Energy storage battery aging analysis method based on adaptive transfer learning

Abstract

The invention relates to the technical field of energy storage batteries, in particular to an energy storage battery aging analysis method based on adaptive transfer learning, which comprises the following steps: extracting a voltage capacity peak position migration path structure, analyzing a frequency modulation instruction and a current response difference to obtain a frequency modulation working condition fluctuation degree, and adjusting mapping to calculate a cross-domain feature migration correlation degree. Evaluating the influence of trajectory offset, updating the correlation weight, comparing the consistency of capacity change, judging the aging evolution stage, and outputting a battery health state index. According to the method, a differential capacity curve peak position migration path structure is screened through a terminal voltage and capacity change relation, the fluctuation degree of an operation section is analyzed in combination with the output power of the converter, a feature migration confidence coefficient weight and a mapping correlation structure are dynamically adjusted according to a track offset relation, and key parameter fluctuation characteristics under complex working conditions are captured. The limitation that a static reference is difficult to adapt to a changeable environment is effectively overcome, and accurate positioning and state judgment of the battery in the health evolution stage are achieved under the continuously fluctuating operation background.

Description

Technical Field

[0001] This invention relates to the field of energy storage battery technology, and in particular to an energy storage battery aging analysis method based on adaptive transfer learning. Background Technology

[0002] The field of energy storage battery technology involves the mutual conversion and storage of electrochemical energy and electrical energy, including battery material research and development, battery cell manufacturing, battery pack structure design, and battery management systems. Power electronic converters and monitoring equipment are used to achieve smooth power output, peak shaving and valley filling, and grid frequency regulation. Traditional energy storage battery aging analysis methods address the health status evolution of energy storage batteries during charge-discharge cycles, such as capacity degradation and increased internal resistance. These methods typically rely on constant current charge-discharge testing, AC impedance spectroscopy analysis, or the establishment of equivalent circuit models as the basis for assessing the degree of battery aging.

[0003] Traditional methods rely on constant current charge-discharge tests or equivalent circuit models as evaluation criteria. However, such static parameter settings are difficult to respond to the frequent fluctuations in frequency modulation power commands of energy storage base stations in actual operation. Test data obtained solely from fixed modes cannot accurately reflect the deep coupling evolution trajectory of dynamic characteristics under varying operating conditions. Long-term use of fixed discrimination structures can lead to deviations in the extraction of health characteristics such as increased internal resistance from the actual situation. It is difficult to accurately track the deep aging evolution path under varying disturbances, which in turn severely restricts the reliability of overall energy storage and smooth output. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an aging analysis method for energy storage batteries based on adaptive transfer learning.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an energy storage battery aging analysis method based on adaptive transfer learning, comprising the following steps:

[0006] S1: Based on the cyclic recording of the battery pack of the energy storage base station, the terminal voltage sequence and capacity sequence during the charging stage are collected, the peak position migration path structure of the differential capacity curve is screened, the order of continuous cyclic peak position changes is compared, the peak position migration path is determined, and the cyclic decay characterization quantity is obtained.

[0007] S2: Based on the cyclic decay characterization quantity, locate the frequency regulation operation section, analyze the change process of the converter output power command, screen the switching position corresponding to the current response segment, compare the current change difference of each segment, and obtain the frequency regulation condition fluctuation.

[0008] S3: Based on the frequency modulation condition fluctuation, filter the time position of the capacity change stage, compare the distribution position of the operating segment features in the source domain feature space, adjust the target segment mapping position, and obtain the cross-domain feature migration correlation degree.

[0009] S4: Based on the cross-domain feature migration correlation, analyze the capacity change process corresponding to the current voltage change trajectory, select key positions of trajectory change to form a reference trajectory, compare the offset relationship between the current trajectory and the reference trajectory, adjust the migration correlation weight structure, and obtain the adaptive migration confidence.

[0010] S5: Based on the adaptive migration confidence, analyze the voltage change path corresponding to the capacity change trajectory during the cyclic advancement process, screen key node segments of capacity change, compare the consistency of capacity change trajectories, adjust the health status discrimination structure, and obtain battery health status indicators.

[0011] The present invention improves upon the following: the cyclic decay characterization parameters include peak voltage migration, peak capacity offset, and response platform displacement; the frequency modulation condition fluctuation includes power command switching frequency, current response change amplitude, and current fluctuation duration range; the cross-domain feature migration correlation includes target operating segment feature coordinates, source domain feature space reference center, and feature space offset description; the adaptive migration confidence includes trajectory offset influence coefficient, feature migration correlation weight, and migration relationship adjustment coefficient; and the battery health status indicators include capacity retention rate, health status level identifier, and remaining usable life estimate.

[0012] The present invention is improved in that the step of obtaining the cyclic decay characterization quantity is specifically as follows:

[0013] S111: Based on the cyclic recording of the battery pack of the energy storage base station, analyze the change process of the terminal voltage sequence during the charging stage, calculate the relationship of the change segment corresponding to the capacity sequence, screen the local turning points of the capacity change rate curve, compare the arrangement order of the voltage coordinates and capacity coordinates corresponding to each turning point, determine the correspondence of the continuous cycle positions, and obtain the peak position coordinate sequence.

[0014] S112: Based on the peak position coordinate sequence, analyze the distribution of peak positions in each cycle, compare the moving directions of voltage coordinates in adjacent cycles, filter the position sequences with the same direction in continuous cycles, calculate the connection relationship of the corresponding capacity coordinate change segments, adjust the time arrangement of peak positions in each cycle, and obtain the peak position migration path structure.

[0015] S113: Based on the peak position migration path structure, analyze the continuous cyclic peak position movement path structure, compare the continuity of the voltage coordinate change direction of each cycle, filter the path segments with the same movement direction, determine the correspondence of the capacity coordinate change process, and obtain the cyclic attenuation characterization quantity.

[0016] The present invention is improved in that the step of obtaining the frequency modulation fluctuation is specifically as follows:

[0017] S211: Based on the cyclic decay characterization quantity, compare the correspondence between the cyclic advancement position and the time series of the energy storage frequency regulation task, filter the time segment in which the power regulation command is continuously applied, determine the distribution of the operating segment corresponding to the power command holding state, and obtain the set of frequency regulation operating segments.

[0018] S212: Based on the set of frequency modulation operation segments, analyze the corresponding power command change process, compare the current change trajectory corresponding to the power command change position, filter the turning point time node of the current change direction, determine the path segment of the current trajectory that deviates from the stable change, and obtain the current response segment sequence.

[0019] S213: Based on the current response segment sequence, compare the differences in the current change paths of each operating segment, filter the current change trajectory offset section position, determine the segment disturbance trajectory, adjust the time arrangement structure of the operating segments, and obtain the frequency modulation condition fluctuation.

[0020] The present invention is improved in that the step of obtaining the cross-domain feature transfer correlation is specifically as follows:

[0021] S311: Based on the frequency modulation operating condition fluctuation, compare the time segments of the voltage change trajectory, filter the operating segments corresponding to the time positions of the capacity change stage, calculate the correspondence between voltage records and capacity records within the segment, and obtain the operating segment feature sequence.

[0022] S312: Based on the feature sequence of the running segment, analyze the voltage coordinates and capacity coordinates, compare the difference between the target running segment coordinates and the source domain feature space reference center coordinates, calculate the voltage coordinate offset and capacity coordinate offset, and obtain the feature space offset.

[0023] S313: Based on the feature space offset, compare the positional relationship of the running segments in the source domain feature space, calculate the mapping relationship of the target running segment coordinates, adjust the corresponding structure of the feature coordinates in the source domain feature space, and obtain the cross-domain feature transfer correlation degree.

[0024] The present invention is improved in that the step of obtaining the adaptive migration confidence is specifically as follows:

[0025] S411: Based on the cross-domain feature migration correlation, analyze the voltage change trajectory sequence, calculate the time difference result of the capacity change sequence, filter the change inflection points in the capacity change sequence, determine the arrangement relationship of the inflection points in the time axis, and obtain the capacity inflection trajectory node set.

[0026] S412: Based on the set of capacity transition trajectory nodes, compare the correspondence between the current trajectory node sequence and the reference trajectory node sequence, calculate the normalized amount of node voltage difference and capacity difference, determine the trajectory offset relationship, and obtain the trajectory offset coefficient.

[0027] S413: Based on the trajectory offset coefficient, analyze the corresponding structure between the cross-domain feature migration correlation sequence, calculate the migration correlation correction amount, compare the distribution of the migration correlation correction amount in the time series, adjust the migration correlation weight, and obtain the adaptive migration confidence.

[0028] The present invention is improved in that the steps for obtaining the battery health status indicators are specifically as follows:

[0029] S511: Based on the adaptive migration confidence, analyze the cyclic advancement position relationship, compare the corresponding structure of the capacity sequence change trajectory and the voltage sequence change path, screen the operating segment where the capacity change shows a turning point, determine the time sequence relationship of the capacity change path, and obtain the capacity trajectory node sequence.

[0030] S512: Based on the capacity trajectory node sequence, compare the morphological relationship of adjacent cyclic capacity trajectories, filter the operating segments where the capacity trajectory direction changes, determine the positional relationship of the capacity trajectory change stages, and obtain the aging stage position sequence.

[0031] S513: Based on the aging stage position sequence, analyze the stage distribution, compare the correspondence between the capacity change trajectory and the voltage change path, filter the trajectory deviation operation segment, determine the capacity trajectory stage change state, adjust the health status discrimination structure, and obtain the battery health status index.

[0032] The present invention is improved in that the terminal voltage sequence refers to the sequence data formed by the battery terminal voltage data recorded by the battery management system at a fixed sampling time during the charging phase in chronological order, and the frequency regulation operation segment refers to the operation time segment triggered by the power regulation command during the period when the energy storage power station performs the grid frequency regulation task.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0034] In this invention, the peak migration path structure of the differential capacity curve is screened by the relationship between terminal voltage and capacity change. Combined with the converter output power analysis, the fluctuation of the operating section is analyzed. The offset relationship between the target segment and the source domain distribution center is calculated to construct feature correlation to map the dynamic disturbance evolution law. Based on the trajectory offset relationship, the feature migration confidence weight and mapping correlation structure are dynamically adjusted to capture the key parameter fluctuation characteristics under complex operating conditions and adjust the discrimination benchmark in real time. This effectively overcomes the limitation of static benchmarks being difficult to adapt to the changing environment and achieves accurate positioning and state discrimination of battery health evolution stage under the background of continuous fluctuation. Attached Figure Description

[0035] Figure 1 This is a flowchart of the main steps of the present invention;

[0036] Figure 2This is a flowchart illustrating the process of obtaining the cyclic decay characterization parameters in this invention.

[0037] Figure 3 This is a flowchart of the process for obtaining the frequency modulation fluctuation in this invention;

[0038] Figure 4 This is a flowchart illustrating the process of obtaining cross-domain feature transfer correlation in this invention.

[0039] Figure 5 This is a flowchart illustrating the process of obtaining adaptive migration confidence in this invention.

[0040] Figure 6 This is a flowchart illustrating the process of obtaining battery health status indicators in this invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the 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, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0043] All user-related information involved in this invention (including but not limited to biometric information, identity verification information, behavioral data, device information, and other data that can be used for identity verification and personalized services) is collected and processed with the user's full knowledge and voluntary consent. The use of data is limited to purposes necessary for providing the technical services of this invention, and reasonable technical and management measures will be taken to ensure the security and confidentiality of users' personal information in terms of information protection and privacy.

[0044] For examples, please refer to Figure 1 This invention provides a technical solution for an energy storage battery aging analysis method based on adaptive transfer learning, comprising the following steps:

[0045] S1: Based on the cyclic recording of the battery pack of the energy storage base station, analyze the relationship between the changes in the terminal voltage sequence and the capacity sequence during the charging stage, screen the peak position migration path structure of the differential capacity curve, compare the order of continuous cyclic peak position coordinate changes, determine whether the peak position migration trend maintains a stable path, and obtain the cyclic decay characterization quantity.

[0046] S2: Based on the cyclic decay characterization quantity, locate the frequency regulation operation section, analyze the command change process of the output power of the converter frequency regulation control unit, screen the command switching position corresponding to the current response segment, compare the distribution differences of the current change process of each operation segment, determine the formation position of the segment disturbance trajectory, and adjust the arrangement relationship of the operation segments to obtain the frequency regulation condition fluctuation.

[0047] S3: Based on the frequency modulation operating condition fluctuation, obtain the operating segment corresponding to the voltage change trajectory, filter the time position of the capacity change stage, compare the distribution position of the operating segment features in the source domain feature space, calculate the offset relationship between the target operating segment features and the source domain feature distribution center, adjust the mapping position of the target operating segment features in the feature space, and obtain the cross-domain feature migration correlation degree.

[0048] S4: Based on the cross-domain feature transfer correlation, analyze the capacity change process of the current voltage change trajectory, screen key positions of trajectory change, form a reference trajectory, compare the offset relationship between the current trajectory and the reference trajectory, judge the degree of influence of trajectory offset on feature transfer results, adjust the weight structure of cross-domain feature transfer correlation, and obtain adaptive transfer confidence.

[0049] S5: Based on adaptive migration confidence, analyze the voltage change path corresponding to the capacity change trajectory during the cycle advancement, screen key node segments of capacity change, compare the consistency of capacity change trajectories, determine the position of the aging evolution stage, and adjust the health status discrimination structure to obtain battery health status indicators.

[0050] Cyclic degradation characteristics include peak voltage shift, peak capacity shift, and response plateau displacement; frequency modulation fluctuation includes power command switching frequency, current response change amplitude, and current fluctuation duration; cross-domain feature migration correlation includes target operating segment feature coordinates, source domain feature space reference center, and feature space offset description; adaptive migration confidence includes trajectory offset influence coefficient, feature migration correlation weight, and migration relationship adjustment coefficient; and battery health status indicators include capacity retention rate, health status level label, and remaining usable life estimate.

[0051] In S1, the terminal voltage sequence refers to the sequence data formed by the battery terminal voltage data recorded by the battery management system at a fixed sampling time during the charging phase, arranged in chronological order; the capacity sequence refers to the capacity change sequence formed by the cumulative charging capacity data recorded by the battery management system based on current integration, arranged in chronological order; the differential capacity curve refers to the curve formed by the relationship between the capacity change and the voltage change, used to describe the electrochemical reaction characteristics of different voltage ranges; the peak position coordinate change refers to the positional change of the voltage and capacity coordinates corresponding to the peak position in the differential capacity curve under different cycle counts; the peak position migration trend refers to the continuous movement direction of the peak position in the voltage coordinate direction as the number of cycles increases; and the stable path refers to the continuous migration trajectory formed by the peak position in similar change directions during continuous cycling.

[0052] In S2, the frequency regulation operation segment refers to the operating time segment triggered by the power regulation command during the energy storage power station's execution of the grid frequency regulation task; the frequency regulation control unit refers to the control module in the energy storage converter that receives the grid frequency regulation signal and generates the battery charging and discharging power control command; the command change process refers to the process of power control signal change formed by the output power command of the frequency regulation control unit changing over time; the current response segment refers to the continuous response time segment formed by the current change of the battery under the action of a certain power command; the operating segment refers to a continuous operating time segment in which the battery maintains the same operating state under the same power command control conditions; the distribution difference refers to the difference in statistical characteristics of the current change characteristics of different operating segments; the disturbance trajectory formation position refers to the time position at which the current change begins to deviate from the stable operating state; the arrangement relationship refers to the sequential connection structure formed by each operating segment according to the time sequence.

[0053] In S3, the voltage change trajectory refers to the continuous voltage curve formed by the change of battery terminal voltage over time; the source domain feature space refers to the feature expression region formed by the standard cycle test data of the energy storage battery. This region is composed of the peak voltage position, capacity plateau position, and polarization change features extracted during the cycle test phase, and is used to represent the distribution range of battery aging characteristics under experimental conditions; the target operating segment feature refers to the set of operating state features extracted from the voltage change trajectory and capacity change phase during the frequency modulation task performed by the energy storage base station. This set of features reflects the battery operating behavior under frequency modulation conditions; the offset relationship refers to the positional difference relationship between the corresponding position of the target operating segment feature in the source domain feature space and the center of the source domain feature distribution. This difference is used to describe the degree of feature change between the frequency modulation condition and the standard cycle condition; the mapping position refers to the feature coordinate position corresponding to the target operating segment feature in the source domain feature space. This position is used to represent the correspondence of the target operating condition feature in the source domain feature structure.

[0054] In S4, the key position refers to the time node corresponding to the change, inflection, abrupt change or peak position in the voltage change trajectory; the feature transfer result refers to the correlation state between the target operating segment features and the source domain feature structure formed after cross-domain feature transfer processing. This correlation state is used to represent the correspondence of frequency modulation operating condition features in the source domain feature structure; the weight structure refers to the importance ratio structure assigned to different feature transfer relationships during the cross-domain feature transfer process. This structure is used to describe the degree of influence of different transfer relationships in aging state judgment.

[0055] In S5, the cycle propagation process refers to the continuous operation process formed by the continuous increase of the number of battery charge and discharge cycles; the consistency degree refers to the similarity between the current capacity change trajectory and the historical capacity change trajectory in terms of the trend; the aging evolution stage position refers to the capacity decay stage position of the battery during the cycle operation; the discrimination structure refers to the battery health status judgment rule structure formed based on the capacity change trajectory, voltage change trajectory and trajectory deviation.

[0056] Please see Figure 2 The specific steps for obtaining the cyclic decay characterization parameters are as follows:

[0057] S111: Based on the cyclic recording of the battery pack of the energy storage base station, analyze the change process of the terminal voltage sequence during the charging stage, calculate the relationship of the change segment corresponding to the capacity sequence, screen the local turning points of the capacity change rate curve, compare the arrangement order of the voltage coordinates and capacity coordinates corresponding to each turning point, determine the correspondence of the continuous cycle positions, and obtain the peak position coordinate sequence.

[0058] Based on historical operation records of energy storage base station battery packs during continuous charging cycles, the system first extracts charging stage records for a specific battery cluster from the database for three consecutive cycles, such as the 15th, 16th, and 17th cycles. The terminal voltage and cumulative capacity data, sampled once per second, are then arranged chronologically. Subsequently, the voltage and capacity changes are read point by point. The difference between the capacity increment and voltage increment between adjacent points is used to calculate the capacity change rate. For example, in a certain record segment, if the voltage increases from 3.405V to 3.410V and the capacity increases from 12.40Ah to 12.44Ah, the corresponding change rate is 0.04Ah ÷ 0.005V = 8Ah / V. This same processing is applied to all sampling points to form a capacity change rate sequence. This sequence is then checked point by point. When the rate value at a certain point is more than 10% higher than the average of the two points before and after it, it is recorded as a local turning point. For example, when the rates of three adjacent points are 7.6, 8.9, and 7.5Ah / V, the intermediate value 8.9... The deviation of approximately 18% from the average value of 7.55 on both sides was recorded as a candidate inflection point. Subsequently, the voltage and capacitance values ​​corresponding to the candidate points were read and sorted from low to high voltage. The distances between adjacent inflection points were then compared. When the voltage difference was less than 0.01V and the capacitance difference was less than 0.05Ah, the two positions were merged. For example, the voltage difference between 3.438V and 3.444V was 0.006V, and the capacitance difference was 0.03Ah. Therefore, the average of the two values, 3.441V and 13.12Ah, was taken as the unified value. The positions are then numbered sequentially in each cycle and compared with the same-numbered positions in adjacent cycles. When the voltage difference between two cycles does not exceed 0.02V and the capacity difference does not exceed 0.10Ah, they are retained as the same peak position. For example, the 16th cycle is 3.441V and 13.12Ah, and the 17th cycle is 3.449V and 13.17Ah. The differences between the two are 0.008V and 0.05Ah, respectively, which meet the threshold requirements. The peak position coordinates are then arranged according to the cycle order to form a peak position coordinate sequence.

[0059] S112: Based on the peak position coordinate sequence, analyze the distribution of peak positions in each cycle, compare the voltage coordinate movement direction of adjacent cycles, filter the position sequences with the same direction in continuous cycles, calculate the connection relationship of the corresponding capacity coordinate change segment, adjust the peak position time arrangement order of each cycle, and obtain the peak position migration path structure.

[0060] The corresponding voltage coordinates in different cycles are extracted one by one according to the number position. For example, the voltage values ​​of a certain number position in the 18th to 22nd cycles are 3.442V, 3.447V, 3.451V, 3.453V and 3.454V respectively. Then the voltage difference between adjacent cycles is calculated. When the difference is greater than 0.002V, it is recorded as voltage upward shift, less than -0.002V is recorded as voltage downward shift, and falling between -0.002V and 0.002V is recorded as stable position. The judgment interval of 0.002V is set according to twice the range of the energy storage battery sampling accuracy of 0.001V. In the above example, the four adjacent differences are 0.005V, 0.004V, 0.002V and 0.001V respectively. The first two are upward shifts and the last two are stable. Therefore, this position does not form a continuous upward shift sequence. The system then checks for three or more consecutive records in the same direction. If this condition is met, the segment is marked as a sequence in the same direction. In this example, the 18th to 22nd loops do not constitute a continuous moving segment. The corresponding capacity value for that segment is then read, for example, [13.01Ah, 13.05Ah, 13.08Ah, 13.10Ah, 13.12Ah]. The capacity difference between adjacent segments is calculated one by one. If the difference is less than 0.06Ah, the connection is maintained; if the difference is greater than 0.06Ah, the connection is broken at that point. For example, if the capacity difference between two points is 0... If the value is 0.08Ah, then this point is taken as the starting point of the new segment. Then, the order of peak appearance is rechecked within the same segment. If the voltage value of the later number in a certain cycle is more than 0.005V lower than the previous number, then the two numbers are swapped. For example, if the voltage of P2 is 3.488V and the voltage of P3 is 3.481V in a certain cycle, the difference is -0.007V, so they are swapped and the corresponding capacity value is adjusted synchronously. The order correction of all cycles is completed in sequence, and each voltage sequence in the same direction is integrated with the corresponding capacity connection segment to form a complete path structure, thereby obtaining the peak migration path structure.

[0061] S113: Based on the peak position migration path structure, analyze the continuous cyclic peak position movement path structure, compare the continuity of the voltage coordinate change direction of each cycle, screen the path segments with the same movement direction, determine the correspondence of the capacity coordinate change process, and obtain the cyclic attenuation characterization quantity.

[0062] For each path, voltage change records are read in cyclical order. For example, the voltage values ​​for a certain path in cycles 25 to 31 are 3.452V, 3.456V, 3.460V, 3.463V, 3.467V, 3.470V, and 3.474V respectively. Then, the voltage difference between adjacent cycles is calculated and the direction of movement is recorded. When four consecutive differences are positive, the segment is considered a stable path. For example, the above sequence forms six positive differences. The system records the activity, then reads the corresponding capacity values ​​[13.18Ah, 13.22Ah, 13.25Ah, 13.28Ah, 13.31Ah, 13.34Ah, 13.36Ah], and calculates the capacity difference between adjacent points. When the difference exceeds 0.03Ah, the point is marked as abnormal and removed from the path. For example, if a point has a capacity of 13.40Ah, resulting in a difference of 0.06Ah with the previous point, then the point is removed, and the preceding and following points are reconnected. The voltage difference between the beginning and end of the path is read as the voltage migration amount. For example, 3.474V minus 3.452V equals 0.022V. Then, the capacity difference between the beginning and end is read to obtain the capacity offset amount. For example, 13.36Ah minus 13.18Ah equals 0.18Ah. Next, the voltage plateau interval in the corresponding cycle of the path is read. For example, the plateau in the 25th cycle is 3.39V to 3.47V, and the plateau in the 31st cycle is 3.36V to 3.45V. The lower limit of the plateau drops by 0.03V, and the upper limit drops by 0.02V. Therefore, the plateau shifts downward as a whole, but the displacement of the upper and lower boundaries is not completely consistent. Then, the number of times the path is consistent in direction in all cycles is counted. When the number of consistent times accounts for more than 70% of the total path length, the path is retained. For example, 5 out of 6 movements are in the same direction, accounting for about 71%, which meets the threshold requirement. Finally, the voltage migration amount, capacity offset amount, and plateau displacement amount corresponding to the path are registered according to the cycle interval, thereby obtaining the cycle attenuation characterization amount.

[0063] Please see Figure 3 The specific steps for obtaining the frequency modulation fluctuation are as follows:

[0064] S211: Based on the cyclic decay characterization quantity, compare the correspondence between the cyclic advancement position and the time series of the energy storage frequency regulation task, screen the time segment where the power regulation command is continuously applied, determine the distribution of the operating segment corresponding to the power command holding state, and obtain the set of frequency regulation operating segments.

[0065] Extract the time series data of frequency regulation tasks from the energy storage power station dispatch system log. Read the start time, end time, and corresponding power value of each frequency regulation command. Then, map the battery cycle progression records to the same time axis. For example, cycles 320 to 326 correspond to 08:10:00 to 09:05:00. Align this time period with each entry in the frequency regulation log. If a power command starts after 08:10:00 and ends before 09:05:00, it is recorded as the corresponding command. Then, read the power setpoint sequence second by second and calculate the power difference between adjacent sampling points. When the difference does not exceed 2kW, it is recorded as a command hold state. When the difference exceeds 2kW, it is recorded as a command change point. 2kW is based on 1M. The minimum adjustment step size of the W frequency modulation unit is set. For example, power sequences of 200kW, 201kW, 200kW, and 199kW are grouped into the same continuous segment. When the value jumps from 199kW to 230kW, the segment is divided at that moment. Then, the duration of each segment is calculated. Segments with a duration of less than 10s are directly deleted. The 10s value is determined based on the PCS output stabilization time. For example, if a segment lasts for 7s, it is removed. Then, the number of sampling points in the segment is counted. When the number is not less than 20, it is recorded as a valid operating segment. Finally, adjacent segments with a time interval of less than 5s and a power difference of no more than 3kW are merged, and the start and end times and corresponding cycle numbers are recorded in chronological order to obtain the set of frequency modulation operating segments.

[0066] S212: Based on the set of frequency modulation operation segments, analyze the corresponding power command change process, compare the current change trajectory corresponding to the power command change position, filter the time nodes of the current change direction inflection, determine the path segments of the current trajectory that deviate from the stable change, and obtain the current response segment sequence.

[0067] The power command sequence and battery current sampling sequence within the specified segment are read segment by segment, and the two types of data are unified to the same time scale. For example, when power is recorded at 1 second and current at 0.5 seconds, the average of two adjacent current points is taken to obtain a 1-second record. Then, the power value change is checked point by point. When the power difference between two adjacent points reaches 5kW or more, it is recorded as a power change node. The 5kW is set according to the station's adjustment accuracy and noise range. For example, a change from 180kW to 192kW is recorded as a change node. Subsequently, 8 seconds of current data are extracted before and after this node to form a local trajectory sequence. Then, the current change difference is calculated point by point. When the current difference changes from positive to negative or from negative to positive and the absolute value before and after the change reaches 3A, it is recorded as a current change. For turning points, such as the current sequences 46A, 50A, 55A, 51A, and 47A, corresponding to differences of 4A, 5A, -4A, and -4A, respectively, the position at 55A is recorded as a turning point. Then, the average current value 4 seconds before the turning point is taken as a stable reference value. If the deviation of the next 3 consecutive sampling points from this reference value exceeds 6A, then this time period is registered as a deviation from the stable change segment. For example, if the reference value is 44A, and the subsequent sampling points are 52A, 54A, and 53A, then all exceed the 6A threshold, and this segment is retained. Then, all deviation segments are connected in chronological order. When the interval between two segments is less than 4 seconds, they are merged into the same segment; otherwise, they are recorded independently, resulting in a current response segment sequence arranged by time.

[0068] S213: Based on the current response segment sequence, compare the differences in the current change path of each operating segment, filter the current change trajectory offset section position, determine the segment disturbance trajectory, adjust the time arrangement structure of the operating segments, and obtain the frequency modulation condition fluctuation.

[0069] The start and end times, duration in seconds, and corresponding current change sequences of each segment are read segment by segment. Multiple segments within the same frequency modulation (FM) band are then placed in the same list for item-by-item comparison. First, the duration of the segments is compared; when the time difference between two segments exceeds 5 seconds, it is recorded as a segment with a duration difference. Next, the starting current and peak current of each segment are read, and the change amplitude is calculated. For example, segment A increases from 40A to 58A with an amplitude of 18A, and segment B increases from 42A to 53A with an amplitude of 11A. The difference of 7A is not recorded. When the difference exceeds 8A, it is marked as an amplitude difference segment. Then, the trajectory offset position is checked, and the current at each sampling point in the segment is compared with the median current at the corresponding time for all segments in the same band. When the deviation at four consecutive sampling points exceeds 5A, it is recorded as a segment with an amplitude difference. The trajectory deviation segment, such as the deviation sequence 6A, 7A, 8A, 6A, is recorded as the deviation position. If the deviation is less than 4 consecutive points, it is not recorded. Then, the direction of current change before and after the deviation segment is checked. When there is a change sequence of rising-falling-rising or falling-rising-falling, it is registered as a disturbance trajectory. Then, the order of the running segments is rearranged according to the number of disturbances. The segments without disturbances are placed first, the segments with single disturbances are placed in the middle, and the segments with multiple disturbances are placed last. The number of power switching, the average peak-valley difference of current, and the percentage of deviation duration within the segment are counted. For example, if a segment has 12 switching within 15 minutes, an average peak-valley difference of 14A, and a deviation duration of 120s, accounting for 13.33%, the frequency modulation condition fluctuation is obtained by classifying the level according to the preset interval.

[0070] Please see Figure 4 The specific steps for obtaining cross-domain feature transfer correlation are as follows:

[0071] S311: Based on the fluctuation of frequency modulation operation, compare the time segments of voltage change trajectory, filter the operating segments corresponding to the time position of capacity change stage, calculate the correspondence between voltage records and capacity records within the segment, and obtain the characteristic sequence of operating segments.

[0072] The terminal voltage trajectory, cumulative capacity trajectory, and operation log of the energy storage base station within the same time period are unified onto the same second-level time axis. Then, the start and end times of the voltage curve are read segment by segment and compared with the start and end times of the capacity change phase. When the overlap length of two time segments reaches 20 seconds or more, it is recorded as a valid corresponding segment. For example, if a voltage record is located from 10:15:00 to 10:16:10 and the capacity phase is located from 10:15:20 to 10:16:00, and the overlap is 40 seconds, it is retained; if the overlap is less than 20 seconds, it is discarded. Subsequently, the voltage and capacity values ​​of each sampling point in the retained segment are read in pairs according to the time sequence, and the voltage values ​​of adjacent sampling points are calculated. For voltage and capacity differences, when the voltage changes in the same direction for 5 consecutive sampling points and the capacity increment fluctuation does not exceed 0.03Ah, it is recorded as a stable corresponding segment. For example, the voltage gradually increases from 3.412V to 3.438V, and the capacity increases from 12.44Ah to 12.58Ah. If the continuous change meets the condition, it is retained. Then, the starting voltage, ending voltage, voltage span, starting capacity, ending capacity, capacity span, and duration within the segment are organized into a group of segment records and numbered according to time sequence. When the interval between two adjacent groups does not exceed 3s, they are merged into the same running segment. If it exceeds 3s, they are registered separately to form a characteristic sequence of running segments arranged by segment.

[0073] S312: Based on the feature sequence of the running segment, analyze the voltage coordinates and capacity coordinates, compare the differences between the target running segment coordinates and the reference center coordinates of the source domain feature space, and calculate the voltage coordinate offset and capacity coordinate offset using the following formula:

[0074] ;

[0075] The feature space offset is obtained, where, Feature space offset refers to the degree of positional shift of the features of a target runtime segment in the source domain feature space. The voltage coordinates of the target running segment represent the voltage position of the target running segment in the feature space, obtained from the voltage change trajectory. The source domain reference center voltage coordinates represent the position of the source domain characteristic space reference center in the voltage dimension. The target runtime segment capacity coordinates represent the capacity position of the target runtime segment in the feature space obtained from the capacity change phase. The source domain reference center capacity coordinates represent the position of the source domain feature space reference center in the capacity dimension. This refers to a voltage reference quantity, used to normalize voltage coordinate offsets, enabling voltage dimension offsets to participate in unified spatial calculations. It refers to the capacity reference quantity, which is used to normalize the capacity coordinate offset so that the offset of the capacity dimension can participate in the unified spatial calculation.

[0076] Feature space offset represents the feature points of the target running segment in the feature space consisting of voltage coordinates and capacity coordinates. and Relative to the source domain reference center and The normalized spatial distance, i.e. the degree of spatial deviation between the battery operating characteristics extracted under frequency regulation conditions and the characteristic center under standard cycle conditions; the voltage-capacity characteristic distribution of energy storage batteries differs between standard cycle test conditions (source domain) and frequency regulation operation conditions (target domain), and the characteristic spatial offset is used to quantify the degree of this difference.

[0077] Based on the feature sequence of the running segment, the coordinates of the target running segment are read, and the voltage coordinates of the target running segment at a certain sampling time are obtained. Target runtime fragment capacity coordinates Then read the coordinates of the source domain feature space reference center and obtain the voltage coordinates of the source domain reference center. Source domain reference center capacity coordinates At the same time, set the voltage reference value. Capacity reference quantity Subsequently, a proportional normalization method was used to process the voltage difference and capacity difference, where the original voltage difference... The corresponding normalized result Original capacity difference The corresponding normalized result This establishes a one-to-one correspondence between the original parameters and the normalized parameters. Based on this, the normalized results are substituted into the formula for calculation. First, the square of the voltage normalization term is calculated to obtain... Then calculate the square of the capacity normalization term to obtain Then, the squares of the two terms are summed to obtain the result. Then take the square root of the sum to get The positional relationship of the running segments in the source domain feature space is determined by the interval boundaries. The intervals are set as follows: when When the target operating segment's voltage and capacity coordinates are close to the source domain reference center, the operating segment is located near the center of the source domain feature space; when When the target runtime segment coordinates are offset to a certain extent relative to the source domain reference center, the runtime segment is still located in the neighborhood of the source domain feature space center; when When the target runtime segment coordinates are outside the reference center of the source domain, the runtime segment is located in the distribution area outside the feature space of the source domain; when When the interval is "far from the reference center," it indicates that there is a significant spatial distance between the target runtime segment's coordinates and the source domain reference center, meaning the runtime segment has deviated from the distribution range of the source domain center. The calculation results are then obtained. By comparing this result with the interval boundary, we can obtain... Therefore, the result falls into the "nearby offset interval", which indicates that the normalized spatial distance formed by the voltage and capacity dimensions of the running segment is in the vicinity of the source domain reference center, and the running segment maintains a continuous distribution relationship with the reference center in the source domain feature space.

[0078] S313: Based on the feature space offset, compare the positional relationship of the running segments in the source domain feature space, calculate the mapping relationship of the target running segment coordinates, adjust the corresponding structure of the feature coordinates in the source domain feature space, and obtain the cross-domain feature transfer correlation degree.

[0079] The voltage span, capacity span, duration, and fluctuation level of each running segment are placed in the same coordinate table. Each item is then compared with a reference position in the source domain feature space, and the differences between the two in each dimension are read. For example, a target segment has a voltage span of 0.026V, a capacity span of 0.14Ah, and a duration of 48s, while the source domain reference position corresponds to 0.020V, 0.11Ah, and 45s. The three differences are 0.006V, 0.03Ah, and 3s, respectively. Then, the distance between this segment and adjacent reference positions is checked. If the total difference with the current reference position is less than the total difference with other reference positions, the mapping is temporarily paused. If, at a given location, the difference in a certain dimension exceeds a preset threshold (e.g., 0.01V for voltage, 0.05Ah for capacity, and 8s for time), a new comparison is made with the adjacent reference location. For example, if the capacity difference exceeds the threshold by 0.07Ah, the comparison is changed to the next reference location. The determined mapping coordinates are then corrected according to the offset size. Segments with small offsets retain their original mappings, segments with offsets in the middle range are moved one level to the nearest reference location, and segments with offsets exceeding two consecutive thresholds are registered separately. Finally, the segment number, mapping coordinates, corrected corresponding location, and offset level are sequentially written into the association table to obtain the cross-domain feature transfer association degree.

[0080] Please see Figure 5 The specific steps for obtaining adaptive migration confidence are as follows:

[0081] S411: Based on cross-domain feature migration correlation, analyze the voltage change trajectory sequence, calculate the time difference result of the capacity change sequence, filter the change inflection points in the capacity change sequence, determine the arrangement relationship of the inflection points in the time axis, and obtain the capacity inflection trajectory node set.

[0082] The system reads terminal voltage records and capacity accumulation records for continuous operation periods from the energy storage base station operation database, and unifies these two types of records to the same time scale. For example, voltage and capacity are arranged in a time series at 1-second intervals. Then, it reads the difference results of adjacent sampling points in the capacity series point by point. For example, when the capacity changes from 12.44Ah, 12.46Ah, 12.49Ah, and 12.50Ah in a certain period, the difference results are 0.02Ah, 0.03Ah, and 0.01Ah, respectively. Subsequently, the direction of the continuous difference values ​​is judged. When the difference value changes from a positive value to a smaller positive value and then rises again, or changes from a larger positive value to a smaller value and then rises again, it is recorded as a turning point. The judgment interval is set to record a valid turning point when the change amplitude of adjacent difference values ​​exceeds 0.015Ah. For example, the difference sequence [0.02Ah, 0.03Ah, 0.01Ah, ... If the difference between the third position and the preceding and following values ​​in [0.04Ah] exceeds 0.02Ah, it is recorded as a turning point. Then, the timestamps corresponding to the turning points are read one by one and arranged in chronological order to form a node list. For example, the three nodes 10:05:12, 10:06:30, and 10:07:45 are recorded in sequence. Then, the time interval between adjacent nodes is checked. If the interval is less than 8s, they are merged into the same turning segment. If the interval is greater than 8s, they are kept as independent records. For example, two nodes with an interval of 5s are merged, and those with an interval of 12s are separated. Then, the sorted node records are aligned with the corresponding voltage trajectory segments one by one. The voltage values ​​corresponding to the time of the node are read, such as 3.428V, 3.435V, and 3.441V. The node time, voltage value, and capacity difference result are registered in sequence as a set of trajectory node records to obtain the capacity turning trajectory node set.

[0083] S412: Based on the capacity transition trajectory node set, compare the correspondence between the current trajectory node sequence and the reference trajectory node sequence, calculate the normalized expression of the node voltage difference and capacity difference, determine the trajectory offset relationship, and use the following formula:

[0084] ;

[0085] The trajectory offset coefficient is obtained, where, The trajectory offset coefficient indicates the overall deviation of the current voltage / capacity change trajectory relative to a reference trajectory. This quantity is a dimensionless index. The total number of trajectory nodes. The current voltage change trajectory is in the [number]th ... Terminal voltage data corresponding to each node location. The reference trajectory is in the 1st Terminal voltage data corresponding to each node location. Refers to the voltage reference quantity. The current capacity change trajectory is in the [number]th [year]. Capacity data corresponding to each node location The reference capacity trajectory is in the 1st Capacity data corresponding to each node location Refers to the baseline quantity of capacity;

[0086] The trajectory offset coefficient is a comprehensive quantitative indicator used to characterize the overall offset between the current operating trajectory and the reference trajectory. This coefficient is synthesized by normalizing the differences between the voltage change trajectory and the capacity change trajectory at the corresponding node positions. It is used to describe the degree of positional difference between the two operating trajectories in the state space. This coefficient is used to quantify the offset state of the battery operating trajectory relative to the reference trajectory, thereby providing a quantitative basis for subsequent migration association weight adjustment and adaptive migration confidence calculation.

[0087] Obtain the voltage sequence from the node set. With capacity sequence Simultaneously, the reference trajectory node sequence is read from the preset sample dataset. and The reference trajectory is derived from the voltage-capacity trajectory node data recorded during the standard cycle test phase of the energy storage battery. Then, the current trajectory is compared node-by-node with the reference trajectory according to the node number, forming a set of voltage difference data and capacity difference data at each node. The differences are then normalized using a reference quantity proportional normalization method, i.e., based on the voltage reference quantity. The voltage difference is scaled proportionally to a capacity reference. Scaling the capacity difference proportionally, voltage reference quantity The voltage range was obtained statistically from the rated operating voltage range of the energy storage battery. The example operating range is 2.9V to 3.65V, with an average voltage of 3.3V. Therefore, the following settings were adopted: V, Capacity reference quantity Based on statistics of the capacity variation range during the frequency modulation operation cycle, the example capacity variation range is 100Ah to 105Ah, with a variation span of 5Ah. Therefore, the following settings are provided: Ah, set the number of nodes in the node sequence. The node index is The original values ​​of the current trajectory voltage are respectively , , , , , The original values ​​of the reference trajectory voltage are respectively , , , , , The original values ​​of the corresponding node capacities are respectively , , , , , The original reference capacity values ​​are respectively , , , , , Based on the above original values, a sequence of absolute voltage differences is first formed. , , , , , Subsequently, the normalized voltage difference sequence was obtained by proportional normalization: 0.00606, 0.00303, 0.00606, 0.00606, 0.00606, 0.00606. The original capacity difference sequence was... , , , , , After proportional normalization, the normalized difference sequence of capacity is obtained: 0.04, 0.04, 0.06, 0.06, 0.08, 0.10. The normalized results are then substituted into the formula for calculation. The first part is the average normalized voltage difference:

[0088] ;

[0089] Normalized sum of square roots of differences:

[0090] ;

[0091] The final calculation yielded Divide the following preset intervals: when When the corresponding trajectory segment is adjacent, it means that the overall offset between the current voltage / capacity change trajectory and the reference trajectory is in a convergent state, and the composite offset remains within a small range; when At that time, the corresponding trajectory separation segment indicates that a continuous offset has formed between the current voltage capacity change trajectory and the reference trajectory, and it is the composite offset entering the intermediate interval; when At that time, the corresponding trajectory extension segment signifies an expanded offset between the current voltage / capacity change trajectory and the reference trajectory; this is the composite offset entering the upper segment interval. Calculated... When comparing this result with the above interval, it can be written as This indicates that the result falls within the trajectory separation segment, which means that the current voltage capacity change trajectory has formed a continuous offset relative to the reference trajectory, and this offset is in the middle range.

[0092] S413: Based on the trajectory offset coefficient, analyze the corresponding structure between the cross-domain feature migration correlation sequence, calculate the migration correlation correction amount, compare the distribution of the migration correlation correction amount in the time series, adjust the migration correlation weight, and obtain the adaptive migration confidence.

[0093] Read the migration correlation value and trajectory offset coefficient corresponding to each running segment, and arrange them into a correspondence table in chronological order. For example, in a certain running segment, the migration correlation records are 0.62, 0.64, 0.66, and 0.63, and the corresponding trajectory offset coefficients are 0.08, 0.10, 0.12, and 0.09. Then, calculate the migration correlation correction for each item by subtracting the offset coefficient from the migration correlation value. For example, 0.62 minus 0.08 equals 0.54, 0.64 minus 0.10 equals 0.54, 0.66 minus 0.12 equals 0.54, and 0.63 minus 0.09 equals 0.54. Then, arrange all the correction results in chronological order and compare the differences between adjacent items one by one. When the difference is less than or equal to 0.02, it is recorded as a stable distribution state; when the difference is greater than 0.02, it is recorded as a fluctuating distribution state. For example, the correction result for a certain segment is 0.54, 0.09 ... When the values ​​are 55, 0.53, and 0.57, the difference between 0.55 and 0.53 (0.02) is still considered stable, while the difference between 0.53 and 0.57 (0.04) is considered fluctuating. The number of stable and fluctuating states is then counted, and the number of adjacent comparisons is used as the statistical base. When the number of stable states accounts for 70% or more of all comparisons, the weight value of that segment is set to 0.8. When the stable proportion is between 40% and 70%, it is set to 0.6. When it is below 40%, it is set to 0.4. For example, if a segment has 10 migration association records, there can be 9 adjacent comparisons, of which 7 are stable states. The stable proportion is 7 / 9 ≈ 77.8%, corresponding to a weight of 0.8. Then, the corrected migration association value is multiplied by the corresponding weight item by item to obtain a new migration association result. For example, 0.54 multiplied by 0.8 equals 0.432. This numerical sequence is recorded in chronological order to form an adaptive migration confidence score.

[0094] Please see Figure 6 The specific steps for obtaining battery health status indicators are as follows:

[0095] S511: Based on adaptive migration confidence, analyze the cyclic advancement position relationship, compare the corresponding structure of the capacity sequence change trajectory and the voltage sequence change path, screen the operating segment where the capacity change shows a turning point, determine the time sequence relationship of the capacity change path, and obtain the capacity trajectory node sequence.

[0096] The start and end times and corresponding confidence levels of consecutive cycles for the same battery cluster are listed in order of cycle number. The capacity and voltage change curves within each cycle are then read sequentially. Capacity and voltage values ​​at the same timestamp are paired and recorded. For example, in the 401st cycle, between 10:10:00 and 10:26:00, the capacity changes from 12.31 Ah to 12.86 Ah, and the voltage changes from 3.402 V to 3.468 V. This segment is then used as a base trajectory. The capacity increments of adjacent sampling points are then examined. When the continuous increment sequence shows a pattern of decreasing from a large value to a local small value and then rising again (e.g., 0.03 Ah, 0.02 Ah, 0.01 Ah, 0.03 Ah), the position corresponding to the local small value is recorded as a capacity inflection point. The inflection point is determined by the difference between adjacent increments reaching 0.01. A 5Ah threshold is used as the recording threshold. When the difference between the position and at least one of the adjacent incremental values ​​reaches or exceeds 0.015Ah, it is recorded as a turning point. If the difference is less than 0.015Ah, it is not recorded. Then, the time position and voltage position corresponding to all turning points are checked one by one. When the voltage change direction corresponding to a turning point is consistent with the direction of the adjacent voltage difference formed by the five sampling points before and after it, the point is retained. For example, if the voltage before and after a turning point continues to rise, the point is retained. If the voltage direction before and after the turning point changes in opposite directions, it is removed. Then, all retained points are arranged in chronological order. If the time interval between two adjacent points is less than or equal to 6s, they are merged into a node segment. If it is greater than 6s, it is registered as an independent node. Finally, the cycle number of the node, the node time, the corresponding capacity value, the corresponding voltage value, and the confidence of the cycle are written into the node table to obtain the capacity trajectory node sequence.

[0097] S512: Based on the capacity trajectory node sequence, compare the morphological relationship of adjacent cyclic capacity trajectories, filter the operating segments where the capacity trajectory direction changes, determine the positional relationship of the capacity trajectory change stages, and obtain the aging stage position sequence.

[0098] Expand the node tables of two adjacent cycles side by side, and compare the number of nodes, node intervals, and the direction of capacity change before and after a node item by item. For example, if 3 nodes are registered in the 401st cycle and 4 nodes are registered in the 402nd cycle, first record the difference in the number of nodes as 1, then check the time offset between corresponding nodes. If the time difference between nodes with the same sequence number does not exceed 12s, they are considered to be in continuous position. If the time difference is greater than 12s but not more than 15s, it is recorded as a position offset. If the time difference exceeds 15s, and the capacity difference also exceeds 0.05Ah, it is marked as a new stage. If the capacity difference does not exceed 0.05Ah, it is still recorded as a long-distance position offset. Then, read the capacity change trend of 5 sampling points before and after each node. If the first segment continuously increases and the second segment continuously decreases, the node is recorded as a direction change node. If both segments increase but the slope slows down significantly, it is not recorded as a direction reversal, but only as a stage transition point. For example, a certain node If the capacity increases by 0.08Ah in the first 5 seconds and only by 0.02Ah in the next 5 seconds, it is classified as a transition point. Then, the direction change nodes in adjacent cycles are aligned in chronological order. If the direction change occurs in two consecutive cycles at similar time positions with a time difference of no more than 12 seconds and the corresponding voltage range difference is no more than 0.01V, then this position is registered as the same change stage. Subsequently, all stage positions are numbered. The first cycle appearing in the first cycle is registered as the first stage, that is, the cycle number of the first occurrence of this stage is located in the first 30% range of the entire cycle range; the continuous offset in the middle is registered as the middle stage, that is, the stage node has a position offset in 3 or more consecutive cycles; the multiple overlapping turns in the latter stage is registered as the latter stage, that is, the overlapping of 3 or more turning nodes occurs in the latter 30% cycle range. The cycle range and node range corresponding to each stage are written into the sequence list to obtain the aging stage position sequence.

[0099] S513: Based on the aging stage location sequence, analyze the stage distribution, compare the correspondence between the capacity change trajectory and the voltage change path, screen the trajectory deviation operation segment, judge the capacity trajectory stage change status, adjust the health status discrimination structure, and obtain the battery health status index.

[0100] The number of cycles, nodes, and corresponding time lengths covered by each stage are statistically analyzed according to the stage number. For example, if a stage covers cycles 401 to 415, a total of 15 cycles, 38 nodes, and a cumulative duration of 210 minutes, then this stage is registered as an independent analysis unit. Subsequently, the capacity change trajectory and voltage change path within this stage are matched one-to-one with the timestamps. The voltage path at each capacity node is checked for synchronization offset. When the voltage deviation of four consecutive nodes exceeds 0.012V, this segment is recorded as a trajectory offset running segment. For example, if the voltage deviations of four adjacent nodes in a stage are 0.013V, 0.015V, 0.014V, and 0.016V, then this segment is recorded. If only two nodes exceed the threshold, it is not recorded. The capacity change status of each stage is then compared. If the capacity difference between nodes within a stage exceeds 60%, the record falls between 0.01Ah and 0.03Ah. If the node position moves smoothly backward, it is recorded as a gradual change phase. If the capacity difference exceeds 0.05Ah three or more times and is accompanied by concentrated node movement, it is recorded as an acceleration phase. If the number of nodes decreases and the capacity difference falls back to within 0.02Ah, it is recorded as a plateau phase. Subsequently, the health status discrimination structure is adjusted according to the phase classification results. Specifically, the judgment level corresponding to the gradual change phase is set as Level 1, the acceleration phase as Level 2, and the plateau phase as Level 3. The final registration is based on the cycle proportion of each phase. For example, when Level 1 accounts for 40%, Level 2 accounts for 35%, and Level 3 accounts for 25%, the corresponding weights of Level 1, Level 2, and Level 3 are 1.0, 0.8, and 0.6, respectively. The capacity retention rate is calculated by weighting the capacity change amplitude of each phase. When the comprehensive score falls within the preset medium level range, the output health status level is medium. Then, based on the capacity decay trend fitting results, the remaining usable life is estimated to be 420 cycles, thus obtaining the battery health status index.

[0101] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An aging analysis method for energy storage batteries based on adaptive transfer learning, characterized in that, Includes the following steps: S1: Based on the cyclic recording of the battery pack of the energy storage base station, the terminal voltage sequence and capacity sequence during the charging stage are collected, the peak position migration path structure of the differential capacity curve is screened, the order of continuous cyclic peak position changes is compared, the peak position migration path is determined, and the cyclic decay characterization quantity is obtained. S2: Based on the cyclic decay characterization quantity, locate the frequency regulation operation section, analyze the change process of the converter output power command, screen the switching position corresponding to the current response segment, compare the current change difference of each segment, and obtain the frequency regulation condition fluctuation. S3: Based on the frequency modulation condition fluctuation, filter the time position of the capacity change stage, compare the distribution position of the operating segment features in the source domain feature space, adjust the target segment mapping position, and obtain the cross-domain feature migration correlation degree. S4: Based on the cross-domain feature migration correlation, analyze the capacity change process corresponding to the current voltage change trajectory, select key positions of trajectory change to form a reference trajectory, compare the offset relationship between the current trajectory and the reference trajectory, adjust the migration correlation weight structure, and obtain the adaptive migration confidence. S5: Based on the adaptive migration confidence, analyze the voltage change path corresponding to the capacity change trajectory during the cyclic advancement process, screen key node segments of capacity change, compare the consistency of capacity change trajectories, adjust the health status discrimination structure, and obtain battery health status indicators.

2. The energy storage battery aging analysis method based on adaptive transfer learning according to claim 1, characterized in that, The cyclic decay characterization parameters include peak voltage migration, peak capacity shift, and response platform displacement. The frequency modulation condition fluctuation includes power command switching frequency, current response change amplitude, and current fluctuation duration range. The cross-domain feature migration correlation includes target operating segment feature coordinates, source domain feature space reference center, and feature space offset description. The adaptive migration confidence includes trajectory offset influence coefficient, feature migration correlation weight, and migration relationship adjustment coefficient. The battery health status indicators include capacity retention rate, health status level identifier, and remaining usable life estimate.

3. The energy storage battery aging analysis method based on adaptive transfer learning according to claim 1, characterized in that, The specific steps for obtaining the cyclic decay characterization quantity are as follows: S111: Based on the cyclic recording of the battery pack of the energy storage base station, analyze the change process of the terminal voltage sequence during the charging stage, calculate the relationship of the change segment corresponding to the capacity sequence, screen the local turning points of the capacity change rate curve, compare the arrangement order of the voltage coordinates and capacity coordinates corresponding to each turning point, determine the correspondence of the continuous cycle positions, and obtain the peak position coordinate sequence. S112: Based on the peak position coordinate sequence, analyze the distribution of peak positions in each cycle, compare the moving directions of voltage coordinates in adjacent cycles, filter the position sequences with the same direction in continuous cycles, calculate the connection relationship of the corresponding capacity coordinate change segments, adjust the time arrangement of peak positions in each cycle, and obtain the peak position migration path structure. S113: Based on the peak position migration path structure, analyze the continuous cyclic peak position movement path structure, compare the continuity of the voltage coordinate change direction of each cycle, filter the path segments with the same movement direction, determine the correspondence of the capacity coordinate change process, and obtain the cyclic attenuation characterization quantity.

4. The energy storage battery aging analysis method based on adaptive transfer learning according to claim 1, characterized in that, The specific steps for obtaining the frequency modulation fluctuation are as follows: S211: Based on the cyclic decay characterization quantity, compare the correspondence between the cyclic advancement position and the time series of the energy storage frequency regulation task, filter the time segment in which the power regulation command is continuously applied, determine the distribution of the operating segment corresponding to the power command holding state, and obtain the set of frequency regulation operating segments. S212: Based on the set of frequency modulation operation segments, analyze the corresponding power command change process, compare the current change trajectory corresponding to the power command change position, filter the turning point time node of the current change direction, determine the path segment of the current trajectory that deviates from the stable change, and obtain the current response segment sequence. S213: Based on the current response segment sequence, compare the differences in the current change paths of each operating segment, filter the current change trajectory offset section position, determine the segment disturbance trajectory, adjust the time arrangement structure of the operating segments, and obtain the frequency modulation condition fluctuation.

5. The energy storage battery aging analysis method based on adaptive transfer learning according to claim 1, characterized in that, The specific steps for obtaining the cross-domain feature transfer correlation are as follows: S311: Based on the frequency modulation operating condition fluctuation, compare the time segments of the voltage change trajectory, filter the operating segments corresponding to the time positions of the capacity change stage, calculate the correspondence between voltage records and capacity records within the segment, and obtain the operating segment feature sequence. S312: Based on the feature sequence of the running segment, analyze the voltage coordinates and capacity coordinates, compare the difference between the target running segment coordinates and the source domain feature space reference center coordinates, calculate the voltage coordinate offset and capacity coordinate offset, and obtain the feature space offset. S313: Based on the feature space offset, compare the positional relationship of the running segments in the source domain feature space, calculate the mapping relationship of the target running segment coordinates, adjust the corresponding structure of the feature coordinates in the source domain feature space, and obtain the cross-domain feature transfer correlation degree.

6. The energy storage battery aging analysis method based on adaptive transfer learning according to claim 1, characterized in that, The specific steps for obtaining the adaptive migration confidence are as follows: S411: Based on the cross-domain feature migration correlation, analyze the voltage change trajectory sequence, calculate the time difference result of the capacity change sequence, filter the change inflection points in the capacity change sequence, determine the arrangement relationship of the inflection points in the time axis, and obtain the capacity inflection trajectory node set. S412: Based on the set of capacity transition trajectory nodes, compare the correspondence between the current trajectory node sequence and the reference trajectory node sequence, calculate the normalized amount of node voltage difference and capacity difference, determine the trajectory offset relationship, and obtain the trajectory offset coefficient. S413: Based on the trajectory offset coefficient, analyze the corresponding structure between the cross-domain feature migration correlation sequence, calculate the migration correlation correction amount, compare the distribution of the migration correlation correction amount in the time series, adjust the migration correlation weight, and obtain the adaptive migration confidence.

7. The energy storage battery aging analysis method based on adaptive transfer learning according to claim 1, characterized in that, The specific steps for obtaining the battery health status indicators are as follows: S511: Based on the adaptive migration confidence, analyze the cyclic advancement position relationship, compare the corresponding structure of the capacity sequence change trajectory and the voltage sequence change path, screen the operating segment where the capacity change shows a turning point, determine the time sequence relationship of the capacity change path, and obtain the capacity trajectory node sequence. S512: Based on the capacity trajectory node sequence, compare the morphological relationship of adjacent cyclic capacity trajectories, filter the operating segments where the capacity trajectory direction changes, determine the positional relationship of the capacity trajectory change stages, and obtain the aging stage position sequence. S513: Based on the aging stage position sequence, analyze the stage distribution, compare the correspondence between the capacity change trajectory and the voltage change path, filter the trajectory deviation operation segment, determine the capacity trajectory stage change state, adjust the health status discrimination structure, and obtain the battery health status index.

8. The energy storage battery aging analysis method based on adaptive transfer learning according to claim 1, characterized in that, The terminal voltage sequence refers to the sequence data formed by the battery terminal voltage data recorded by the battery management system at a fixed sampling time during the charging phase, arranged in chronological order. The frequency regulation operation segment refers to the operating time segment triggered by the power regulation command during the period when the energy storage power station performs the grid frequency regulation task.

Images