Battery capacity calibration and health factor extraction method for low-voltage station area

By collecting and weighted fusion of battery data in real time, the accuracy and robustness of capacity calibration and health factor extraction in low-voltage power grid energy storage systems have been solved, enabling efficient and intelligent assessment and management of battery status.

CN121477010APending Publication Date: 2026-02-06QUJING POWER SUPPLY BUREAU YUNNAN POWER GRID CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202511792881.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-01
Publication Date
2026-02-06

AI Technical Summary

Technical Problem

Existing technologies for battery capacity calibration and health factor extraction in low-voltage power grid energy storage systems are unable to adapt to complex and dynamic operating conditions, resulting in decreased accuracy and applicability of capacity estimation, lagging health factor assessment behind the actual battery condition, insufficient fusion and utilization of multi-source heterogeneous data, and difficulty in fully mining the correlation information between data.

Method used

By acquiring real-time battery terminal voltage, current, temperature and SOC data, the equivalent capacity is calculated using the ampere-hour integration method. Combined with sliding window weighted fitting and operating condition adaptive weight correction, multi-dimensional health factors are extracted and weighted fused to form a comprehensive health characterization.

Benefits of technology

It significantly improves the accuracy and real-time performance of capacity calibration, can dynamically adjust parameters, reflect the battery aging process, reduce the impact of abnormal operating conditions, improve the accuracy and robustness of health monitoring, and support online tracking of battery capacity and health status throughout the entire life cycle.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121477010A_ABST
    Figure CN121477010A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of energy storage battery health state evaluation, and provides a low-voltage station area battery capacity calibration and health factor extraction method, which comprises the following steps: acquiring battery end voltage, current, temperature and SOC data in real time; a complete charge-discharge cycle is automatically divided according to a preset rule, and the equivalent discharge capacity is calculated through an ampere-hour integral; constructing an exponential type working condition correction factor based on the circulating temperature and the root-mean-square current to correct the capacity, and obtaining a dynamic calibration capacity in a sliding window by adopting exponential weight or weighted least square fitting; meanwhile, multi-dimensional health factors are extracted from voltage, current, temperature and SOC signals, and comprehensive health characterization is formed through adaptive weighted fusion in a data driving mode. According to the method, online and dynamic capacity estimation and multi-dimensional health assessment of the energy storage battery in the transformer area under the complex and unsteady working condition are achieved, the capacity calibration precision and the life prediction reliability are remarkably improved, and intelligent operation and maintenance and risk early warning are facilitated.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage battery state of health evaluation, and particularly relates to a low-voltage power supply area battery capacity calibration and health factor extraction method. BACKGROUND

[0002] With the development of distributed power supply and smart grid, low-voltage power supply area energy storage gradually becomes an important means to smooth load fluctuation, improve power supply reliability and promote renewable energy access. Lithium ion batteries are widely used due to their high energy density and long cycle life, and their capacity calibration, SOH (state of health) evaluation and life prediction are directly related to the safety and economy of the energy storage system.

[0003] The prior art mainly uses laboratory standard constant current-constant voltage (CC-CV) calibration, current integration (Ah measurement) in the BMS and internal resistance measurement methods for health evaluation; in recent years, data-driven methods based on temperature rise, terminal voltage characteristics and machine learning have also been gradually introduced.

[0004] The existing low-voltage power supply area energy storage system battery capacity calibration and health factor extraction technology has the following shortcomings: Traditional methods rely on idealized experimental conditions for capacity calibration, making it difficult to accurately evaluate the complex and dynamic working environment of the actual area, resulting in a decrease in the accuracy and applicability of capacity estimation; some methods do not fully consider the random disturbance effects of charging and discharging current, temperature and other working condition factors during data processing, and are easily affected by abnormal data and sudden working conditions, resulting in large fluctuations in capacity and health evaluation results; the selection and weight distribution of health factors generally use fixed or empirical setting methods, lack adaptive optimization based on actual operating data, and cannot dynamically reflect the importance changes of each health factor at different life stages and working conditions, resulting in a lag in the comprehensive health characterization of the battery true state; the existing methods still have deficiencies in the fusion and utilization of multi-source heterogeneous data, making it difficult to fully mine the correlation information between data and limiting the accuracy and comprehensiveness of health factor extraction.

[0005] Therefore, it is necessary to provide a low-voltage power supply area battery capacity calibration and health factor extraction method to solve the problems in the prior art. SUMMARY

[0006] The purpose of the present application is to solve the problems in the prior art and provide a low-voltage power supply area battery capacity calibration and health factor extraction method to solve the problems in the background art.

[0007] In order to achieve the above purpose, the present application adopts the following technical solutions: A low-voltage power supply area battery capacity calibration and health factor extraction method, comprising the following steps: S1. Real-time acquisition of battery terminal voltage V(t), current I(t), temperature T(t), and SOC data in the energy storage system of the distribution area; S2. The collected data is automatically divided into several complete charge-discharge cycles according to the preset cycle division rules; the equivalent capacity is calculated using the ampere-hour integration method for each complete charge-discharge cycle. ; S3, regarding the above The dynamic calibration capacity is obtained by performing adaptive weight correction based on operating conditions and using sliding window weighted fitting. ; S4. Extract multidimensional health factors from the voltage, current, temperature and SOC data, and form a comprehensive health characterization through weighted fusion.

[0008] Preferably, the preset cycle division rule in S2 is: A loop is considered to have ended and a new loop begins when any of the following conditions are met: (a) Current is continuous and in the same direction and the SOC changes monotonically. ; (b) Absolute value of current And the duration exceeded 30 minutes, of which .

[0009] Preferably, the ampere-hour integration method in S2 is: ; in, This is the start time; The termination time; Let be the instantaneous current within the cycle; when the actual sampling is discrete data, the above formula can be approximated as: ; in, For the first Number of sampling points in the next loop This represents the sampling time interval.

[0010] Preferably, the adaptive weight correction for operating conditions in S3 adopts an exponential parameterization form: ; in, This is the average temperature of the cycle. This is the root mean square value of the cyclic current; parameter The data is obtained by log-linearizing historical data and then fitting it using the least squares method or ridge regression.

[0011] Preferably, the sliding window weighting in S3 uses exponentially decreasing weights: ; in, A positive decay coefficient controls the rate of decrease of the weight.

[0012] Preferably, the health factors extracted in S4 include at least one of the following: Normalized capacity: ; in, This is the initial standard capacity of the battery; Normalized internal resistance: ; in, The initial internal resistance, This represents the resistance within the current cycle. After the complete cycle is completed and the system has been allowed to stand for 30 minutes, a 10-second 1C discharge pulse is applied. The voltage difference 10ms before and after the pulse is divided by the current to obtain the result. Isovoltage rise time during charging phase: ; in, This is the actual charging cutoff voltage; Discharge phase isothermal time: ; in, This refers to the battery's nominal voltage.

[0013] Preferably, S4 further includes performing variational mode decomposition or ensemble empirical mode decomposition on the voltage or current curve, and calculating the energy of each intrinsic mode component. ; and the Pearson correlation coefficient between the energy of each intrinsic mode component and SOH. The absolute value is normalized and used as the weight to obtain the result, i.e. .

[0014] Preferably, the method further includes performing a 1C constant current-constant voltage standard capacity calibration every 30 to 100 cycles to correct for long-term algorithm drift; the standard capacity calibration calculation formula is as follows: ; in, For standard constant current, and These are the start and end times of constant current charging and constant voltage charging, respectively. This refers to the instantaneous current during the constant voltage phase. For scenarios where the sampled data is discrete, the integral part can be approximated as a summation: ; in, This represents the current value at the kth sampling point during the constant voltage phase. The time interval between two adjacent sampling points; This represents the total number of sampling points during the constant pressure phase.

[0015] Preferably, in step S3, before performing the operating condition weight correction, the current data is preprocessed by median filtering and moving average.

[0016] Preferably, when the dynamically calibrated capacity after sliding window weighted fitting is lower than the initial standard capacity for three consecutive times... When the battery reaches 70% capacity, it is considered to have reached the end of its lifespan.

[0017] The present invention discloses a method for calibrating the capacity of batteries in low-voltage radio stations and extracting health factors, which has the following beneficial effects.

[0018] This invention, by integrating a data-driven and condition-adaptive capacity calibration algorithm, can effectively filter abnormal operating conditions and observation noise in the complex actual operating environment of low-voltage power grid areas, significantly improving the accuracy and real-time performance of capacity calibration and overcoming the dependence of existing capacity estimation methods on ideal experiments or single operating conditions. The proposed adaptive window capacity fusion and operating condition weight correction method can dynamically adjust capacity estimation parameters for diverse operating conditions encountered by energy storage batteries in actual operation, such as charging and discharging current fluctuations, extreme temperatures, and large current surges. This enables accurate correction for various abnormal or sudden situations, demonstrating good robustness and adaptability. By extracting and fusing multiple health factors such as temperature changes, battery internal resistance, and isobaric charging / isobaric discharging time, a multi-dimensional and multi-level health assessment system is established. This system comprehensively reflects the battery aging process and health status, improving the accuracy and reliability of health monitoring and effectively avoiding the risks of misjudgment and omission caused by single health factors. It supports full-cycle, online dynamic tracking of battery capacity and health status. Through regular standard operating condition calibration and sliding window trend analysis, it achieves continuous correction of capacity parameters and dynamic optimization of health assessment, providing efficient and intelligent maintenance and management decision-making basis for energy storage systems. Attached Figure Description

[0019] Figure 1 This is the battery capacity degradation curve; Figure 2 The curve showing the change in charging time with the number of charge / discharge cycles; Figure 3 The curve showing the change of battery charging and discharging voltage over time; Figure 4 The curves showing the changes in battery constant voltage rise charging time, battery internal resistance, and constant voltage drop discharge time with the increase of charge and discharge cycles; Figure 5 The curve shows the change in battery temperature rise time with the increase of charge / discharge cycles. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0022] As disclosed in the background section, traditional methods often rely on idealized experimental conditions for capacity calibration, making it difficult to accurately assess the complex and dynamic working environment of actual power stations, resulting in decreased accuracy and applicability of capacity estimation. Some methods fail to fully consider the random disturbance effects of operating conditions such as charging and discharging current and temperature during data processing, making them susceptible to abnormal data and sudden operating conditions, leading to significant fluctuations in capacity and health assessment results. The selection and weighting of health factors generally adopt fixed or empirical settings, lacking adaptive optimization capabilities based on actual operating data, and failing to dynamically reflect the changes in the importance of each health factor at different life stages and operating conditions, resulting in a lag between comprehensive health characterization and the actual state of the battery. Existing methods still have shortcomings in the fusion and utilization of multi-source heterogeneous data, making it difficult to fully mine the correlation information between data, thus limiting the accuracy and comprehensiveness of health factor extraction.

[0023] To address the aforementioned problems, this invention proposes a method for calibrating the battery capacity and extracting health factors in low-voltage radio stations, comprising the following steps: S1. Real-time acquisition of battery terminal voltage V(t), current I(t), temperature T(t), and SOC data in the energy storage system of the distribution area; S2. The collected data is automatically divided into several complete charge-discharge cycles according to the preset cycle division rules; the equivalent capacity is calculated using the ampere-hour integration method for each complete charge-discharge cycle. ; S3, to The dynamic calibration capacity is obtained by performing adaptive weight correction based on operating conditions and using sliding window weighted fitting. ; S4. Extract multidimensional health factors from voltage, current, temperature and SOC data, and form a comprehensive health characterization through weighted fusion.

[0024] This invention addresses the complex and variable operating environment of batteries in substation energy storage systems. By collecting multi-source data such as voltage, current, SOC, and temperature across all time periods, and employing an adaptive window capacity fusion algorithm and operating condition weighting correction, it achieves high-precision, real-time battery capacity calibration. A sliding window weighted regression is used to analyze capacity changes and periodically correct capacity estimation drift using standard operating conditions. Simultaneously, health factors such as temperature changes, internal resistance, and constant-voltage ramp-up charging time are extracted to comprehensively assess battery health from multiple dimensions. This invention significantly improves the accuracy and robustness of capacity calibration and health assessment, supports intelligent operation and maintenance and lifespan prediction of energy storage systems, and is suitable for the widespread application of low-voltage substation energy storage.

[0025] The above plan will be explained in detail below.

[0026] A method for calibrating the battery capacity and extracting health factors in low-voltage radio stations includes the following steps: S1. Real-time acquisition of battery terminal voltage V(t), current I(t), temperature T(t), and SOC data in the energy storage system of the distribution area; In this embodiment, multi-source data of the battery in the energy storage system throughout the year at different times are collected, including battery terminal voltage V(t), current I(t), SOC change, and temperature T(t). Other optional parameters, such as operating efficiency, can be expanded according to the specific system configuration.

[0027] S2. The collected data is automatically divided into several complete charge-discharge cycles according to the preset cycle division rules; the equivalent capacity is calculated using the ampere-hour integration method for each complete charge-discharge cycle. ; Preferably, in this embodiment, the preset cyclic partitioning rule in S2 is: A loop is considered to have ended and a new loop begins when any of the following conditions are met: (a) Current is continuous and in the same direction and the SOC changes monotonically. ; (b) Absolute value of current And the duration exceeded 30 minutes, of which .

[0028] Preferably, in this embodiment, the ampere-hour integration method in S2 is: ; in, This is the start time; The termination time; Let be the instantaneous current within the cycle; when the actual sampling is discrete data, the above formula can be approximated as: ; in, For the first Number of sampling points in the next loop This represents the sampling time interval.

[0029] The current changes during battery charging and discharging are affected by control strategies and load disturbances. A sampling frequency of at least 1-10Hz is required to capture the main characteristics of current changes. For fast dynamic response scenarios, the sampling frequency is increased to 50Hz. The maximum permissible sampling interval Δtmax should not exceed 1 second; otherwise, it will lead to a decrease in the accuracy of capacity calculation and miss the detection of instantaneous large current surges.

[0030] At this point, the acquired current signal contains high-frequency jitter, noise, and occasional anomalies. To reduce the impact of noise on the accuracy of capacity estimation, jitter-resistant and noise-reducing preprocessing should be performed before calculation.

[0031] Preferably, in this embodiment, before performing the operating condition weight correction in S3, the current data is preprocessed by median filtering and moving average.

[0032] Specifically, in this embodiment, the collected raw current data needs to be checked for data integrity. The check includes identifying missing data points, abnormal timestamps, or duplicate sampling issues. For missing data, linear interpolation can be used to complete or correct the data based on the number and distribution of missing points, ensuring the continuity and accuracy of subsequent processing.

[0033] Secondly, outliers are identified and removed. A median filtering method is used, with a window of 3 or 5 points. The median value within this window is used to replace the center point, effectively eliminating spike noise and occasional abnormal impact points. Extreme values ​​exceeding the physically permissible range are directly pruned or removed using upper and lower thresholds.

[0034] Noise reduction and smoothing are performed using a moving average method. Setting an appropriate window length or cutoff frequency based on the sampling frequency effectively suppresses high-frequency noise and signal jitter, preserving the true trend of current changes.

[0035] S3, to The dynamic calibration capacity is obtained by performing adaptive weight correction based on operating conditions and using sliding window weighted fitting. ; Specifically, in this embodiment, to mitigate the impact of abnormal operating conditions (such as high current and extreme temperature) on the instantaneous fluctuations in capacity estimation, an operating condition weighting correction factor is introduced. This factor can be determined by real-time operating conditions (such as current temperature). Current amplitude This is obtained adaptively through linear or nonlinear functions, specifically set as follows: ; Preferably, in this embodiment, the adaptive weight correction for the working condition in S3 adopts an exponential parameterization form: ; in, This is the average temperature of the cycle. This is the root mean square value of the cyclic current; parameter The data is obtained by log-linearizing historical data and then fitting it using the least squares method or ridge regression.

[0036] Specifically, in this embodiment, the parameters By collecting historical experimental data under different temperatures and currents, the error between the estimated capacity and the actual capacity can be minimized, and the least squares method or ridge regression method can be used for fitting and solving the problem. Specifically, in the experimental dataset, based on the temperature, current, and actual capacity for each cycle, the aforementioned exponential relationship is constructed, and the optimal parameters are obtained through batch regression using a fitting algorithm. Ultimately, this correction factor can adaptively compensate for the systematic bias in capacity estimation caused by abnormal operating conditions such as extreme temperatures and high currents.

[0037] The fitting process is as follows: 1. Experimental data collection; Under different temperature and current conditions, a set of sample data was collected, and the temperature statistics for each cycle were recorded. Root mean square current and the actual measured capacity correction factor. (This is calculated by comparing the actual capacity under operating conditions with the standard capacity).

[0038] 2. Data conversion; To facilitate linear fitting, take the logarithm of both sides of the above equation: ; This transforms the problem into multiple linear regression.

[0039] 3. Parameter fitting; Construct a matrix form: make ,make ; A matrix is ​​formed from all samples. ; Use least squares or ridge regression to fit the parameters and solve... : ; After obtaining the parameters, the system can be adjusted according to real-time conditions under new operating conditions. and calculate This is used for capacity estimation correction. By collecting operating condition data, logarithmic linearization, and least squares / ridge regression modeling, the following can be obtained: The parameterized expression effectively compensates for the influence of temperature and current on capacity estimation.

[0040] Meanwhile, to reflect the capacity change trend of the battery at different stages and suppress observation noise, a sliding window weighted regression is used to fit the trend of the capacity of the current cycle and the previous M cycles: ; in, For the first in the window The weights for each cycle can be assigned different values ​​depending on the health stage.

[0041] Preferably, in this embodiment, the weight The exponential decreasing method is used to determine the value, highlighting the impact of recently cyclical data within the window on the current capacity change trend. Let the sliding window contain a total of... There are 1 loop, and the latest loop number is . In S3, the sliding window weighting uses exponentially decreasing weights: ; in, A positive attenuation coefficient controls the rate of decrease in weights. This means that data closer to the current cycle has a higher weight, reflecting its stronger representativeness of the current capacity state; while earlier cycles are assigned lower weights, effectively suppressing historical noise interference with trend fitting. The attenuation coefficient λ can be optimized by adjusting parameters based on actual capacity fluctuation characteristics, ultimately achieving smooth and sensitive tracking of battery capacity change trends.

[0042] Preferably, in this embodiment, when the dynamically calibrated capacity after sliding window weighted fitting is lower than the initial standard capacity for three consecutive times... When the battery reaches 70% capacity, it is considered to have reached the end of its lifespan.

[0043] Preferably, in this embodiment, a 1C constant current-constant voltage standard capacity calibration is performed every 30 to 100 cycles to correct long-term algorithm drift; in this embodiment, the 1C constant current-constant voltage standard capacity calibration is performed every 50 cycles, and the standard capacity calibration calculation formula is as follows: ; in, For standard constant current, and These are the start and end times of constant current charging and constant voltage charging, respectively. This refers to the instantaneous current during the constant voltage phase. For scenarios where the sampled data is discrete, the integral part can be approximated as a summation: ; in, This represents the current value at the kth sampling point during the constant voltage phase. The time interval between two adjacent sampling points; This represents the total number of sampling points during the constant pressure phase.

[0044] The capacity during the constant current stage is obtained by multiplying the constant current by the duration. During the constant voltage stage, the instantaneous current flowing through the battery is integrated over time, and the two are then added together to obtain the total charging capacity. This ensures the accuracy of the capacity calibration under standard operating conditions and avoids errors caused by algorithm drift correction.

[0045] In summary, this method enables dynamic and accurate calibration of the actual usable capacity of batteries under complex operating conditions in the energy storage system of the power distribution area, providing a solid foundation for subsequent health factor extraction and lifespan prediction.

[0046] To simulate the irregular charging and discharging conditions of batteries in real-world operation, the project team collected 12 months of operational data from multiple energy storage power stations in a certain region during 2023, including key parameters such as the number of charge-discharge cycles, changes in SOC (State of Charge), temperature changes, and efficiency degradation. A comprehensive 24-hour time-domain control scenario was set up. This series of batteries uses non-standard charge-discharge cycles to simulate the actual working environment, randomly selecting charge and discharge currents from 1.50A, 2.25A, 3.00A, 3.75A, and 4.50A. A charge-discharge cycle is defined as 5 minutes or reaching the cutoff voltage. The capacity threshold at which the battery cannot provide sufficient energy or power (70% of its capacity) during use is used as the end-of-life threshold. Standard charge-discharge cycles and constant-voltage cycling were used to calibrate the battery capacity during the experiment. The capacity degradation curves of RW9-12 batteries are as follows. Figure 1 As shown.

[0047] S4. Extract multidimensional health factors from voltage, current, temperature, and SOC data, and form a comprehensive health characterization through weighted fusion. The aim is to extract health factors with high correlation to battery capacity degradation from battery voltage, current, and temperature data as model inputs to improve the accuracy of the model in predicting battery RUL.

[0048] In this embodiment, the health factor extraction process is as follows: For the collected multi-source operating data (including voltage, current, SOC, temperature, capacity, etc.), a sliding window statistical feature extraction method is used to calculate the mean value within each window period. Standard deviation skewness kurtosis Maximum and minimum value difference Statistical characteristics, etc.

[0049] Based on charging time The health factors are extracted, and their average value and volatility can be expressed as follows: ; in, For the first Duration of the first charging process This represents the sample size. Changes in charging time during battery aging often serve as a good indicator of battery health.

[0050] Preferably, in this embodiment, the health factors extracted in S4 include at least one of the following: Normalized capacity: ; in, This is the initial standard capacity of the battery; Normalized internal resistance: ; in, The initial internal resistance, This represents the resistance within the current cycle. After a complete cycle and a 30-minute rest period, a 10-second 1C discharge pulse is applied. The voltage difference 10ms before and after the pulse is divided by the current to obtain the resistance within the current cycle in this embodiment. The hybrid pulse power characteristic test (HPPC) method was used to measure the internal resistance in ohms: after each complete charge-discharge cycle, the battery was allowed to rest for 30 minutes before a 10-second 1C discharge pulse was applied. The voltage difference within 10ms before and after the pulse, divided by the current, was the internal resistance in ohms. .

[0051] Isovoltage rise time during charging phase: ; in, This is the actual charging cutoff voltage; Discharge phase isothermal time: ; in, This refers to the battery's nominal voltage.

[0052] Specifically, in this embodiment, Figure 2The figure shows the curves of constant current and constant voltage charging time for four batteries as the number of charge-discharge cycles increases. As can be seen from Figure 2, the constant current charging time gradually decreases while the constant voltage charging time gradually increases with the increase in the number of charge-discharge cycles. During the battery charge-discharge cycle, the battery capacity decay process can be dynamically reflected based on changes in battery voltage. In this embodiment, taking the ternary lithium battery RW9 as an example, health factors are extracted from the voltage data. Its typical voltage plateau is around 3.7V, and the high SOC region (4.0-4.2V) is suitable for extracting aging characteristics. Figure 2 The figure shows the voltage variation curve of the RW9 battery during charge and discharge cycles.

[0053] Figure 3 Each curve in the diagram represents a complete charge-discharge process. Figure 3 In the diagram, 'a' represents the charging stage. Points A1 and A2 indicate when the battery reaches 4.1V (representing the characteristic voltage point of the high SOC range), and points B1 and B2 indicate when the battery reaches 4.2V (representing the charging cutoff voltage). The time difference between A1 and B1, and between A2 and B2, can be defined as the constant voltage rise charging time. This time period reflects the charging rate and efficiency of the battery as it approaches full charge. Figure 3 In the diagram, 'b' represents the voltage change during the discharge phase. At the beginning of discharge, the battery's internal resistance increases with the number of battery uses, mainly reflected in a larger instantaneous voltage drop. In the middle stage of the discharge curve, i.e., the stable discharge plateau period, the battery voltage drop is relatively stable. Points C1 and C2 can represent the moment the battery discharges to 4.0V (representing the characteristic voltage point of the high SOC range), and points D1 and D2 can represent the moment the battery discharges to 3.7V (representing the nominal voltage). The time difference between C1 and D1, and between C2 and D2, is the constant voltage drop discharge time, which can serve as an important indicator for evaluating battery discharge performance. The calculation formula is as follows: ; ; ; The formula above represents the time taken for a ternary lithium battery to drop from 4.2V (charging cutoff voltage) to 4.1V (representing the characteristic voltage point of the high SOC range) and from 4.0V (representing the characteristic voltage point of the high SOC range) to 3.7V (nominal voltage) during discharge. This definition reflects the capacity release characteristics of the battery in high or low voltage ranges. In practical terms, "constant voltage rise time" and "constant voltage drop time" generally refer to the duration of the constant voltage charging phase (i.e., the current gradually decreases while the voltage is constant) and the duration of the constant voltage discharge phase (if there is constant voltage discharge) or the discharge phase in a specific range, respectively. However, directly defining "constant voltage rise / drop time" as the time interval between two voltage points is often used as a simplified measurement method in research. Its essence is to characterize the battery behavior within a specific voltage range, which aligns with practical engineering applications.

[0054] Indicates the constant voltage rise charging time; Indicates the ohmic internal resistance of the battery; Indicates the endpoint voltage of the battery's sudden drop; Indicates the initial voltage drop of the battery; Indicates constant current discharge current; Indicates the equal voltage drop discharge time; t 4.2V t 4.1V t 4.0V and t 3.7V These represent the time it takes for a lithium-ion battery to first discharge to 4.2, 4.1, 4.0, and 3.7V, respectively.

[0055] Analyzing the constant-voltage-rise charging time and constant-voltage-drop discharging time provides a deeper understanding of battery performance at different charging and discharging stages. Furthermore, analyzing the voltage drop during discharge offers crucial information about changes in battery internal resistance, which is significant for predicting battery lifespan and performance degradation. Figure 4 The figures show the curves of the battery's constant voltage rise charging time, battery internal resistance, and constant voltage drop discharge time as the number of charge-discharge cycles increases.

[0056] The curves showing the changes in constant voltage rise charging time, battery internal resistance, and constant voltage drop discharge time with the number of charge-discharge cycles show that as the number of charge-discharge cycles increases, the constant voltage rise charging time gradually decreases, while the battery internal resistance and constant voltage drop discharge time gradually increase.

[0057] Based on temperature, health factors are extracted. During the charging and discharging process of a lithium-ion battery, the battery temperature rises due to the thermal effect. The formula for calculating the time it takes for the temperature to rise from the start of discharge to its peak value during lithium-ion battery discharge is: ; In the formula: This indicates the time required for the battery temperature to rise to its maximum value; Indicates the moment when the battery temperature reaches its peak. This indicates the moment the battery begins charging. The curve showing the change in battery temperature rise time with increasing charge / discharge cycles is as follows. Figure 5 As shown.

[0058] Preferably, in this embodiment, S4 further includes performing variational mode decomposition or ensemble empirical mode decomposition on the voltage or current curve, and calculating the energy of each intrinsic mode component. ; and the Pearson correlation coefficient between the energy of each intrinsic mode component and SOH. The absolute value is normalized and used as the weight to obtain the result, i.e. .

[0059] Specifically, in this embodiment, variational mode decomposition (VMD) or empirical mode decomposition (EMD) is applied to the charge / discharge curve. When decomposing the charge / discharge curve, variational mode decomposition (VMD) is used. During VMD decomposition, the initial value of the mode number K is generally 4–8, and can be adjusted appropriately according to the signal complexity and decomposition effect; the penalty parameter α is recommended to be set between 2000–5000, and can be corrected according to the actual sampling rate. To reduce endpoint effects, it is recommended to perform endpoint mirroring extension processing on the signal before decomposition. The energy of each intrinsic mode component (IMF) can be defined as... (i.e., signal energy). If significant mode mixing is encountered, ensemble empirical mode decomposition (EEMD) can be used as an alternative. During EEMD, the noise amplitude can be taken as 0.1–0.3 times the standard deviation of the original signal, and the number of repeated noise decompositions is generally set to 50–200. The IMF energy and dominant frequency characteristics obtained from the decomposition are used for subsequent health feature extraction and weighted fusion analysis.

[0060] Obtain several intrinsic mode components (IMFs), and then calculate the energy of each IMF. and clock speed The combination constitutes the degenerate feature vector of the signal. : ; and The values ​​are assigned to energy and frequency weights. =0.5, =0.5; clock speed This refers to the dominant frequency of each intrinsic mode component, which is used for subsequent feature extraction and weighted fusion analysis. The formula is as follows: ; K is the IMF component index; This represents the number of sampling points in the spectrum after performing a Fourier transform on the IMF. Let i be the frequency value of the i-th frequency point; For the k-th IMF at frequency The spectral amplitude at that location (i.e., the amplitude after performing an FFT on the IMF). The spectral energy at that frequency point; where, For the first The health weights of each IMF, where M is the total number of components. It can be learned from the correlation of historical health status.

[0061] Health weight This can be determined using a historical health status correlation learning method. In practice, it is first necessary to collect historical battery data over a period of time, including the energy of the k-th IMF component obtained from each cycle decomposition. And corresponding health status indicators (capacity retention rate, SOH) for the corresponding period. Then, the Pearson correlation coefficient is used to measure the linear correlation between IMF energy and health status, and its calculation formula is as follows: ; in, Indicates the number of historical samples. and These are the means of the corresponding indicators. To ensure that all weights are positive and sum to 1, the absolute values ​​of the correlation coefficients of each IMF are taken and normalized. The final health weight can be expressed as: ; Where M represents the total number of IMF components. Thus, the weights automatically learned from a large amount of historical data can effectively reflect the actual correlation between each IMF component and the battery's health status, making subsequent feature fusion more scientific and adaptable.

[0062] Finally, the above-mentioned multidimensional health factors are adaptively weighted and fused to form a comprehensive health characterization. : ; in, For each individual health factor, These are weighting coefficients, which can be dynamically optimized using data-driven methods.

[0063] Dynamic optimization of weighting coefficients is the key to achieving comprehensive health characterization H. Itotal The key to accuracy. Specifically, The determination process employs a data-driven approach, combining historical operational data with health factors. Correlation analysis was conducted among the factors to uncover their actual contribution to the characterization of battery health status.

[0064] First, a large amount of health factors under different operating conditions, along with corresponding battery lifespan, capacity calibration, and other labeled data, were collected. Principal component analysis (PCA) was used to initially screen out the feature factors with significant influence. Then, machine learning algorithms were used to build a predictive model between health factors and the target health state. During model training, an adaptive loss function was used to weight each health factor. Automatic adjustment and optimal allocation.

[0065] After training, the weight coefficients Dynamically assign values ​​to parameters based on specific operating conditions, lifespan stages, or environmental conditions to ensure... It exhibits higher accuracy and robustness throughout the entire battery lifespan.

[0066] Through the above multi-level and multi-algorithm collaborative extraction, the resulting health factor system can sensitively reflect the multi-dimensional degradation characteristics of battery performance, greatly improving the accuracy and robustness of subsequent lifespan modeling and health assessment.

[0067] Furthermore, this embodiment incorporates experimental simulations and data augmentation.

[0068] 1. Model Selection and Tool Determination Choosing the Thevenin equivalent circuit model (a second-order Thevenin circuit is optional) and the thermal model can reflect the voltage response, internal resistance change and temperature characteristics during charging and discharging.

[0069] Model using Matlab / Simulink, or implement the model using the open-source Python library PyBaMM.

[0070] Simulations using Thevenin equivalent circuit and thermal model under various operating conditions yield high-quality data covering diverse operational scenarios. By integrating and correcting this data with measured data, the accuracy and adaptability of the model parameters are effectively improved. These simulation experiments provide an essential data foundation and technical support for achieving the efficiency, accuracy, and scientific rigor of this method in the health management of distribution network-level energy storage batteries.

[0071] 2. Model parameter settings Based on the measured open-circuit voltage and internal resistance of the battery ( ),capacitance( ), thermal parameters ( Parameters are calibrated using methods such as )

[0072] The parameters can be obtained by fitting previously measured data (e.g., using gradient descent or genetic algorithm fitting). To simulate the battery aging process, the model parameters (e.g., internal resistance R0, R1) are not fixed but are gradually updated with the number of charge-discharge cycles. Specifically, the update rule is: every 50 cycles, the parameters are linearly adjusted based on measured data (e.g., capacity decay rate). For example, R... 0,new =R 0,initial ×(1+0.001×n), where n is the current cycle number and 0.001 is the empirical decay coefficient based on historical data (obtainable through least squares fitting). This ensures that the model can dynamically reflect aging characteristics such as increased battery internal resistance and capacity decay.

[0073] 3. Multi-condition input design A series of charge / discharge current conditions, ambient temperature, and initial state of charge (SOC) values ​​are set for simulation under different scenarios. Table 1 Examples of Thevenin equivalent circuit model parameters 3.2 Data Augmentation Detailed implementation process: 1. Typical operating condition settings Designed for various scenarios including overcharge (charging voltage up to 4.3V), undercharge (charging cutoff at 3.8V), high temperature (ambient temperature 55℃), fast charge / discharge (3C / 5C), low temperature (0℃), and short circuit.

[0074] 2. Batch Simulation Simulations were run in batches under various scenarios to collect data on changes in SOC, voltage, current, and temperature over time.

[0075] 3. Data output and organization The experimental dataset is composed of key outputs for each operating condition.

[0076] Table 2 Simulation conditions and main output data 3.3 Fusion of Simulation and Measured Data 1. Data alignment Alignment analysis was performed by selecting measured and simulated output data (such as voltage curves, temperature curves, etc.) under the same operating conditions.

[0077] 2. Error Analysis and Model Correction Calculate the error of key outputs (such as RMSE). If the error exceeds the limit, adjust the model parameters (such as R0, R1, C1, Rth, etc.) to make the output closer to the actual measurement.

[0078] 3. Data fusion output The corrected simulation data is mixed with the measured samples to form a high-quality dataset for training health factor extraction and remaining lifespan prediction.

[0079] Table 3 Comparison of Key Indicators between Simulation and Actual Measurement After completing the experimental simulation and data augmentation steps, the conclusion analysis mainly focuses on the accuracy of the simulation model, its multi-condition coverage, and the effectiveness of data fusion. By comparing the simulation data with the measured data under different typical conditions (such as standard, overcharge, undercharge, high temperature, and fast charge / discharge), it can be found that the constructed Thevenin equivalent circuit-thermal effect joint model (with parameters dynamically updated with the number of cycles) can well reflect the dynamic response characteristics of the battery under complex conditions. The voltage curve and temperature change trend output by the simulation are less than 3% of the measured results under most conditions, meeting the requirements of engineering applications. The simulation data of various faults and extreme scenarios greatly enriches the health factor extraction and model training sample library, improving the generalization ability of the subsequent RUL prediction model. This simulation and data augmentation process not only verifies the applicability of the parameterized model, but also effectively makes up for the limited number of actual condition samples, providing a high-quality and comprehensive data foundation for health factor extraction and battery life prediction. At the same time, the data fusion process, through adaptive correction of model parameters, further narrows the gap with real conditions, improving the robustness and accuracy of the battery health assessment system. Overall, simulation and data augmentation provide solid data and methodological support for subsequent health assessments and RUL predictions.

[0080] As can be seen from the above embodiments, the method of the present invention can efficiently and automatically transform massive amounts of raw operational data into interpretable parameter features, and achieve reliable evaluation of battery RUL with the assistance of deep learning, effectively supporting the scientific operation and maintenance and management of substation-level energy storage equipment.

[0081] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Substitutions may include replacements of some structures, devices, or method steps, or may be complete technical solutions. Equivalent substitutions or modifications made to the technical solutions and inventive concepts of the present invention should all be covered within the scope of protection of the present invention.

Claims

1. A method for calibrating the battery capacity and extracting health factors in a low-voltage radio station area, characterized in that, Includes the following steps: S1. Real-time acquisition of battery terminal voltage V(t), current I(t), temperature T(t), and SOC data in the energy storage system of the distribution area; S2. The collected data is automatically divided into several complete charge-discharge cycles according to the preset cycle division rules; The equivalent capacity is calculated using the ampere-hour integral method for each complete charge-discharge cycle. ; S3, regarding the above The dynamic calibration capacity is obtained by performing adaptive weight correction based on operating conditions and using sliding window weighted fitting. ; S4. Extract multidimensional health factors from the voltage, current, temperature and SOC data, and form a comprehensive health characterization through weighted fusion.

2. The method for calibrating the battery capacity and extracting health factors in a low-voltage radio station area as described in claim 1, characterized in that, The preset cyclic partitioning rule in S2 is as follows: A loop is considered to have ended and a new loop begins when any of the following conditions are met: (a) Current is continuous and in the same direction and the SOC changes monotonically. ; (b) Absolute value of current And the duration exceeded 30 minutes, of which .

3. The method for calibrating the battery capacity and extracting health factors for low-voltage radio stations as described in claim 1 or 2, characterized in that, The ampere-hour integration method in S2 is as follows: ; in, This is the start time; The termination time; Let be the instantaneous current within the cycle; when the actual sampling is discrete data, the above formula can be approximated as: ; in, For the first Number of sampling points in the next loop This represents the sampling time interval.

4. The method for calibrating the battery capacity and extracting health factors for low-voltage radio stations as described in claim 1, characterized in that, The adaptive weight correction for operating conditions in S3 adopts an exponential parameterization form: ; in, This is the average temperature of the cycle. This is the root mean square value of the cyclic current; parameter The data is obtained by log-linearizing historical data and then fitting it using the least squares method or ridge regression.

5. The method for calibrating the battery capacity and extracting health factors for low-voltage radio stations as described in claim 1 or 4, characterized in that, In S3, the sliding window weighting uses exponentially decreasing weights: ; in, A positive decay coefficient controls the rate of decrease of the weight.

6. The method for calibrating the battery capacity and extracting health factors for low-voltage radio stations as described in claim 1, characterized in that, The health factors extracted in S4 include at least one of the following: Normalized capacity: ; in, This is the initial standard capacity of the battery; Normalized internal resistance: ; in, The initial internal resistance, This represents the resistance within the current cycle. After the complete cycle is completed and the system has been allowed to stand for 30 minutes, a 10-second 1C discharge pulse is applied. The voltage difference 10ms before and after the pulse is divided by the current to obtain the result. Isovoltage rise time during charging phase: ; in, This is the actual charging cutoff voltage; Discharge phase isothermal time: ; in, This refers to the battery's nominal voltage.

7. The method for calibrating the battery capacity and extracting health factors for low-voltage radio stations as described in claim 1, characterized in that, S4 further includes performing variational mode decomposition or ensemble empirical mode decomposition on the voltage or current curves, and calculating the energy of each intrinsic mode component. ; and the Pearson correlation coefficient between the energy of each intrinsic mode component and SOH. The absolute value is normalized and used as the weight to obtain the result, i.e. .

8. The method for calibrating the battery capacity and extracting health factors for low-voltage radio stations as described in claim 1, characterized in that, It also includes performing a 1C constant current-constant voltage standard capacity calibration every 30 to 100 cycles to correct for long-term algorithm drift; the standard capacity calibration calculation formula is: ; in, For standard constant current, and These are the start and end times of constant current charging and constant voltage charging, respectively. This refers to the instantaneous current during the constant voltage phase. For scenarios where the sampled data is discrete, the integral part can be approximated as a summation: ; in, This represents the current value at the kth sampling point during the constant voltage phase. The time interval between two adjacent sampling points; This represents the total number of sampling points during the constant pressure phase.

9. The method for calibrating the battery capacity and extracting health factors for low-voltage radio stations as described in claim 1, characterized in that, Before performing the operating condition weight correction in S3, the current data is first preprocessed by median filtering and moving average.

10. The method for calibrating the battery capacity and extracting health factors for low-voltage radio stations as described in claim 1, characterized in that, When the dynamically calibrated capacity after sliding window weighted fitting is lower than the initial standard capacity for three consecutive times... When the battery reaches 70% capacity, it is considered to have reached the end of its lifespan.

Citation Information

Cited By

  • Lithium battery capacity estimation method and system using partial charging data

    CN121763122A

  • A lithium battery capacity estimation method and system using partial charging data

    CN121763122B