CN-122017608-A - Energy storage battery aging analysis method based on self-adaptive migration learning

CN122017608ACN 122017608 ACN122017608 ACN 122017608ACN-122017608-A

Abstract

The invention relates to the technical field of energy storage batteries, in particular to an energy storage battery aging analysis method based on self-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, adjusting and mapping to calculate a cross-domain characteristic migration association degree, evaluating track offset influence, updating association weight, comparing capacity change consistency, judging an aging evolution stage, and outputting a battery health state index. According to the invention, through the relation between terminal voltage and capacity change, a differential capacity curve peak position migration path structure is screened, the fluctuation degree of an operation section is analyzed by combining the output power of the converter, the characteristic migration confidence weight and mapping association structure are dynamically adjusted according to the track deviation relation, the key parameter fluctuation characteristic under complex working conditions is captured, the limitation that a static reference is difficult to adapt to a changeable environment is effectively overcome, and the accurate positioning and state discrimination of the battery in the healthy evolution stage are realized under the continuous fluctuation operation background.

Inventors

WANG SHUHAI
PANG YANWEN
ZHAO KANGNING
WANG GUOLI
GUO HAIYAN
XING MINGMING
GAO JICHAO

Assignees

临沂大学

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (8)

1. The energy storage battery aging analysis method based on self-adaptive transfer learning is characterized by comprising the following steps of: S1, based on the cycle record of a battery pack of an energy storage base station, collecting a voltage sequence and a capacity sequence of a charging stage end, screening a peak position migration path structure of a differential capacity curve, comparing a continuous cycle peak position change sequence, and judging a peak position migration path to obtain a cycle attenuation characterization quantity; S2, positioning a frequency modulation operation section based on the cyclic attenuation characterization quantity, analyzing the change process of the output power instruction of the converter, screening the corresponding switching position of the current response segment, and comparing the current change difference of each segment to obtain the fluctuation degree of the frequency modulation working condition; s3, screening time positions of capacity change stages based on the fluctuation degree of the frequency modulation working condition, comparing distribution positions of the running fragment features in a source domain feature space, and adjusting mapping positions of target fragments to obtain cross-domain feature migration association degree; S4, analyzing a capacity change process corresponding to the current voltage change track based on the cross-domain feature migration association degree, screening track change key positions to form a reference track, comparing the current track with the reference track offset relation, and adjusting a migration association weight structure to obtain self-adaptive migration confidence degree; and S5, analyzing a voltage change path corresponding to the capacity change track in the cyclic propulsion process based on the self-adaptive migration confidence, screening capacity change key node sections, comparing the consistency degree of the capacity change track, and adjusting a health state judging structure to obtain a battery health state index.
2. The method for aging analysis of an energy storage battery based on adaptive migration learning according to claim 1, wherein the cyclic attenuation characterization values include peak voltage migration values, peak capacity migration values and reaction platform displacement values, the frequency modulation condition fluctuation degree includes power command switching frequencies, current response variation amplitudes and current fluctuation duration intervals, the cross-domain feature migration correlation degree includes target operation segment feature coordinates, source domain feature space reference centers and feature space migration description values, the adaptive migration confidence degree includes track migration influence coefficients, feature migration correlation weights and migration relation adjustment coefficients, and the battery health state indexes include capacity retention rates, health state grade identifications and residual available life estimations.
3. The method for aging analysis of an energy storage battery based on adaptive transfer learning according to claim 1, wherein the step of obtaining the cyclic attenuation characterization quantity is specifically as follows: S111, based on the cyclic record of the battery pack of the energy storage base station, analyzing the change process of the terminal voltage sequence in the charging stage, calculating the change section relation corresponding to the capacity sequence, screening the local turning positions of the capacity change rate curve, comparing the voltage coordinates corresponding to the turning positions with the arrangement sequence of the capacity coordinates, and judging the corresponding relation of the continuous cyclic positions to obtain the coordinate sequence of the peak position; s112, analyzing the distribution condition of the peak positions of each cycle based on the coordinate sequence of the peak positions, comparing the moving directions of adjacent cycle voltage coordinates, screening the consistent position sequence of the directions in continuous cycles, calculating the connection relation of the corresponding capacity coordinate change sections, and adjusting the time arrangement sequence of the peak positions of each cycle to obtain a peak position migration path structure; s113, analyzing a continuous circulation peak position moving path structure based on the peak position moving path structure, comparing continuous conditions of the changing directions of the circulating voltage coordinates, screening path sections with consistent moving directions, and judging the corresponding relation of the capacity coordinate changing process to obtain the circulating attenuation characterization quantity.
4. The method for aging analysis of an energy storage battery based on adaptive transfer learning according to claim 1, wherein the step of obtaining the fluctuation degree of the frequency modulation condition is specifically as follows: S211, based on the cyclic attenuation characterization quantity, comparing the corresponding relation between the cyclic propulsion position and the time sequence of the energy storage frequency modulation task, screening time sections continuously acted by the power regulation instruction, and judging the distribution condition of the operation sections corresponding to the power instruction holding state to obtain a frequency modulation operation section set; S212, analyzing the corresponding power instruction change process based on the frequency modulation operation section set, comparing the current change track corresponding to the power instruction change position, screening turning time nodes of the current change direction, judging the path section of the current track deviating from stable change, and obtaining a current response fragment sequence; s213, comparing the difference conditions of the current change paths of the running segments based on the current response segment sequences, screening the positions of the offset sections of the current change tracks, judging the segment disturbance tracks, and adjusting the time arrangement structure of the running segments to obtain the fluctuation degree of the frequency modulation working condition.
5. The method for aging analysis of an energy storage battery based on adaptive migration learning according to claim 1, wherein the step of obtaining the cross-domain feature migration association degree is specifically as follows: S311, comparing time sections of the voltage change tracks based on the fluctuation degree of the frequency modulation working condition, screening an operation section corresponding to the time position of the capacity change stage, and calculating the corresponding relation between the voltage record and the capacity record in the section to obtain an operation fragment characteristic sequence; S312, analyzing voltage coordinates and capacity coordinates based on the operation fragment characteristic sequence, comparing the difference relation between the target operation fragment coordinates and the source domain characteristic space reference center coordinates, and calculating voltage coordinate offset and capacity coordinate offset to obtain characteristic space offset; s313, based on the characteristic space offset, comparing the position relation of the operation fragment in the source domain characteristic space, calculating the mapping relation of the target operation fragment coordinates, and adjusting the corresponding structure of the characteristic coordinates in the source domain characteristic space to obtain the cross-domain characteristic migration association.
6. The method for aging analysis of an energy storage battery based on adaptive migration learning according to claim 1, wherein the step of obtaining the adaptive migration confidence is specifically: s411, analyzing a voltage change track sequence based on the cross-domain feature migration association degree, calculating a time difference result of the capacity change sequence, screening change turning positions in the capacity change sequence, and judging an arrangement relation of the turning positions in a time axis to obtain a capacity turning track node set; S412, based on the capacity turning track node set, comparing the corresponding relation between the current track node sequence and the reference track node sequence, calculating the normalized quantity of the node voltage difference quantity and the capacity difference quantity, and judging the track offset relation to obtain a track offset coefficient; S413, based on the track offset coefficient, analyzing a corresponding structure between the track offset coefficient and the cross-domain characteristic migration association degree sequence, calculating migration association correction quantity, comparing the distribution state of the migration association correction quantity in the time sequence, and adjusting migration association weight to obtain the self-adaptive migration confidence degree.
7. The method for aging analysis of an energy storage battery based on adaptive transfer learning according to claim 1, wherein the step of obtaining the battery state of health index specifically comprises: S511, analyzing a cyclic propulsion position relation based on the self-adaptive migration confidence, comparing a capacity sequence change track with a voltage sequence change path corresponding structure, screening an operation section with turning of capacity change, and judging a time sequence relation of the capacity change path to obtain a capacity track node sequence; s512, comparing the form relation of adjacent circulating capacity trajectories based on the capacity trajectory node sequence, screening a running section with the changed capacity trajectory direction, and judging the position relation of a capacity trajectory change stage to obtain a position sequence of an aging stage; S513, based on the aging stage position sequence, analyzing stage distribution conditions, comparing the corresponding relation between the capacity change track and the voltage change path, screening track deviation operation sections, judging the state of the capacity track stage change, and adjusting a health state judging structure to obtain battery health state indexes.
8. The method for aging analysis of an energy storage battery based on adaptive transfer learning according to claim 1, wherein the terminal voltage sequence refers to sequence data formed by a battery management system according to a time sequence of battery terminal voltage data recorded by a fixed sampling time in a charging stage, and the frequency modulation operation section refers to an operation time section formed by triggering a power adjustment instruction during the execution of a grid frequency modulation task by an energy storage power station.

Description

Energy storage battery aging analysis method based on self-adaptive migration learning Technical Field The invention relates to the technical field of energy storage batteries, in particular to an energy storage battery aging analysis method based on self-adaptive transfer learning. Background The technical field of energy storage batteries relates to the interconversion and energy storage of electrochemical energy and electric energy, and comprises battery material research and development, battery cell manufacturing, battery pack structural design and a battery management system, and smooth output, peak clipping and valley filling and power grid frequency adjustment of the electric energy are realized through a power electronic converter and monitoring equipment. The traditional aging analysis method of the energy storage battery refers to the problem of evolution of the energy storage battery in the state of health such as capacity fading, internal resistance increase and the like in the process of charge and discharge cycles, and generally relies on constant-current charge and discharge tests, alternating-current impedance spectrum analysis or establishment of an equivalent circuit model, so that the aging degree of the battery is evaluated. In the traditional method, constant-current charge and discharge tests are relied on or an equivalent circuit model is established as an evaluation basis, the static parameters are set to be difficult to respond to frequency modulation power instructions frequently fluctuating by an energy storage base station in actual operation, test data obtained by a fixed mode alone cannot accurately reflect deep coupling evolution tracks of dynamic characteristics under variable working conditions, extraction of healthy characteristics such as internal resistance increase and the like is deviated from the actual conditions due to long-term use of a fixed judging structure, the deep aging evolution path is difficult to accurately track under variable disturbance, and the reliability degree of integral energy storage and smooth output is severely restricted. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides an energy storage battery aging analysis method based on self-adaptive transfer learning. In order to achieve the above purpose, the invention adopts the following technical scheme that the energy storage battery aging analysis method based on self-adaptive transfer learning comprises the following steps: S1, based on the cycle record of a battery pack of an energy storage base station, collecting a voltage sequence and a capacity sequence of a charging stage end, screening a peak position migration path structure of a differential capacity curve, comparing a continuous cycle peak position change sequence, and judging a peak position migration path to obtain a cycle attenuation characterization quantity; S2, positioning a frequency modulation operation section based on the cyclic attenuation characterization quantity, analyzing the change process of the output power instruction of the converter, screening the corresponding switching position of the current response segment, and comparing the current change difference of each segment to obtain the fluctuation degree of the frequency modulation working condition; s3, screening time positions of capacity change stages based on the fluctuation degree of the frequency modulation working condition, comparing distribution positions of the running fragment features in a source domain feature space, and adjusting mapping positions of target fragments to obtain cross-domain feature migration association degree; S4, analyzing a capacity change process corresponding to the current voltage change track based on the cross-domain feature migration association degree, screening track change key positions to form a reference track, comparing the current track with the reference track offset relation, and adjusting a migration association weight structure to obtain self-adaptive migration confidence degree; and S5, analyzing a voltage change path corresponding to the capacity change track in the cyclic propulsion process based on the self-adaptive migration confidence, screening capacity change key node sections, comparing the consistency degree of the capacity change track, and adjusting a health state judging structure to obtain a battery health state index. The invention is improved in that the cyclic attenuation characterization quantity comprises peak voltage migration quantity, peak capacity migration quantity and reaction platform displacement quantity, the fluctuation degree of the frequency modulation working condition comprises power command switching frequency, current response change amplitude and current fluctuation duration, the cross-domain characteristic migration association degree comprises target operation fragment characteristic coordinates, a source domain characteristic space r