Cloud-edge collaborative intelligent bms system, vehicle and energy storage system

Abstract

The application discloses a cloud-edge collaborative intelligent BMS system, a vehicle and an energy storage system, which comprises a vehicle-mounted edge-end BMS device, a cloud-end data center and a vehicle-cloud communication module connecting the two; the vehicle-mounted edge-end BMS device comprises a sampling module, an incremental capacity curve extraction module, a cumulative morphology distance construction module, a battery individual descriptor generation module, a multi-task state estimation module, an active sample screening module, a physical consistency verification module and a hierarchical safety control module. The system constructs a cumulative morphology distance CIMD through the morphology difference of dQ / dV curves of adjacent charge and discharge cycles, and simultaneously uses the CIMD as a battery individual descriptor, a degradation space coordinate of active sample screening and a degradation independent variable of SOH / RUL output physical consistency verification. The application can improve the stability, individual adaptability and safety and reliability of battery health state and residual life estimation.

Description

Technical Field

[0001] This invention belongs to the field of new energy vehicle power battery management technology, specifically involving a cloud-edge collaborative intelligent BMS system, vehicle, and energy storage system. Background Technology

[0002] Lithium-ion batteries are widely used in new energy vehicles, energy storage power stations, and battery swapping systems due to their high energy density, long cycle life, and good environmental adaptability. To ensure the safe, reliable, and efficient operation of the battery system, the battery management system needs to monitor battery voltage, current, temperature, and other operating data in real time, and estimate key states such as state of charge (SOC), state of health (SOH), and remaining life (RUL).

[0003] In practical applications, power batteries operate under complex and dynamic conditions for extended periods. Significant differences in driving habits, charging strategies, ambient temperature, and discharge rates among different vehicles lead to marked individual and condition-specific variations in battery degradation paths. For long-life batteries such as lithium iron phosphate batteries, capacity decay is not significant in the early to mid-life stages, making traditional macroscopic health indicators based on capacity or internal resistance insufficient to reflect early microscopic degradation in a timely manner. Furthermore, existing intelligent battery management systems (BMS) often employ a single state of equilibrium (SOH) estimation model, which is susceptible to local operating conditions, sensor noise, and limited sample migration, resulting in short-term fluctuations, localized divergences, and even non-physical increases in SOH.

[0004] Currently, deep learning models are used for battery state estimation. However, deep learning models typically exhibit black-box characteristics, with their outputs lacking clear physical boundary constraints. If the model output is broadcast directly to the vehicle controller or energy storage control system without verification, it may affect power scheduling, thermal management, or safety protection decisions under model mismatch, sensor anomalies, or extreme operating conditions. Furthermore, the scale of vehicle-side operational data is enormous. Uploading all data to the cloud for training would significantly increase communication bandwidth and cloud computing burden; while random uploading makes it difficult to guarantee that the uploaded data covers critical degradation stages and scarce operating conditions.

[0005] Therefore, how to construct a feature scale that can characterize the micro-degradation of batteries in a physical BMS system, and use this feature scale for individualized modeling, active sample selection, multi-task state estimation and physical safety verification, is a technical problem that current intelligent BMS systems urgently need to solve. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a cloud-edge collaborative intelligent BMS system, vehicle, and energy storage system, which solves the problems of unstable health status estimation of power batteries under cross-operating conditions, limited sample size, and long-term operation, insufficient adaptation to individual differences, low efficiency of vehicle-to-cloud data upload, and lack of physical safety constraints in the output of artificial intelligence models.

[0007] To achieve the above objectives, the present invention provides the following solution: A cloud-edge collaborative intelligent BMS system includes: The vehicle-mounted edge BMS device, the cloud data center, and the vehicle-cloud communication module connecting the vehicle-mounted edge BMS device and the cloud data center; The vehicle-mounted edge BMS device includes: The sampling module is used to collect voltage, current, and temperature data during battery operation; The incremental capacity curve extraction module is used to construct the dQ / dV curve during the battery charging or discharging process based on the voltage, current and time data; The cumulative morphological distance construction module is used to calculate the morphological differences of the dQ / dV curves of adjacent cycles or adjacent comparable segments, and to generate the cumulative morphological distance CIMD based on the morphological differences. The battery individual descriptor generation module is used to generate battery individual descriptors based on the CIMD sequence and dQ / dV curve characteristics of the early operating stage of the battery. The multi-task state estimation module is used to output the battery health status (SOH), remaining life (RUL), and degradation physical parameters in parallel based on the real-time operating condition timing window and the individual battery descriptor. The active sample screening module is used to determine high-value samples to be uploaded based on prediction uncertainty, CIMD degradation space diversity, and domain distribution distance. The physical conformance verification module is used to perform physical conformance verification on SOH and / or RUL outputs with CIMD as the degradation independent variable, and to perform output correction or security control based on the verification results. The graded safety control module is used to perform graded safety actions based on the physical consistency verification results and the real-time battery status. The cloud data center is used to update at least some parameters of the multi-task state estimation module based on the high-value samples, and to send the updated parameters to the vehicle edge BMS device.

[0008] Preferably, the cumulative morphological distance construction module calculates the Pearson correlation coefficient r between adjacent dQ / dV curves and obtains the single-cycle morphological damage amount according to 1-r; the cumulative morphological distance construction module performs outlier removal, sliding median filtering, temperature compensation, and rate compensation on the single-cycle morphological damage amount before accumulating it to obtain a CIMD sequence for characterizing the degree of micro-degradation of the battery.

[0009] Preferably, the multi-task state estimation module includes a feature projection submodule, a shared temporal encoder, a SOH prediction head, a RUL prediction head, and a physical parameter prediction head. The shared temporal encoder employs Transformer, LSTM, GRU, TCN, a one-dimensional convolutional network, or a combination thereof.

[0010] Preferably, the training loss function of the multi-task state estimation module includes SOH prediction loss, RUL prediction loss, physical consistency loss, and monotonicity constraint loss, which are used to make the local health state estimation mutually constrained with the long-term lifetime trend.

[0011] Preferably, the active sample screening module is used to determine high-value samples based on prediction uncertainty, CIMD degradation space diversity, and domain distribution distance, with a comprehensive score of:

[0012] in, U(x j ) To predict uncertainty, D cimd (x j ) For CIMD degradation space diversity, D domain (x j ) For the domain distribution distance, w 1 、w 2 、w 3 For weighting coefficients; when Score(x j ) When the sample exceeds the preset threshold, the active sample screening module uploads the sample to the cloud data center.

[0013] Preferably, the cloud data center is used to receive high-value samples uploaded by the vehicle and generate training signals based on capacity calibration data, historical degradation database of the same type of battery, laboratory degradation data, pseudo-labels of teacher models, semi-supervised learning results or offline capacity test results. It adaptively fine-tunes at least some parameters of the multi-task state estimation module and sends the updated lightweight parameters to the vehicle edge BMS device.

[0014] Preferably, the physical consistency verification module performs verification based on the following degradation relationship:

[0015] in, λ and γ For degenerate physical parameters; When the physical consistency verification module detects that the SOH output does not meet the monotonic degradation trend, the rate of change is abnormal, or the deviation from the degradation relationship exceeds a preset threshold, the vehicle edge BMS device does not directly broadcast the abnormal output to the vehicle bus or energy storage control system, but instead performs output correction or safety control.

[0016] Preferably, the graded safety actions include at least one of the following: increasing sampling frequency, intercepting abnormal outputs, correcting SOH / RUL outputs, enhancing thermal management, limiting charging and discharging power, disconnecting high-voltage relays, reporting fault codes, and providing cloud-based early warnings.

[0017] The present invention also provides a vehicle, including: a power battery pack and a cloud-edge collaborative intelligent BMS system.

[0018] The present invention also provides an energy storage system, including: an energy storage battery cluster and a cloud-edge collaborative intelligent BMS system.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs CIMD by using the difference in the shape of adjacent dQ / dV curves, which can characterize the differences in micro-degradation of the battery in the early stage when capacity decay is not obvious; 2. This invention injects early CIMD features as individual battery descriptors into the multi-task state estimation module, thereby improving the adaptability of the intelligent BMS system to the differences between different individual batteries; 3. This invention simultaneously estimates SOH, RUL and degradation physical parameters, so that the local health state estimation and long-term lifetime trend are mutually constrained, thereby improving the stability of cross-condition estimation; 4. This invention uses proactive sample screening based on prediction uncertainty, CIMD degradation space diversity, and domain distribution distance, which can reduce redundant data uploads and improve the value of cloud-based fine-tuned samples; 5. This invention uses CIMD as a degenerate independent variable for physical consistency verification, which can impose boundary constraints on the output of the deep learning model and reduce the impact of non-physical output on vehicle control or energy storage control. 6. The present invention adopts a hierarchical safety control strategy, which can perform early warning, correction, power limiting, thermal management enhancement or high voltage protection actions when there is abnormal degradation, sensor mismatch or increased thermal risk, thereby improving system safety. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the overall structure of the cloud-edge collaborative intelligent BMS system of the present invention.

[0022] Figure 2 This is a schematic diagram of the functional modules of the vehicle-mounted edge BMS device of the present invention; Figure 3 This is a schematic diagram of the process for constructing CIMD based on the morphological differences of dQ / dV curves in this invention. Figure 4 This is a flowchart illustrating the battery individual descriptor generation and injection multi-task state estimation module of the present invention. Figure 5 This is a schematic diagram of the active sample screening and vehicle-cloud collaborative fine-tuning process of the present invention; Figure 6 This is a schematic diagram of the process of physical consistency verification using CIMD as the degenerate independent variable in this invention; Figure 7 This is a schematic diagram of the hierarchical safety control process of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Example 1 like Figure 1 As shown, the present invention provides a cloud-edge collaborative intelligent BMS system, comprising: The vehicle-mounted edge BMS device, the cloud data center, and the vehicle-cloud communication module connecting the vehicle-mounted edge BMS device and the cloud data center; like Figure 2 As shown, the vehicle-mounted edge BMS device includes: The sampling module is used to collect voltage, current, and temperature data during battery operation; The incremental capacity curve extraction module is used to construct the dQ / dV curve during the battery charging or discharging process based on the voltage, current and time data; The cumulative morphological distance construction module is used to calculate the morphological differences of the dQ / dV curves of adjacent cycles or adjacent comparable segments, and to generate the cumulative morphological distance CIMD based on the morphological differences. The battery individual descriptor generation module is used to generate battery individual descriptors based on the CIMD sequence and dQ / dV curve characteristics of the early operating stage of the battery. The multi-task state estimation module is used to output the battery health status (SOH), remaining life (RUL), and degradation physical parameters in parallel based on the real-time operating condition timing window and the individual battery descriptor. The active sample screening module is used to determine high-value samples to be uploaded based on prediction uncertainty, CIMD degradation space diversity, and domain distribution distance. The physical conformance verification module is used to perform physical conformance verification on SOH and / or RUL outputs with CIMD as the degradation independent variable, and to perform output correction or security control based on the verification results. The graded safety control module is used to perform graded safety actions based on the physical consistency verification results and the real-time battery status. The cloud data center is used to update at least some parameters of the multi-task state estimation module based on the high-value samples, and to send the updated parameters to the vehicle edge BMS device.

[0026] As one embodiment of the present invention, the cumulative morphological distance construction module calculates the Pearson correlation coefficient r between adjacent dQ / dV curves and obtains the single-cycle morphological damage amount according to 1-r.

[0027] Furthermore, let g be the incremental capacity curve corresponding to the k-th cycle. k (V), the incremental capacity curve corresponding to the (k-1)th cycle is g. k-1 (V), calculate the Pearson correlation coefficient r between the two under a unified voltage range and a unified sampling grid. k The single-cycle morphological damage amount is obtained according to the following formula:

[0028] Furthermore, the cumulative morphological distance construction module performs outlier removal, sliding median filtering, temperature compensation, and rate compensation on the single-cycle morphological damage amount before accumulating it to obtain a CIMD sequence used to characterize the degree of microscopic degradation of the battery.

[0029] Furthermore, after performing outlier removal, sliding median filtering, temperature compensation, and magnification compensation on the single-loop morphological damage amount, the corrected single-loop morphological damage amount is obtained, and then accumulated to obtain the CIMD sequence:

[0030] As one embodiment of the present invention, the battery individual descriptor includes at least one of the following: early CIMD mean, early CIMD growth slope, early single-cycle morphological damage mean, dQ / dV peak position, dQ / dV peak height, dQ / dV peak area, early internal resistance change rate, early capacity estimate, and early operating condition statistical characteristics.

[0031] In one embodiment of the present invention, the multi-task state estimation module includes a feature projection submodule, a shared temporal encoder, a SOH prediction head, a RUL prediction head, and a physical parameter prediction head. The shared temporal encoder may employ Transformer, LSTM, GRU, TCN, a one-dimensional convolutional network, or a combination thereof.

[0032] Furthermore, the training loss function of the multi-task state estimation module includes SOH prediction loss, RUL prediction loss, physical consistency loss, and monotonicity constraint loss, which are used to make the local health state estimation mutually constrained with the long-term lifetime trend.

[0033] As one embodiment of the present invention, the active sample screening module calculates the comprehensive score of the samples according to the following formula:

[0034] in, U(x j ) To predict uncertainty, D cimd (x j ) For CIMD degradation space diversity, D domain (x j ) For the domain distribution distance, w 1 、w 2 、w 3 These are the weighting coefficients.

[0035] Furthermore, when the comprehensive score of the sample exceeds a preset threshold, the active sample screening module uploads the corresponding running window, CIMD value, working condition label, model output and statistical features to the cloud data center through the vehicle-cloud communication module.

[0036] As one embodiment of the present invention, the physical consistency verification module performs verification based on the following degradation relationship:

[0037] in, λand γ Degenerate physical parameters can be output from a physical parameter prediction head, or obtained by fitting historical data, laboratory calibration, or cloud model updates.

[0038] When the physical consistency verification module detects that the SOH output does not meet the monotonically degrading trend, has an abnormal rate of change, or deviates from the degradation relationship by more than a preset threshold, the on-board edge BMS device does not directly broadcast the abnormal output to the vehicle bus or energy storage control system. Instead, it performs output correction or safety control. The safety control includes at least one of the following: abnormal output interception, SOH / RUL output correction, enhanced thermal management, charging and discharging power limiting, high-voltage relay disconnection, fault code reporting, and cloud-based early warning.

[0039] In one embodiment of the present invention, the graded safety control module is used to perform graded safety actions based on the physical consistency verification results and the real-time battery status. The graded safety actions include at least one of the following: increasing sampling frequency, intercepting abnormal outputs, correcting SOH / RUL outputs, enhancing thermal management, limiting charge and discharge power, disconnecting high-voltage relays, reporting fault codes, and providing cloud-based early warnings.

[0040] As one embodiment of the present invention, the cloud data center is used to receive high-value samples uploaded by the vehicle, and generate training signals based on capacity calibration data, historical degradation database of the same type of battery, laboratory degradation data, pseudo-labels of teacher models, semi-supervised learning results or offline capacity test results, adaptively fine-tuning at least some parameters of the multi-task state estimation module, and sending the updated lightweight parameters to the vehicle edge BMS device.

[0041] Furthermore, the cloud data center freezes at least a portion of the feature extraction layer in the multi-task state estimation module and updates at least one of the SOH prediction head, RUL prediction head, and physical parameter prediction head.

[0042] As one embodiment of the present invention, the computing unit in the vehicle edge BMS device is implemented by at least one of MCU, DSP, FPGA, ASIC, SoC or automotive-grade AI accelerator.

[0043] As one embodiment of the present invention, the vehicle-to-cloud communication module is at least one of a T-Box, a 5G communication module, a 4G communication module, an Ethernet communication module, a Wi-Fi communication module, or an energy storage system gateway.

[0044] Example 2 This invention also provides a cloud-edge collaborative intelligent BMS system, comprising: an in-vehicle edge BMS device, a cloud data center, and a vehicle-cloud communication module. The vehicle-cloud communication module can be a T-Box, a 5G communication module, a 4G communication module, an Ethernet communication module, a Wi-Fi communication module, or an energy storage system gateway.

[0045] The on-board edge BMS device is installed in new energy vehicles, energy storage power stations, or battery swapping systems to collect battery operation data in real time and perform local state estimation and safety control. The cloud data center is used to manage, label, train, and distribute parameters to high-value samples uploaded from the vehicle.

[0046] In this embodiment, the on-board edge BMS device first acquires voltage, current, temperature, and time data during battery operation through a sampling module. The sampling module may include a voltage sampling analog front-end, a current sensor, a temperature sensor, a sampling bus, an analog-to-digital converter, and an embedded controller interface. The sampled data undergoes time synchronization, outlier removal, and normalization before being fed into subsequent modules.

[0047] The incremental capacity curve extraction module calculates the capacity change based on the sampled current and time data, and constructs a dQ / dV curve by combining it with voltage changes. To reduce the impact of sampling noise and operating condition differences on the curve shape, the system can perform resampling, smoothing filtering, temperature normalization, and rate correction on the dQ / dV curve.

[0048] like Figure 3 As shown, the cumulative morphological distance construction module calculates the morphological similarity of dQ / dV curves for adjacent cycles or adjacent comparable segments under a unified voltage range and a unified sampling grid. In this embodiment, the system calculates the Pearson correlation coefficient r between adjacent dQ / dV curves. k and 1-r k As a single-cycle morphological damage measure, after outlier removal, moving median filtering, temperature compensation, and rate compensation, the single-cycle morphological damage measure is accumulated to obtain the CIMD sequence. This CIMD sequence is used to statistically characterize the degree of micro-degradation of the battery.

[0049] like Figure 4As shown, the battery individual descriptor generation module extracts individual features during the early operation phase of the battery. The early operation phase can be the first N complete charge-discharge cycles, where N can be from 20 to 100, preferably 50. The individual descriptor can include the early CIMD mean, early CIMD growth slope, early single-cycle morphological damage mean, early single-cycle morphological damage variance, dQ / dV main peak position, dQ / dV main peak height, dQ / dV main peak area, curve peak position shift rate, temperature-normalized CIMD growth rate, rate-normalized CIMD growth rate, early internal resistance change rate, early capacity estimate, early coulombic efficiency, resting voltage recovery characteristics, and early operating condition statistical characteristics. These features can be combined to form a 15-dimensional or other-dimensional battery individual descriptor and stored in the vehicle-side non-volatile memory.

[0050] The multi-task state estimation module reads the real-time operating condition timing window and the battery individual descriptor. The operating condition timing window consists of features such as voltage, current, temperature, estimated SOC, time interval, charge / discharge rate, temperature change rate, voltage change rate, and historical statistics. In one implementation, the window length L=64, the feature dimension F=9, and the sliding step size is 8. These parameters can be adjusted according to the vehicle's computing power and sampling frequency.

[0051] The individual battery descriptor is mapped to a background feature vector with the same dimension as the latent space of the main model via the feature projection submodule, and then fused with the temporal window features before being input into a shared temporal encoder. The shared temporal encoder can employ Transformer, LSTM, GRU, TCN, a one-dimensional convolutional network, or a combination thereof. The latent states output by the shared temporal encoder are fed into the SOH prediction head, RUL prediction head, and physical parameter prediction head, respectively, thus outputting SOH, RUL, and degradation physical parameters in parallel. The RUL prediction head uses a non-negative activation function to ensure that the remaining lifetime output is non-negative.

[0052] like Figure 5 As shown, the active sample screening module performs a high-value assessment on the current operating window. Prediction uncertainty U(x) j The CIMD degeneracy space diversity (D) can be obtained through multiple forward propagations, model ensembles, or output variance calculations. cimd (x j The CIMD distance (D) can be calculated by the distance between the current sample CIMD value and the historical uploaded sample CIMD coverage area. domain (x j The score can be calculated using Mahalanobis distance, maximum mean difference, or feature space distance. When the comprehensive score exceeds a preset threshold, the active sample screening module uploads the running window and its corresponding CIMD value, operating condition label, model output, and statistical features to the cloud data center.

[0053] After receiving high-value samples, the cloud data center generates training signals based on capacity calibration data, historical degradation databases of batteries of the same model, laboratory degradation data, pseudo-labels from teacher models, semi-supervised learning results, maintenance and testing data, or offline capacity test results. During cloud-based fine-tuning, the underlying feature extraction layer, location encoding layer, or part of the temporal encoding layer in the multi-task state estimation module can be frozen, updating only at least one of the SOH prediction head, RUL prediction head, and physical parameter prediction head. After fine-tuning, the cloud distributes a lightweight parameter package to the vehicle-mounted edge BMS device. The vehicle-mounted edge BMS device updates its local model parameters after passing version verification, integrity verification, and security verification.

[0054] like Figure 6 As shown, the physical consistency verification module receives the SOH, RUL, and degenerate physical parameters output by the multi-task state estimation module and synchronously reads the CIMD sequence. The physical consistency verification module uses CIMD as the degenerate independent variable and compares the SOH sequence output by the model with a preset degenerate relation. In this embodiment, the degenerate relation is...

[0055] If the SOH output is detected to not meet the monotonic degradation trend, has an abnormal rate of change, or deviates from the degradation relationship by more than a preset threshold, the system will not directly broadcast the abnormal SOH / RUL output to the vehicle controller or energy storage control system, but will instead perform output correction, abnormal marking, or safety control.

[0056] In one implementation, the output correction module performs trend compensation based on historical prediction sequences. The system can compare the Bayesian Information Criterion (BIC) of the linear trend correction model and the constant bias correction model. When the BIC of the linear trend correction model is significantly better than that of the constant bias correction model, the linear trend correction is adopted; otherwise, the constant bias correction is adopted to improve output smoothness and physical consistency.

[0057] like Figure 7 As shown, the graded safety control module performs graded actions based on the physical consistency verification results and the real-time battery status. Level 1 warning involves recording the anomaly and increasing the sampling frequency when the deviation slightly exceeds the first threshold. Level 2 correction involves intercepting the abnormal SOH / RUL output and broadcasting the corrected state estimate when the deviation exceeds the second threshold. Level 3 control involves sending a power limiting or thermal management enhancement request to the vehicle controller when the deviation exceeds the third threshold and is accompanied by a rate of temperature increase, internal resistance increase, or voltage anomaly. Level 4 protection involves triggering a high-voltage relay disconnection, a shutdown request, or energy storage system isolation protection when the deviation exceeds the fourth threshold and multi-sensor cross-verification indicates a serious safety risk.

[0058] Therefore, this invention uses the CIMD obtained by constructing the dQ / dV curve morphology difference to simultaneously generate individual battery descriptors, actively screen samples, and verify physical consistency, enabling the intelligent BMS system to achieve more stable SOH / RUL estimation, more efficient vehicle-cloud collaborative updates, and more reliable safety control under complex dynamic operating conditions.

[0059] Example 3 The present invention also provides a vehicle, including: a power battery pack and a cloud-edge collaborative intelligent BMS system.

[0060] Example 4 The present invention also provides an energy storage system, including: an energy storage battery cluster and a cloud-edge collaborative intelligent BMS system.

[0061] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A cloud-edge collaborative intelligent BMS system, characterized in that, include: The vehicle-mounted edge BMS device, the cloud data center, and the vehicle-cloud communication module connecting the vehicle-mounted edge BMS device and the cloud data center; The vehicle-mounted edge BMS device includes: The sampling module is used to collect voltage, current, and temperature data during battery operation; The incremental capacity curve extraction module is used to construct the dQ / dV curve during the battery charging or discharging process based on the voltage, current and time data; The cumulative morphological distance construction module is used to calculate the morphological differences of the dQ / dV curves of adjacent cycles or adjacent comparable segments, and to generate the cumulative morphological distance CIMD based on the morphological differences. The battery individual descriptor generation module is used to generate battery individual descriptors based on the CIMD sequence and dQ / dV curve characteristics of the early operating stage of the battery. The multi-task state estimation module is used to output the battery health status (SOH), remaining life (RUL), and degradation physical parameters in parallel based on the real-time operating condition timing window and the individual battery descriptor. The active sample screening module is used to determine high-value samples to be uploaded based on prediction uncertainty, CIMD degradation space diversity, and domain distribution distance. The physical conformance verification module is used to perform physical conformance verification on SOH and / or RUL outputs with CIMD as the degradation independent variable, and to perform output correction or security control based on the verification results. The graded safety control module is used to perform graded safety actions based on the physical consistency verification results and the real-time battery status. The cloud data center is used to update at least some parameters of the multi-task state estimation module based on the high-value samples, and to send the updated parameters to the vehicle edge BMS device.

2. The cloud-edge collaborative intelligent BMS system as described in claim 1, characterized in that, The cumulative morphological distance construction module calculates the Pearson correlation coefficient r between adjacent dQ / dV curves and obtains the single-cycle morphological damage amount according to 1-r. The cumulative morphological distance construction module performs outlier removal, sliding median filtering, temperature compensation, and rate compensation on the single-cycle morphological damage amount before accumulating it to obtain a CIMD sequence used to characterize the degree of micro-degradation of the battery.

3. The cloud-edge collaborative intelligent BMS system as described in claim 3, characterized in that, The multi-task state estimation module includes a feature projection submodule, a shared temporal encoder, a SOH prediction head, a RUL prediction head, and a physical parameter prediction head. The shared temporal encoder employs Transformer, LSTM, GRU, TCN, a one-dimensional convolutional network, or a combination thereof.

4. The cloud-edge collaborative intelligent BMS system as described in claim 3, characterized in that, The training loss function of the multi-task state estimation module includes SOH prediction loss, RUL prediction loss, physical consistency loss, and monotonicity constraint loss, which are used to constrain the local health state estimation with the long-term lifetime trend.

5. The cloud-edge collaborative intelligent BMS system as described in claim 4, characterized in that, The active sample screening module is used to determine high-value samples based on prediction uncertainty, CIMD degradation space diversity, and domain distribution distance, with a comprehensive score as follows: in, U(x j ) To predict uncertainty, D cimd (x j ) For CIMD degradation space diversity, D domain (x j ) For the domain distribution distance, w 1 、w 2 、w 3 For weighting coefficients; when Score(x j ) When the sample exceeds the preset threshold, the active sample screening module uploads the sample to the cloud data center.

6. The cloud-edge collaborative intelligent BMS system as described in claim 5, characterized in that, The cloud data center is used to receive high-value samples uploaded from the vehicle and generate training signals based on capacity calibration data, historical degradation database of the same type of battery, laboratory degradation data, pseudo-labels of teacher models, semi-supervised learning results or offline capacity test results. It adaptively fine-tunes at least some parameters of the multi-task state estimation module and sends the updated lightweight parameters to the vehicle edge BMS device.

7. The cloud-edge collaborative intelligent BMS system as described in claim 6, characterized in that, The physical consistency verification module performs verification based on the following degradation relationship: in, λ and γ For degenerate physical parameters; When the physical consistency verification module detects that the SOH output does not meet the monotonic degradation trend, the rate of change is abnormal, or the deviation from the degradation relationship exceeds a preset threshold, the vehicle edge BMS device does not directly broadcast the abnormal output to the vehicle bus or energy storage control system, but instead performs output correction or safety control.

8. The cloud-edge collaborative intelligent BMS system as described in claim 7, characterized in that, The graded safety actions include at least one of the following: increasing sampling frequency, intercepting abnormal outputs, correcting SOH / RUL outputs, enhancing thermal management, limiting charging and discharging power, disconnecting high-voltage relays, reporting fault codes, and providing cloud-based early warnings.

9. A vehicle, characterized in that, include: The power battery pack and the cloud-edge collaborative intelligent BMS system as described in any one of claims 1 to 8.

10. An energy storage system, characterized in that, include: Energy storage battery clusters and the cloud-edge collaborative intelligent BMS system as described in any one of claims 1 to 8.

Images