A lithium battery soc-soh joint prediction method and related device

Abstract

The application discloses a lithium battery SOC-SOH joint prediction method and related devices. The application obtains lithium battery multi-modal electrochemical data and generates a working condition label, performs layered data processing based on the working condition label to obtain standardized multi-modal data; determines a dynamic coulomb efficiency according to temperature, rate, last sampling period SOH and self-discharge loss, integrates to obtain real-time available capacity and determine a reference SOH; takes the reference SOH as an aging state reference, performs correlation screening and aging constraint dimension reduction on an electrochemical mechanism feature set to obtain model input features; inputs the model input features into an electrochemical mechanism constraint joint prediction model, outputs SOC, SOH and SOH prediction confidence interval, and realizes closed-loop correction through SOH back substitution and layered update of online incremental data.

Description

Technical Field

[0001] This application belongs to the field of lithium battery state detection technology, specifically relating to a lithium battery SOC-SOH joint prediction method and related device. Background Technology

[0002] Lithium-ion batteries are widely used in energy storage power stations, new energy vehicles, and distributed energy systems. Their operating status directly affects system safety, economy, and service life. State of Charge (SOC) and State of Harmony (SOH) are used to reflect the degree of battery capacity decay and aging. Accurately and stably obtaining SOC and SOH is a crucial foundation for safe battery operation and maintenance, life management, and secondary utilization.

[0003] During long-term charge and discharge, lithium batteries experience irreversible aging phenomena such as lithium loss, active material degradation, and SEI film thickening, leading to capacity reduction, increased internal resistance, and inaccurate state estimation. Especially under complex operating conditions such as low temperature, pulse fast charging, and long-term shallow charge and discharge, battery operating characteristics will change significantly. If the SOC and SOH estimates are inaccurate, it can easily affect thermal runaway warning, charge and discharge control, and operation and maintenance decisions.

[0004] Existing battery state prediction methods mainly include the ampere-hour integration method and data-driven machine learning methods. The ampere-hour integration method typically estimates the state of charge (SOC) based on the integration of charge and discharge currents, and combines it with OCV calibration to reduce errors; data-driven methods use historical data such as current, voltage, and temperature to train prediction models to achieve SOC or state of equilibrium (SOH) estimation. Some solutions also attempt to introduce physical constraint models or transfer learning methods to improve the model's generalization ability.

[0005] However, existing methods still have significant shortcomings: the ampere-hour integration method typically uses a fixed coulombic efficiency, making it difficult to compensate for long-term drift caused by temperature, rate capability, self-discharge, and aging coupling; pure data models lack microscopic mechanism constraints such as dQ / dV phase transition and DRT impedance, easily producing prediction results that violate electrochemical laws; and independent SOC-SOH estimation can lead to mutual amplification of errors. Therefore, a joint prediction method that can integrate electrochemical mechanisms, support online updates, and achieve closed-loop mutual correction of SOC-SOH is needed. Summary of the Invention

[0006] This application addresses the technical problems of existing lithium battery SOC-SOH prediction methods, such as long-term integral drift, insufficient mechanism constraints, poor adaptability across operating conditions, and mutual amplification of estimation errors, by providing a joint prediction method and related apparatus for lithium battery SOC-SOH.

[0007] To achieve the above objectives, this application adopts the following technical solution: The first aspect of this application provides a method for jointly predicting the SOC-SOH of lithium batteries, including the following steps: Acquire multimodal electrochemical data of lithium batteries and generate operating condition labels for the multimodal electrochemical data based on the operating status of lithium batteries; Based on the operating condition labels, multimodal electrochemical data are hierarchically processed to obtain standardized multimodal data; The dynamic coulombic efficiency is determined based on real-time cell temperature, instantaneous charge / discharge rate, SOH of the previous sampling period, and resting self-discharge loss. The real-time available capacity is obtained by integrating the charge / discharge current based on the dynamic coulombic efficiency, and the reference SOH is determined based on the real-time available capacity. An electrochemical mechanism feature set was constructed based on standardized multimodal data. The reference state of steam (SOH) was used as the aging state benchmark. The electrochemical mechanism feature set was then subjected to correlation screening and aging constraint dimensionality reduction to obtain the model input features. Based on the model input features and the baseline SOH, the electrochemical mechanism constraint joint prediction model is trained, and the model input features are input into the electrochemical mechanism constraint joint prediction model to obtain the real-time SOC, real-time SOH and SOH prediction confidence interval; The real-time SOH is used as the process for determining the dynamic coulombic efficiency by substituting the SOH back into the next sampling period. Online incremental data is formed based on the online acquired multimodal electrochemical data. When the incremental update conditions are met, the electrochemical mechanism-constrained joint prediction model is updated in a hierarchical manner based on the online incremental data to obtain the updated SOC-SOH joint prediction results.

[0008] Furthermore, the multimodal electrochemical data includes basic time-series flow cytometry data, capacity increment curve data, and impedance relaxation characteristic data. The basic time-series flow cytometry data includes instantaneous charge and discharge current of the cell, single cell voltage, terminal temperature, ambient temperature, cumulative cycle count, real-time SOC, start-stop time of a single charge and discharge cycle, resting time, constant current charging time, and constant voltage charging time. The capacity increment curve data consists of sampling points of the dQ / dV discrete curve generated based on the changes in charge and voltage within the SOC range. The impedance relaxation characteristic data consists of relaxation time and impedance amplitude data obtained by DRT transformation based on the low-frequency impedance spectrum.

[0009] Furthermore, the operating condition labels include low-temperature standby operating condition, steady-state energy storage operating condition, and high-power pulse fast charging operating condition; Hierarchical data processing of multimodal electrochemical data based on operating condition labels, including: Missing data under different operating conditions are filled in using corresponding filling methods; A two-tiered screening of outlier data is performed based on statistical outlier rules and electrochemical safety boundaries. The filtered data were segmented and standardized according to temperature and magnification zones to obtain standardized multimodal data.

[0010] Furthermore, the dynamic coulombic efficiency is expressed through η(T, I, SOH, Q). loss ) represents, where T is the real-time cell temperature, I is the instantaneous charge / discharge rate, SOH is the SOH of the previous sampling period, and Q is... loss The static self-discharge loss is considered; the real-time available capacity is determined according to the following relationship:

[0011] Among them, Q real For real-time available capacity, I(t) is the charging and discharging current during the sampling time; The baseline SOH is determined based on the real-time available capacity, including: The correspondence between SOC and open-circuit voltage is established based on the equivalent circuit identification parameters, and open-circuit voltage calibration is triggered in the low charge range or the high charge range. The calibrated capacity sequence is subjected to gradient adaptive variable window smoothing to obtain smoothed capacity, and then... Determine the reference SOH; Among them, Q smooth To smooth out capacity, Q rated This refers to the battery's factory rated capacity.

[0012] Furthermore, an electrochemical mechanism feature set was constructed based on standardized multimodal data. Using the baseline SOH as the aging state benchmark, the electrochemical mechanism feature set was subjected to correlation screening and aging constraint dimensionality reduction to obtain the model input features, including: Time-series statistical features, capacity increment phase transition features, and impedance relaxation features are constructed based on standardized multimodal data; Calculate the correlation between each feature and the benchmark SOH, and select highly correlated features; Highly correlated features are filtered based on lithium loss boundary constraints and active material decay boundary constraints, and then the filtered highly correlated features are subjected to dimensionality reduction with monotonically decay constraints to obtain the model input features.

[0013] Furthermore, the electrochemical mechanism-constrained joint prediction model includes a multi-channel aging feature extraction layer, an electrochemical physical constraint layer, a cross-operating condition adaptive fusion layer, and an adaptive weight output layer. The multi-channel aging feature extraction layer is used to extract the deep temporal features corresponding to the temporal statistical features, capacity increment phase transition features, and impedance relaxation features, respectively. The electrochemical physical constraint layer is used to limit the SOH prediction results based on the capacity decay law and the electrochemical aging boundary. The cross-condition adaptive fusion layer is used to adjust model parameters based on sample tasks corresponding to different temperatures, different rates, and different battery materials. The adaptive weighted output layer is used to fuse features from different modalities and output real-time SOC, real-time SOH, and SOH prediction confidence intervals.

[0014] Furthermore, the electrochemical mechanism-constrained joint prediction model is updated hierarchically based on online incremental data, including: During the online inference phase, multimodal real-time data is cached within a preset time range using a time-series sliding window. Incremental update is triggered when the preset number of complete charge-discharge cycles is reached; During the incremental update process, the underlying feature extraction parameters and electrochemical physical constraint parameters in the electrochemical mechanism-constrained joint prediction model are frozen, and the cross-condition adaptive fusion parameters and adaptive weight output parameters are updated.

[0015] A second aspect of this application provides a lithium battery SOC-SOH joint prediction system, comprising: The data acquisition module is used to acquire multimodal electrochemical data of lithium batteries and generate operating condition labels for the multimodal electrochemical data according to the operating status of lithium batteries. The data processing module is used to perform hierarchical data processing on multimodal electrochemical data based on operating condition labels to obtain standardized multimodal data. The capacity calculation module is used to determine the dynamic coulombic efficiency based on the real-time cell temperature, instantaneous charge-discharge rate, SOH of the previous sampling period, and resting self-discharge loss, and to integrate the charge-discharge current based on the dynamic coulombic efficiency to obtain the real-time available capacity, and to determine the reference SOH based on the real-time available capacity. The feature construction module is used to construct an electrochemical mechanism feature set based on standardized multimodal data, and to perform correlation screening and aging constraint dimensionality reduction on the electrochemical mechanism feature set using the benchmark SOH as the aging state benchmark to obtain the model input features; The joint prediction module is used to train the electrochemical mechanism-constrained joint prediction model based on model input features and benchmark SOH, and input the model input features into the electrochemical mechanism-constrained joint prediction model to obtain real-time SOC, real-time SOH and SOH prediction confidence interval; The update module is used to replace the real-time SOH with the SOH in the next sampling period to determine the dynamic coulombic efficiency. It also generates online incremental data based on the online acquired multimodal electrochemical data. When the incremental update conditions are met, the electrochemical mechanism-constrained joint prediction model is updated in a hierarchical manner based on the online incremental data to obtain the updated SOC-SOH joint prediction results.

[0016] A third aspect of this application provides a computer device, including: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the lithium battery SOC-SOH joint prediction method as described above.

[0017] The first aspect of this application provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed as described above for the lithium battery SOC-SOH joint prediction method.

[0018] Compared with the prior art, this application has the following beneficial effects: This application overcomes the problem of existing methods that rely solely on basic electrical time-series data and struggle to characterize microscopic aging mechanisms by acquiring multimodal electrochemical data of lithium batteries and generating operating condition labels. This is achieved through hierarchical data processing based on operating condition labels, which preserves effective features for different operating conditions such as low temperature, steady-state energy storage, and high-power pulse fast charging, avoiding the weakening of aging information under extreme conditions by globally uniform preprocessing. Furthermore, the application establishes a dynamic coulombic efficiency determination process based on real-time cell temperature, instantaneous charge / discharge rate, state of equilibrium (SOH) of the previous sampling period, and resting self-discharge loss, and then substitutes the real-time SOH back into the dynamic coulombic efficiency determination process for the next sampling period. The closed-loop link of ampere-hour integral and health status feedback correction continuously reduces the long-term drift of the ampere-hour integral. By using the benchmark SOH as the aging state benchmark to perform correlation screening and aging constraint dimensionality reduction on the electrochemical mechanism feature set, pseudo-correlation features inconsistent with the battery aging law can be eliminated, improving the consistency of the mechanism input of the model. By using the electrochemical mechanism constraint joint prediction model to synchronously output real-time SOC, real-time SOH and SOH prediction confidence interval, and combining online incremental data for hierarchical incremental updates, the model's cross-operating condition adaptability can be improved while reducing the computing power cost of full retraining, and reducing the problem of mutual amplification of SOC and SOH estimation errors, thereby improving the accuracy, stability and safety early warning capability of lithium battery full life cycle state prediction. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the joint prediction method for SOC-SOH of lithium batteries in this application. Figure 2This is a schematic diagram of the results of the lithium battery SOC-SOH joint prediction method system in this application; Figure 3 This is a schematic diagram of the electronic device structure according to a preferred embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In this application, DRT stands for Distribution of Relaxation Times; SOC stands for State of Charge; SOH stands for State of Health; SEI stands for Solid Electrolyte Interphase; LLI stands for Loss of Lithium Inventory; and LAM stands for Loss of Active Material.

[0023] This application provides a joint prediction method for SOC-SOH of lithium batteries, which can be applied to scenarios such as online monitoring of battery clusters in energy storage power stations, management of power batteries for new energy vehicles, evaluation of battery cascade utilization, and early warning of battery thermal runaway risks. This method addresses the problems of long-term drift in ampere-hour integrals, lack of electrochemical constraints in pure data models, weak generalization ability across operating conditions, and mutual amplification of SOC-SOH estimation errors in existing technologies.

[0024] like Figure 1 As shown in the embodiments of this application, the lithium battery SOC-SOH joint prediction method may include the following steps: S101: Acquire multimodal electrochemical data of lithium battery and generate operating condition labels for the multimodal electrochemical data based on the operating status of lithium battery.

[0025] In this embodiment, the multimodal electrochemical data includes basic time-series flow cytometry data, capacity increment curve data, and impedance relaxation characteristic data.

[0026] Basic time-series streaming data can include instantaneous charging and discharging current of the battery cell, single-cell voltage, terminal temperature, ambient temperature, cumulative cycle count, real-time state of charge (SOC), start-stop time of a single charge / discharge cycle, resting time, constant current charging time, and constant voltage charging time. This basic time-series streaming data is used to characterize the real-time operating status of lithium batteries during charging, discharging, resting, and cyclic aging processes.

[0027] The capacity increment curve data is generated based on the voltage and charge changes within the SOC range. Specifically, after each complete charge-discharge cycle, voltage and charge change data can be collected segmentally within the SOC range to calculate the charge change dQ and voltage change dV, obtaining the sampling points for the dQ / dV discrete curve. The dQ / dV capacity increment curve can be used to reflect the electrochemical phase transition characteristics of the battery within different SOC ranges.

[0028] Impedance relaxation characteristic data are obtained from the low-frequency impedance spectrum through DRT transformation. Specifically, a small AC excitation can be applied after the lithium battery has been left to stand, the low-frequency impedance spectrum can be collected, and the relaxation time and impedance amplitude related to the SEI film, charge transfer, and solid-phase diffusion can be obtained through DRT transformation.

[0029] When acquiring multimodal electrochemical data, various types of data can be bound to cell number, battery pack number and acquisition timestamp, and stored according to cell number, date and operating condition information for subsequent incremental reading, segmented retrieval and historical playback.

[0030] In one optional implementation, multimodal electrochemical data can be acquired in parallel through multiple synchronous acquisition channels, with a uniform sampling period of 1 second, and timestamps recorded synchronously during the acquisition process. All tagged multimodal data is uploaded to a distributed time-series database via a time-series message queue, and stored in partitions with a three-level index based on cell number, date, and operating condition. This approach supports incremental reading, segmented retrieval, and historical playback of large-scale battery cluster operating data.

[0031] In this embodiment, operating condition tags are generated based on the lithium battery's operating status. These tags can include low-temperature standby, steady-state energy storage, and high-power pulse fast charging conditions. Specifically, the low-temperature standby condition corresponds to an operating state where the ambient temperature is below 0°C and the battery has been idle for more than 2 hours without charging or discharging operations; the steady-state energy storage condition corresponds to a low-rate, stable charging and discharging state without instantaneous power pulses; and the high-power pulse fast charging condition corresponds to a high charging / discharging rate with significant instantaneous power fluctuations. By using these operating condition tags, subsequent data processing can employ different strategies for different operating scenarios, avoiding the weakening of effective aging characteristics by using a single processing method.

[0032] In one optional implementation, the low-temperature standby condition can be set to an ambient temperature below 0°C and a static state for more than 2 hours without charging or discharging operations; the steady-state energy storage condition can be set to a stable charge and discharge rate within 0.5C with no instantaneous power pulses; and the high-power pulse fast charging condition can be set to a charge and discharge rate of not less than 1C and an instantaneous power fluctuation exceeding 30% of the rated power. The above-mentioned stratification standards can be adjusted according to different battery types, site operation strategies, and data sampling conditions.

[0033] S102, based on the operating condition label, performs hierarchical data processing on the multimodal electrochemical data to obtain standardized multimodal data.

[0034] In this embodiment, hierarchical data processing includes missing data completion, abnormal data filtering, and segmented standardization processing.

[0035] For missing data, a differentiated completion method can be used according to the operating condition label. For missing data under steady-state energy storage conditions, interpolation can be performed using the weighted average of adjacent time-series sampling points, with higher weights for sampling points closer to the missing point, to preserve the time-series variation trend during stable charging and discharging. For missing data under low-temperature standby conditions, the median of historical samples with the same temperature range and resting time can be used to fill the missing data, reducing the impact of voltage drift on the interpolation results under low-temperature conditions. For missing data under high-power pulse fast charging conditions, a second-order Thevenin equivalent circuit model can be used to identify internal resistance and polarization parameters based on the effective sampling points before and after, and to simulate and complete the missing current and voltage sampling values, preserving the pulse transient characteristics.

[0036] For outlier data, a two-layer screening process can be implemented. The first layer uses box plots to calculate characteristic quartiles and upper and lower boundaries, initially marking outlier sampling points that exceed the boundaries. The second layer introduces electrochemical safety boundaries for fine-tuning, which can include cell voltage boundaries, charge / discharge current boundaries, and temperature boundaries. Data that simultaneously meets the statistical outlier criteria and exceeds the electrochemical safety boundaries is identified as sensor failure anomaly data and removed. Pulse transient data that is only marked as outlier by statistical rules but still falls within the electrochemical safety boundaries can be retained and assigned a lower confidence weight for subsequent feature calculations. This avoids the accidental deletion of effective transient aging information during high-power charge / discharge processes.

[0037] The filtered data can be segmented and standardized according to temperature and scaling factor partitions. Specifically, the temperature can be divided into low-temperature, normal-temperature, and high-temperature ranges, and the scaling factor can be divided into low-scaling, medium-scaling, and high-scaling ranges. Within each temperature and scaling factor partition, the mean and standard deviation are calculated, and the feature data within that partition are standardized using z-scores. Segmented standardization reduces the impact of scale differences between different temperatures and scaling factors on the model input, ensuring that subtle aging features under extreme conditions are not suppressed by the global scale.

[0038] In one optional implementation, the temperature partition may include three intervals: [-20℃, 0℃], [0℃, 25℃], and [25℃, 45℃]; the scaling factor partition may include three intervals: 0.5C and below, 0.5C to 1C, and 1C to 2C. For the feature data within each temperature and scaling factor cross-partition, z-score normalization is performed according to the following relationship:

[0039] Where, x std Here, x represents the original feature data, μ represents the mean of the feature within the corresponding partition, and s represents the standard deviation of the feature within the corresponding partition. By reading the corresponding statistical parameters by partition, the impact of extreme operating condition data on the overall standardization scale can be reduced.

[0040] Through the hierarchical data processing described above, standardized multimodal data can be obtained. This standardized multimodal data is used for subsequent real-time available capacity calculations, electrochemical mechanism feature construction, and input to joint prediction models.

[0041] S103 determines the dynamic coulombic efficiency based on real-time cell temperature, instantaneous charge / discharge rate, SOH of the previous sampling period, and resting self-discharge loss. It then integrates the charge / discharge current based on the dynamic coulombic efficiency to obtain the real-time usable capacity. Finally, it determines the reference SOH based on the real-time usable capacity. In this embodiment, dynamic coulombic efficiency is used to characterize the charge conversion efficiency of a lithium battery under the combined effects of current temperature, rate, aging state, and self-discharge state. Traditional ampere-hour integration typically uses a fixed coulombic efficiency, which is difficult to compensate for long-term integration drift caused by temperature changes, rate changes, battery aging, and static self-discharge. This application corrects the ampere-hour integration process using dynamic coulombic efficiency.

[0042] Specifically, the dynamic coulomb efficiency function is constructed as: η(T, I, SOH, Q) loss ); Where T is the real-time cell temperature, I is the instantaneous charge / discharge rate, SOH is the SOH of the previous sampling period, and Q is the current temperature. lossThis represents the self-discharge loss during rest. The coefficients of the dynamic coulombic efficiency function can be obtained through battery accelerated aging calibration tests and can be continuously updated during subsequent closed-loop correction.

[0043] The real-time available capacity is obtained by integrating the charge and discharge current based on the dynamic coulomb efficiency.

[0044] Among them, Q real The real-time available capacity is represented by I(t), which is the charging and discharging current during the sampling time.

[0045] Furthermore, a correspondence between State of Charge (SOC) and Open Circuit Voltage (OCV) can be established based on equivalent circuit identification parameters, and open circuit voltage calibration can be triggered in low-charge and high-charge ranges. Specifically, a preset number of sampling periods can be used as the identification window to identify the ohmic internal resistance, polarization resistance, and polarization capacitance, and the SOC-OCV curve can be fitted based on the identification parameters. When the battery operates in the low-charge or high-charge range, OCV calibration is triggered, and the standard charge quantity obtained by back-calculation from the OCV is used as a reference to correct the error in the ampere-hour integrated cumulative charge quantity.

[0046] After completing capacity calculation and OCV calibration, the capacity sequence is smoothed using gradient adaptive variable window smoothing to obtain smoothed capacity. Specifically, the difference in capacity gradient between two adjacent cycles can be calculated. When the capacity change is relatively stable, a longer sliding window is used for smoothing to suppress random noise; when the capacity change is abrupt, a shorter sliding window is used for smoothing to avoid over-smoothing out the true capacity decay characteristics.

[0047] The reference SOH can be determined according to the following relationship:

[0048] Among them, Q smooth To smooth out capacity, Q rated This refers to the battery's factory rated capacity.

[0049] Through the above processing, a baseline SOH can be obtained for model training and closed-loop calibration. Simultaneously, this baseline SOH can also serve as input to the dynamic coulomb efficiency function for the next sampling period.

[0050] S104. An electrochemical mechanism feature set is constructed based on standardized multimodal data. The electrochemical mechanism feature set is then subjected to correlation screening and aging constraint dimensionality reduction using the benchmark SOH as the aging state benchmark to obtain the model input features.

[0051] In this embodiment, the baseline SOH obtained in S103 is not only used to characterize the battery health status of the current cycle, but also serves as an aging state benchmark for electrochemical mechanism feature screening, model training, and subsequent model correction. Based on this baseline SOH, the degree of correlation between various electrochemical features and battery capacity decay can be determined, thereby avoiding the misinterpretation of statistical fluctuations unrelated to the aging process as valid features input into the model.

[0052] Specifically, an electrochemical mechanism feature set is constructed based on the standardized multimodal data obtained from S102. The electrochemical mechanism feature set includes time-series statistical features, capacity increment phase transition features, and impedance relaxation features.

[0053] Time-series statistical features are used to characterize the external operating conditions and charging and discharging behavior of lithium batteries. These features may include average temperature rise during charging and discharging, rate of temperature rise change, percentage of constant current charging time, percentage of constant voltage charging time, mean current, current variance, peak current, mean voltage, voltage variance, resting time per cycle, number of days between cycles, and total ampere-hours of charging and discharging.

[0054] Capacity increment phase transition characteristics are used to characterize the phase transition process of battery materials and its changes with aging. These characteristics can include the peak value of the main peak and the peak value of the secondary peak of the dQ / dV curve, the SOC position corresponding to the main peak, the SOC range span of the phase transition peak, the integral area under the peak, the shift of the charge and discharge phase transition peak, and the dQ / dV difference between the high and low SOC segments.

[0055] Impedance relaxation characteristics are used to characterize the growth of the internal interface film, changes in charge transfer resistance, and changes in diffusion processes within a battery. These characteristics can include the SEI film relaxation time, charge transfer relaxation time, solid-phase diffusion relaxation time, impedance amplitude corresponding to each relaxation, the ratio of the real to the imaginary part of the low-frequency impedance, and the rate of impedance change with cycling.

[0056] After obtaining the electrochemical mechanism feature set, the correlation between each feature and the reference SOH is calculated, and highly correlated features are selected based on the correlation. In an optional embodiment, the mutual information coefficient can be used to calculate the correlation between each feature and the reference SOH, and features whose mutual information coefficients meet preset conditions are retained.

[0057] Furthermore, highly correlated features are filtered based on lithium loss boundary constraints and active material decay boundary constraints. By using lithium loss boundary constraints and active material decay boundary constraints, impedance features and capacity increment phase transition features that have a reasonable changing trend with cycle aging can be retained, while spurious correlated features that fluctuate irregularly with aging or contradict the battery aging pattern can be eliminated.

[0058] The filtered highly correlated features are subjected to dimensionality reduction with a monotonically decaying constraint to obtain the model input features. Dimensionality reduction with a monotonically decaying constraint can compress the feature dimension while maintaining the physical meaning of the reduced features consistent with the SOH decay trend, reducing the risk of physically distorted features caused by unconstrained dimensionality reduction.

[0059] Through the above processing, the obtained model input features simultaneously contain runtime timing information, capacity increment phase transition information, and impedance relaxation aging information, and form a corresponding relationship with the benchmark SOH. This can be used for the training, correction, and online inference of subsequent electrochemical mechanism-constrained joint prediction models.

[0060] S105, based on model input features and benchmark SOH, trains the electrochemical mechanism-constrained joint prediction model, and inputs the model input features into the electrochemical mechanism-constrained joint prediction model to obtain real-time SOC, real-time SOH and SOH prediction confidence interval.

[0061] In this embodiment, the electrochemical mechanism-constrained joint prediction model is used to synchronously output real-time SOC, real-time SOH, and SOH prediction confidence intervals based on model input features. Unlike models that estimate SOC or SOH separately, this model places SOC estimation and SOH estimation within the same prediction framework, allowing the two types of state variables to mutually constrain each other during subsequent closed-loop correction.

[0062] Specifically, the electrochemical mechanism-constrained joint prediction model includes a multi-channel aging feature extraction layer, an electrochemical physical constraint layer, a cross-operating condition adaptive fusion layer, and an adaptive weight output layer.

[0063] The multi-channel aging feature extraction layer is specifically a multi-channel iTransformer feature extraction layer; it is used to process timing statistical features, capacity increment phase transition features, and impedance relaxation features respectively, and outputs deep timing features corresponding to different modes. In one specific implementation, three parallel feature extraction branches can be set up to input timing statistical features, dQ / dV phase transition features, and DRT impedance relaxation features respectively, so as to capture long-period aging dependence, short-time pulse change, and impedance attenuation information respectively.

[0064] The electrochemical physical constraint layer is specifically the PINN electrochemical physical constraint layer, used to incorporate capacity decay laws and electrochemical aging boundaries into the model training process. Specifically, a multi-objective loss can be constructed based on SOH prediction error, capacity decay physical constraint error, and electrochemical aging boundary constraint error, enabling the model to reduce prediction errors while avoiding outputs that contradict battery aging laws. For example, capacity decay constraints limit the overall SOH decay trend during cycling, while lithium loss boundary constraints limit the SOH decay rate to within a reasonable aging range.

[0065] The cross-condition adaptive fusion layer is specifically a MAML meta-learning fusion layer; it is used to adjust model parameters according to sample tasks corresponding to different operating conditions, enabling the model to adapt to different operating conditions such as low-temperature standby, steady-state energy storage, and high-power pulse fast charging. In one specific implementation, meta-tasks can be constructed based on battery samples with multiple temperatures, multiple rates, and different material systems, allowing the model to complete correction with a small number of samples when faced with small sample data from new batches of batteries or extreme operating conditions.

[0066] The adaptive weighted output layer dynamically calculates the contribution weights of time-series statistical features, capacity increment phase transition features, and impedance relaxation features to the prediction results, and fuses these modal features to output real-time SOC, real-time SOH, and SOH prediction confidence interval. Real-time SOC reflects the remaining battery capacity, real-time SOH reflects the battery's health status, and the SOH prediction confidence interval characterizes the uncertainty of the SOH prediction results.

[0067] During the offline training phase, the model input features corresponding to historical samples can be used as input, the baseline SOH obtained from S103 can be used as the SOH training label, and the initial model can be trained by combining the corresponding SOC calibration value. During the online operation phase, the model can be corrected based on the model input features obtained from newly acquired data and the baseline SOH, so that the model parameters are adapted to the current cell batch, operating conditions, and aging stage.

[0068] In one optional implementation, after model training is complete, the model can be comprehensively evaluated using mean absolute error (MAE), root mean square error (RMSE), goodness-of-fit R² for the SOH decay curve, and coverage of the SOH prediction confidence interval. If the comprehensive evaluation result does not meet the preset accuracy requirements, the model parameters can be readjusted and training can continue; if the comprehensive evaluation result meets the preset accuracy requirements, the online prediction stage begins.

[0069] S106 uses the real-time SOH as the process for determining the dynamic coulombic efficiency by substituting the SOH back into the next sampling period, and forms online incremental data based on the online acquired multimodal electrochemical data. When the incremental update conditions are met, the electrochemical mechanism-constrained joint prediction model is updated in a hierarchical manner based on the online incremental data to obtain the updated SOC-SOH joint prediction results.

[0070] In this embodiment, the real-time SOH is not only used as the model output, but also as the input variable of the dynamic coulombic efficiency function in the next sampling period, used to correct the subsequent ampere-hour integral capacity calculation. Specifically, the real-time SOH output by S105 is used as the SOH of the next sampling period and substituted back into the dynamic coulombic efficiency function η(T, I, SOH, Q). lossThis updates the coulomb efficiency correction parameter for the next sampling period. The updated dynamic coulomb efficiency is then used to calculate the real-time available capacity.

[0071] The online incremental data originates from the multimodal electrochemical data continuously collected during the online inference phase. Specifically, during online operation, basic time-series flow cytometry data, capacity increment curve data, and impedance relaxation characteristic data of the lithium battery are continuously acquired according to S101, layered data processing is performed according to S102, and the corresponding model input features are constructed according to S104. The latest data collected, processed, and generated online together constitute the online incremental data.

[0072] In one optional implementation, a time-series sliding window can be configured to cache online-acquired multimodal electrochemical data. Once the window capacity reaches a preset time range, the oldest time-series sample is automatically discarded, retaining only the latest data for model calibration or fine-tuning. This approach avoids the need for long-term storage of massive amounts of historical data throughout the entire data lifecycle, reducing the data storage pressure for online updates.

[0073] When the preset number of complete charge-discharge cycles is reached, the incremental update condition is determined to be met, and a hierarchical incremental update is performed on the electrochemical mechanism constraint joint prediction model based on online incremental data. During the hierarchical incremental update process, the low-level feature extraction parameters and electrochemical physical constraint parameters in the model are frozen, and only the cross-condition adaptive fusion parameters and adaptive weight output parameters are updated. Since the low-level feature extraction parameters and electrochemical physical constraint parameters carry relatively stable aging mechanism information, freezing them during the incremental update can reduce computational power consumption and reduce the risk of model physical constraint distortion caused by online small sample updates.

[0074] After the update, the updated electrochemical mechanism-constrained joint prediction model continues to receive real-time model input features and outputs updated real-time SOC, real-time SOH, and SOH prediction confidence intervals. Thus, on the one hand, dynamic coulombic efficiency is corrected through real-time SOH substitution, continuously reducing ampere-hour integral drift; on the other hand, online incremental data drives hierarchical model updates, enabling the model to adapt to differences in cell batches, changes in operating conditions, and aging stages, ultimately improving the accuracy, stability, and online adaptability of lithium battery SOC-SOH joint prediction.

[0075] In one optional implementation, a fixed-memory time-series sliding window is opened during the online inference phase to continuously cache nearly 30 days of multimodal real-time acquired data. Once the window capacity reaches a preset time range, the earliest time-series samples are automatically discarded, and only newly added data is retained for model fine-tuning. An incremental update is triggered after every 500 complete charge-discharge cycles. During the update, the underlying feature extraction parameters and electrochemical-physical constraint parameters are frozen, and only the cross-condition adaptive fusion parameters and adaptive weight output parameters are opened to achieve lightweight online adaptation.

[0076] In one embodiment of this application, such as Figure 2 As shown, a lithium battery SOC-SOH joint prediction system is provided, comprising: The data acquisition module is used to acquire multimodal electrochemical data of lithium batteries and generate operating condition labels for the multimodal electrochemical data according to the operating status of lithium batteries. The data processing module is used to perform hierarchical data processing on multimodal electrochemical data based on operating condition labels to obtain standardized multimodal data. The capacity calculation module is used to determine the dynamic coulombic efficiency based on the real-time cell temperature, instantaneous charge-discharge rate, SOH of the previous sampling period, and resting self-discharge loss, and to integrate the charge-discharge current based on the dynamic coulombic efficiency to obtain the real-time available capacity, and to determine the reference SOH based on the real-time available capacity. The feature construction module is used to construct an electrochemical mechanism feature set based on standardized multimodal data, and to perform correlation screening and aging constraint dimensionality reduction on the electrochemical mechanism feature set using the benchmark SOH as the aging state benchmark to obtain the model input features; The joint prediction module is used to train the electrochemical mechanism-constrained joint prediction model based on model input features and benchmark SOH, and input the model input features into the electrochemical mechanism-constrained joint prediction model to obtain real-time SOC, real-time SOH and SOH prediction confidence interval; The update module is used to replace the real-time SOH with the SOH in the next sampling period to determine the dynamic coulombic efficiency. It also generates online incremental data based on the online acquired multimodal electrochemical data. When the incremental update conditions are met, the electrochemical mechanism-constrained joint prediction model is updated in a hierarchical manner based on the online incremental data to obtain the updated SOC-SOH joint prediction results.

[0077] Specific limitations regarding the lithium battery SOC-SOH joint prediction system can be found in the limitations of the lithium battery SOC-SOH joint prediction method described above, and the corresponding technical effects can be obtained equivalently, so they will not be repeated here. Each module in the aforementioned lithium battery SOC-SOH joint prediction system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0078] Figure 3 An internal structural diagram of a computer device is shown in one embodiment. This computer device may specifically be a terminal or a server. Figure 3As shown, the computer device includes a processor, memory, network interface, display, camera, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a lithium battery SOC-SOH joint prediction method. The display screen can be an LCD screen or an e-ink display screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device casing, or an external keyboard, touchpad, or mouse.

[0079] As will be understood by those skilled in the art, computer equipment Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0080] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0081] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0082] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0083] The above embodiments only illustrate several preferred implementations of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make several improvements and substitutions without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A method for jointly predicting SOC-SOH of lithium batteries, characterized in that, Includes the following steps: Acquire multimodal electrochemical data of lithium batteries and generate operating condition labels for the multimodal electrochemical data based on the operating status of lithium batteries; Based on the operating condition labels, multimodal electrochemical data are hierarchically processed to obtain standardized multimodal data; The dynamic coulombic efficiency is determined based on real-time cell temperature, instantaneous charge / discharge rate, SOH of the previous sampling period, and resting self-discharge loss. The real-time available capacity is obtained by integrating the charge / discharge current based on the dynamic coulombic efficiency, and the reference SOH is determined based on the real-time available capacity. An electrochemical mechanism feature set was constructed based on standardized multimodal data. The reference state of steam (SOH) was used as the aging state benchmark. The electrochemical mechanism feature set was then subjected to correlation screening and aging constraint dimensionality reduction to obtain the model input features. Based on the model input features and the baseline SOH, the electrochemical mechanism constraint joint prediction model is trained, and the model input features are input into the electrochemical mechanism constraint joint prediction model to obtain the real-time SOC, real-time SOH and SOH prediction confidence interval; The real-time SOH is used as the process for determining the dynamic coulombic efficiency by substituting the SOH back into the next sampling period. Online incremental data is formed based on the online acquired multimodal electrochemical data. When the incremental update conditions are met, the electrochemical mechanism-constrained joint prediction model is updated in a hierarchical manner based on the online incremental data to obtain the updated SOC-SOH joint prediction results.

2. The lithium battery SOC-SOH joint prediction method according to claim 1, characterized in that, Multimodal electrochemical data includes basic time-series flow cytometry data, capacity increment curve data, and impedance relaxation characteristic data. The basic time-series flow cytometry data includes instantaneous charge and discharge current of the cell, cell voltage, terminal temperature, ambient temperature, cumulative cycle count, real-time SOC, start-stop time of a single charge and discharge cycle, resting time, constant current charging time, and constant voltage charging time. The capacity increment curve data consists of sampling points of the dQ / dV discrete curve generated based on the changes in charge and voltage within the SOC range. The impedance relaxation characteristic data consists of relaxation time and impedance amplitude data obtained by DRT transformation based on the low-frequency impedance spectrum.

3. The lithium battery SOC-SOH joint prediction method according to claim 1, characterized in that, Operating condition labels include low-temperature standby operating condition, steady-state energy storage operating condition, and high-power pulse fast charging operating condition; Hierarchical data processing of multimodal electrochemical data based on operating condition labels, including: Missing data under different operating conditions are filled in using corresponding filling methods; A two-tiered screening of outlier data is performed based on statistical outlier rules and electrochemical safety boundaries. The filtered data were segmented and standardized according to temperature and magnification zones to obtain standardized multimodal data.

4. The lithium battery SOC-SOH joint prediction method according to claim 1, characterized in that, Dynamic coulombic efficiency is obtained through η(T, I, SOH, Q) loss ) represents, where T is the real-time cell temperature, I is the instantaneous charge / discharge rate, SOH is the SOH of the previous sampling period, and Q is... loss The static self-discharge loss is considered; the real-time available capacity is determined according to the following relationship: Among them, Q real For real-time available capacity, I(t) is the charging and discharging current during the sampling time; The baseline SOH is determined based on the real-time available capacity, including: The correspondence between SOC and open-circuit voltage is established based on the equivalent circuit identification parameters, and open-circuit voltage calibration is triggered in the low charge range or the high charge range. The calibrated capacity sequence is subjected to gradient adaptive variable window smoothing to obtain smoothed capacity, and then... Determine the reference SOH; Among them, Q smooth To smooth out capacity, Q rated This refers to the battery's factory rated capacity.

5. The lithium battery SOC-SOH joint prediction method according to claim 1, characterized in that, An electrochemical mechanism feature set was constructed based on standardized multimodal data. Using the baseline SOH as the aging state benchmark, the feature set was subjected to correlation screening and aging constraint dimensionality reduction to obtain the model input features, including: Time-series statistical features, capacity increment phase transition features, and impedance relaxation features are constructed based on standardized multimodal data; Calculate the correlation between each feature and the benchmark SOH, and select highly correlated features; Highly correlated features are filtered based on lithium loss boundary constraints and active material decay boundary constraints, and then the filtered highly correlated features are subjected to dimensionality reduction with monotonically decay constraints to obtain the model input features.

6. The lithium battery SOC-SOH joint prediction method according to claim 1, characterized in that, The electrochemical mechanism-constrained joint prediction model includes a multi-channel aging feature extraction layer, an electrochemical physical constraint layer, a cross-condition adaptive fusion layer, and an adaptive weight output layer. The multi-channel aging feature extraction layer is used to extract the deep temporal features corresponding to the temporal statistical features, capacity increment phase transition features, and impedance relaxation features, respectively. The electrochemical physical constraint layer is used to limit the SOH prediction results based on the capacity decay law and the electrochemical aging boundary. The cross-condition adaptive fusion layer is used to adjust model parameters based on sample tasks corresponding to different temperatures, different rates, and different battery materials. The adaptive weighted output layer is used to fuse features from different modalities and output real-time SOC, real-time SOH, and SOH prediction confidence intervals.

7. The lithium battery SOC-SOH joint prediction method according to claim 1, characterized in that, Hierarchical incremental updates are performed on the joint prediction model constrained by electrochemical mechanisms based on online incremental data, including: During the online inference phase, multimodal real-time data is cached within a preset time range using a time-series sliding window. Incremental update is triggered when the preset number of complete charge-discharge cycles is reached; During the incremental update process, the underlying feature extraction parameters and electrochemical physical constraint parameters in the electrochemical mechanism-constrained joint prediction model are frozen, and the cross-condition adaptive fusion parameters and adaptive weight output parameters are updated.

8. A lithium battery SOC-SOH joint prediction system, characterized in that, include: The data acquisition module is used to acquire multimodal electrochemical data of lithium batteries and generate operating condition labels for the multimodal electrochemical data according to the operating status of lithium batteries. The data processing module is used to perform hierarchical data processing on multimodal electrochemical data based on operating condition labels to obtain standardized multimodal data. The capacity calculation module is used to determine the dynamic coulombic efficiency based on the real-time cell temperature, instantaneous charge-discharge rate, SOH of the previous sampling period, and resting self-discharge loss, and to integrate the charge-discharge current based on the dynamic coulombic efficiency to obtain the real-time available capacity, and to determine the reference SOH based on the real-time available capacity. The feature construction module is used to construct an electrochemical mechanism feature set based on standardized multimodal data, and to perform correlation screening and aging constraint dimensionality reduction on the electrochemical mechanism feature set using the benchmark SOH as the aging state benchmark to obtain the model input features; The joint prediction module is used to train the electrochemical mechanism-constrained joint prediction model based on model input features and benchmark SOH, and input the model input features into the electrochemical mechanism-constrained joint prediction model to obtain real-time SOC, real-time SOH and SOH prediction confidence interval; The update module is used to replace the real-time SOH with the SOH in the next sampling period to determine the dynamic coulombic efficiency. It also generates online incremental data based on the online acquired multimodal electrochemical data. When the incremental update conditions are met, the electrochemical mechanism-constrained joint prediction model is updated in a hierarchical manner based on the online incremental data to obtain the updated SOC-SOH joint prediction results.

9. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the lithium battery SOC-SOH joint prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1-7, which is the lithium battery SOC-SOH joint prediction method.

Images