies. At the same time, it avoids false alarms caused by isolated fluctuations when the compartments are dispersed.

[0054] By comparing the aggregation characterization coefficient with the preset aggregation threshold, the risky battery compartments are automatically divided into two categories: "aggregated areas" and "non-aggregated areas". When the aggregation coefficient exceeds the threshold, it indicates that multiple units with abnormal heating or stress are concentrated, which may lead to increased heat diffusion in the area or a chain reaction of fatigue in the overall structure. When the aggregation coefficient is lower than or equal to the threshold, it indicates that the abnormal battery compartments are relatively dispersed and are not likely to form a large-scale risk spread. By combining the electro-thermal-mechanical coupling characteristics, it can accurately identify high-risk blocks that require group collaborative intervention and avoid over-handling isolated anomalies. This improves the accuracy of early warning while optimizing the target allocation of operation and maintenance resources.

[0055] Please see Figure 4 As shown, this is a logic diagram for the filtering module to filter the first abnormal battery compartment in this embodiment. In this embodiment, the filtering module includes: The first screening unit is used to determine the risk battery compartment as the first abnormal battery compartment based on the first determination result when the evaporation rate is greater than a preset rate threshold. The second screening unit is used to input all the battery temperature rise rates and the battery voltages into the preset artificial intelligence model based on the second determination result, so as to screen out a number of the second abnormal battery compartments.

[0056] The preset rate threshold is the safe limit of the normal evaporation rate of the electrolyte in a high-altitude, low-pressure environment. It depends on the volatility characteristics of the battery chemical system, the permeability coefficient of the casing sealing material, and the ambient air pressure and temperature conditions. It is usually set between 0.2 mg / h and 0.6 mg / h. In this embodiment, it is set to 0.4 mg / h, which can accurately distinguish between the weak volatilization caused by the diurnal temperature difference or air pressure change and the significant evaporation rate caused by the encapsulation failure or thermal overload, thereby ensuring the accuracy and stability of the first abnormal battery compartment screening.

[0057] In the first screening unit, a battery compartment identified as a cluster is marked as the first abnormal battery compartment only when the electrolyte evaporation rate exceeds a preset rate threshold. This accurately distinguishes the risk of chemical leakage caused by encapsulation failure or internal overheating. In the second screening unit, the temperature rise rate and voltage data of all battery compartments to be treated are input into an artificial intelligence model. By utilizing the model's deep learning capabilities on thermo-electric behavior, it identifies second abnormal battery compartments that, although they do not show obvious clustering characteristics, still pose an abnormal risk in dynamic thermal response and voltage change patterns. This approach can capture high-risk compartments with regional chemical leakage and also detect early signs of thermal runaway in dispersed compartments, ensuring comprehensive early warning coverage and the lowest false alarm rate.

[0058] Specifically, the early warning module includes: The selection unit is used to select the overlapping battery compartments to be processed in the first abnormal battery compartment and the second abnormal battery compartment to obtain a number of warning battery compartments. A recording unit, connected to the selection unit, is used to record the duration of the warning battery compartment when it is received. An early warning module, which is connected to the recording unit, is used to issue an early warning to the early warning battery compartment when the duration exceeds a preset duration threshold.

[0059] The preset continuous threshold is the minimum time required for continuous monitoring of abnormal states. It depends on the thermal response characteristics of different battery types, the frequency of fluctuations in the operating environment, and the system's fault tolerance strategy. It is usually set between 10 minutes and 2 hours. In this embodiment, it is set to 30 minutes, which can effectively eliminate false judgments caused by occasional or short-term interference, improve the stability and accuracy of abnormal identification, and ensure that the system responds in a timely manner when there is a real risk of thermal instability, thus avoiding the spread of faults.

[0060] By establishing a cross-screening mechanism and a time-based cumulative criterion, the stability and reliability of anomaly identification are enhanced. Specifically, the module prioritizes battery compartments that simultaneously exhibit both the first anomaly (affected by abnormal evaporation rate and aggregation distribution) and the second anomaly (identified by an artificial intelligence model as exhibiting abnormal thermoelectric trends) as key monitoring targets. This avoids the risk of false alarms caused by fluctuations in a single indicator. Furthermore, the recording unit quantifies and tracks the duration of anomalies in these warning battery compartments. Combined with a preset duration threshold, this effectively filters out short-term disturbances or occasional fluctuations, thereby improving the effectiveness and accuracy of the warnings. Through a three-dimensional cross-judgment method of "spatial aggregation + trend identification + time continuity," the module further ensures the early detection capability of potential thermal runaway or structural anomalies in battery compartments, providing a reliable basis for scheduling and intervention.

[0061] Specifically, the correction module includes: The overlap rate calculation unit is used to calculate the overlap rate of the first abnormal battery compartment and the second abnormal battery compartment at each time within the preset correction period, Pt=∣At∩Bt∣ / ∣At∪Bt∣, where Pt is the overlap rate at time t within the preset correction period, At is the set of the first abnormal battery compartments at time t, and Bt is the set of the second abnormal battery compartments at time t. An overlap fluctuation calculation unit is used to calculate the standard deviation of all the overlap rates to obtain the overlap fluctuation value; A correction unit, connected to the overlapping fluctuation calculation unit, is used to reduce the preset voltage threshold according to the relative deviation between the overlapping fluctuation value and the preset overlapping fluctuation threshold when the overlapping fluctuation value is greater than the preset overlapping fluctuation threshold. H'=H×[1-k×(V-V0) / V0], where H' is the preset voltage threshold after reduction, H is the preset voltage threshold before reduction, V is the overlapping fluctuation value, V0 is the preset overlapping fluctuation threshold, and k is the preset threshold adjustment coefficient.

[0062] The preset threshold adjustment coefficient is a proportional factor used by the system to regulate the correction magnitude when performing dynamic voltage threshold correction. Its value depends on the balance requirements of anomaly identification sensitivity and false alarm tolerance in the target application scenario, and is usually set between 0.05 and 0.25. In this embodiment, it is set to 0.2, which can moderately lower the preset voltage threshold when significant fluctuations are detected, enhance the system's response capability to risk signals, and avoid frequent false alarms due to excessive correction, thereby improving the robustness and practicality of the overall early warning system.

[0063] By calculating the overlap rate of the first and second abnormal battery compartments at each moment within a preset correction period, a temporal overlap rate sequence is formed. The standard deviation of this sequence is then calculated to obtain an overlap fluctuation value, which measures the intensity of fluctuation between the two identification results over time. When this fluctuation value exceeds a set overlap fluctuation threshold, it indicates significant inconsistency or instability in the identification of abnormal battery compartments. Based on this, the system introduces a preset threshold adjustment coefficient, dynamically lowering the preset voltage threshold according to the relative deviation between the overlap fluctuation value and the threshold. This allows for flexible responses to environmental changes or system identification biases, improving identification sensitivity when volatility increases, and promoting earlier exposure of risky battery compartments. This effectively enhances the diagnostic accuracy and stability of the system in distributed scenarios, avoiding missed or delayed detections due to identification bias.

[0064] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-based battery management system, characterized by, The method comprises the following steps: The acquisition module is used to acquire the shell shrinkage, support stress, evaporation rate of electrolyte, battery temperature rise rate and battery voltage of each battery cabin to be processed in the distributed battery cabin cluster of the high-altitude wind-solar storage combined power station in real time; The identification module is connected with the acquisition module and is used to identify a plurality of concerned battery cabins according to the battery voltage and a preset voltage threshold, and identify a plurality of risk battery cabins according to the battery voltage, the shell shrinkage and the support stress of each concerned battery cabin; The determination module is connected with the identification module and the acquisition module respectively, and is used to determine the position of each risk battery cabin according to the number and position of the risk battery cabin in the battery cluster to which the distributed battery cabin cluster belongs, to obtain a first determination result and a second determination result; The screening module is connected with the determination module and the acquisition module respectively, and is used to screen a plurality of first abnormal battery cabins according to the first determination result and the evaporation rate of each risk battery cabin, and screen a plurality of second abnormal battery cabins according to the second determination result, the battery temperature rise rate of all the battery cabins to be processed, the battery voltage and a preset artificial intelligence model; The warning module is connected with the screening module and is used to issue a warning for the battery cabin to be processed which is overlapped in the first abnormal battery cabin and the second abnormal battery cabin; The correction module is connected with the screening module and the identification module respectively, and is used to correct the preset voltage threshold according to the change characteristics of the overlap rate of the first abnormal battery cabin and the second abnormal battery cabin in a preset correction period.

2. The AI-based battery management system of claim 1, wherein, The identification module comprises: The concerned identification unit is used to identify a plurality of concerned battery cabins from all the battery cabins to be processed according to the comparison result of the battery voltage and the preset voltage threshold; The risk identification unit is connected with the concerned identification unit and is used to identify a plurality of risk battery cabins from all the concerned battery cabins according to the battery voltage, the shell shrinkage and the support stress.

3. The AI-based battery management system of claim 2, wherein, The risk identification unit comprises: The first similarity calculation subunit is used to calculate a first change similarity according to the battery voltage and the shell shrinkage in a preset risk identification time length; The second similarity calculation subunit is used to calculate a second change similarity according to the battery voltage and the support stress in the risk identification time length; The risk identification subunit is connected with the first similarity calculation subunit and the second similarity calculation subunit respectively, and is used to identify a plurality of risk battery cabins from all the concerned battery cabins according to the first change similarity and the second change similarity.

4. The AI-based battery management system of claim 3, wherein, The risk identification subunit is used to calculate a risk representation coefficient according to the relative deviation of the first change similarity and a preset first change threshold, and the relative deviation of the second change similarity and a preset second change threshold, and identify a plurality of risk battery cabins from all the concerned battery cabins according to the comparison result of the risk representation coefficient and a preset standard representation coefficient.

5. The AI-based battery management system of claim 4, wherein, The determination module comprises: The acquisition unit is configured to acquire a number of the risk battery compartments in each of the battery clusters to obtain a risk number, and acquire position coordinates of the risk battery compartments in each of the battery clusters to obtain a plurality of risk coordinates; The coefficient calculation unit is connected with the acquisition unit and configured to calculate the aggregation representation coefficient of each of the risk battery compartments according to a comparison result of the risk number and a preset number threshold and the risk coordinates; The position determination unit is connected with the coefficient calculation unit and configured to determine whether each of the risk battery compartments is in an aggregation area according to a comparison result of the aggregation representation coefficient and a preset density threshold to obtain a plurality of first determination results and a plurality of second determination results.

6. The AI-based battery management system of claim 5, wherein, The coefficient calculation unit includes: The coefficient first calculation sub-unit is configured to calculate the aggregation representation coefficient according to all the risk coordinates and a preset reference coordinate when the risk number is less than the preset number threshold; The coefficient second calculation sub-unit is configured to calculate the aggregation representation coefficient according to each of the risk coordinates and a preset coefficient calculation distance when the risk number is greater than or equal to the preset number threshold.

7. The AI-based battery management system of claim 6, wherein, The position determination unit includes: The first determination sub-unit is configured to determine that each of the risk battery compartments is in the aggregation area when the aggregation representation coefficient of each of the risk battery compartments is greater than a preset aggregation coefficient threshold to obtain a plurality of first determination results; The second determination sub-unit is configured to determine that each of the risk battery compartments is not in the aggregation area when the aggregation representation coefficient of each of the risk battery compartments is less than or equal to the preset aggregation coefficient threshold to obtain a plurality of second determination results.

8. The AI-based battery management system of claim 7, wherein, The screening module includes: The first screening unit is configured to screen a plurality of first abnormal battery compartments from all the risk battery compartments according to a comparison result of the evaporation rate and a preset rate threshold based on the first determination results; The second screening unit is configured to screen a plurality of second abnormal battery compartments by inputting all the battery temperature rise rates and the battery voltages into the preset artificial intelligence model based on the second determination results. 9.The AI-based battery management system of claim 8, wherein, The early warning module includes: The selection unit is configured to select the to-be-handled battery compartments that overlap in the first abnormal battery compartments and the second abnormal battery compartments to obtain a plurality of early warning battery compartments; The recording unit is connected with the selection unit and configured to record when the early warning battery compartments are obtained to obtain a duration; The early warning module is connected with the recording unit and configured to issue an early warning for the early warning battery compartments when the duration is greater than a preset duration threshold.

10. The AI-based battery management system of claim 9, wherein, The correction module includes: The overlap rate calculation unit is configured to calculate an overlap rate of the first abnormal battery compartments and the second abnormal battery compartments at each time in the preset correction period; The overlap fluctuation calculation unit is configured to calculate a standard deviation of all the overlap rates to obtain an overlap fluctuation value; The correction unit is connected with the overlap fluctuation calculation unit and configured to correct the preset voltage threshold according to a relative deviation of the overlap fluctuation value and a preset overlap fluctuation threshold when the overlap fluctuation value is greater than the preset overlap fluctuation threshold.

Images