Search

CN-122017655-A - Battery health prediction method, system and storage medium based on self-adaptive feature fusion and double-branch frequency domain decomposition

CN122017655ACN 122017655 ACN122017655 ACN 122017655ACN-122017655-A

Abstract

The application relates to the technical field of new energy and artificial intelligence, in particular to a battery health prediction method, a system and a storage medium based on self-adaptive feature fusion and double-branch frequency domain decomposition. The method comprises the steps of sorting and forming a fusion key feature sequence based on attention mechanism fusion, dividing the fusion key feature sequence into a low-frequency feature sequence and a high-frequency feature sequence through a threshold value, taking the low-frequency feature sequence as a sampling sequence of a battery degradation trend, taking the high-frequency feature sequence as a sampling sequence of a capacity regeneration phenomenon, and respectively obtaining a battery degradation trend feature and a nonlinear fluctuation feature after extraction and capture of a two-way gating circulation unit and a complex neural network, so that prediction of a battery health result is carried out based on the two features. The method aims to solve the problem of capturing the characteristics of the long-term attenuation trend and capacity regeneration phenomenon of the battery.

Inventors

  • YANG BO
  • DONG SHA
  • ZHANG CHEN
  • JIANG LIN

Assignees

  • 昆明理工大学

Dates

Publication Date
20260512
Application Date
20260414

Claims (10)

  1. 1. The battery health prediction method based on self-adaptive feature fusion and double-branch frequency domain decomposition is characterized by comprising the following steps of: S10, acquiring fusion key characteristics of a battery to be predicted, wherein the fusion key characteristics are obtained by weighting and fusing candidate battery health characteristics of the battery to be predicted by adopting an attention mechanism; S20, acquiring fusion key features acquired in a preset period of time, sequencing the fusion key features according to a time sequence to form a fusion key feature sequence, determining a sequence representation corresponding to the fusion key feature sequence in a frequency domain, and dividing the sequence representation into a low-frequency feature sequence and a high-frequency feature sequence according to a preset frequency threshold; s30, extracting battery degradation trend characteristics in the low-frequency characteristic sequence based on a bidirectional gating circulating unit, and extracting nonlinear fluctuation characteristics in the high-frequency characteristic sequence based on a complex-valued neural network; And S40, generating a battery health prediction result of the battery to be predicted based on the battery degradation trend characteristic and the nonlinear fluctuation characteristic.
  2. 2. The method for predicting battery health based on adaptive feature fusion and dual-branch frequency domain decomposition according to claim 1, wherein in step S30, the step of extracting the characteristic of degradation trend of the battery comprises: S31, defining a low-frequency signal sequence: ; In the formula, Length of sequence, for each time instant The low frequency branch circuit carries out gate control state update through the GRU unit: ; ; ; ; In the formula, On behalf of the update gate, Representing a reset gate and, Representing a hidden state of the object, 、 And Respectively representing the input weight matrix of the update gate, resetting the input weight matrix of the gate, and candidate hidden state; 、 And Respectively represent the hidden state weight matrix of the update gate, the hidden state weight matrix of the reset gate, the hidden state weight matrix of the candidate hidden state, 、 And Respectively representing an offset vector of an update gate, an offset vector of a reset gate and an offset vector of a candidate hidden state; representing an offset vector of the update gate, resetting the offset vector of the gate, and candidate hidden state; s32, introducing a bidirectional structure to extract low-frequency signal input History information of (a) And future information : ; ; Obtaining the characteristic of degradation trend of the battery Expressed as: ; The nonlinear fluctuation feature extraction step comprises the following steps: s33, the high frequency signal of the high frequency branch is processed Expressed in complex form : ; In the formula, Is the amplitude information of the amplitude of the signal, Is phase information, T represents the length of a time sequence participating in modeling; S32, the complex high-frequency signal is obtained by complex linear mapping process : ,l=1,2,...,L; Wherein W represents a complex weight matrix, b represents complex bias, and L represents a neural network layer; definition: ; Will be The expansion is calculated as real and imaginary parts: ; the activation function takes the form of a complex valued nonlinear mapping: ,l=1,2,...,L; s34, obtaining nonlinear fluctuation characteristics by stacking complex-valued neural network layers : ; In the formula, Is a complex-valued mapping function.
  3. 3. The method for predicting battery health based on adaptive feature fusion and dual-branch frequency domain decomposition according to claim 1, wherein step S40 comprises: S41, respectively calculating frequency domain energy distribution of the low-frequency characteristic sequence and the high-frequency characteristic sequence: ; ; In the formula, The sequence of low-frequency features is represented, Representing a high frequency signature sequence; A frequency domain energy distribution representing a low frequency signature sequence; A frequency domain energy distribution representing a high frequency signature sequence; S42, calculating the corresponding attention weights according to the frequency domain energy distribution of the two components: ; ; In the formula, Attention weights for low frequency feature sequences; Attention weights for high frequency feature sequences; s43, carrying out feature fusion according to the attention weights of the two to obtain fusion features : ; In the formula, Is characterized by the degradation tendency of the battery, Representing nonlinear fluctuation characteristics; s44, according to the fusion characteristics And predicting to generate a battery health prediction result of the battery to be predicted.
  4. 4. The method for predicting battery health based on adaptive feature fusion and dual-branch frequency domain decomposition according to claim 1, wherein said selecting the candidate battery health features comprises: Selecting initial characteristics in the collected pretreatment battery operation data set; determining a correlation coefficient between the initial feature and an measured battery health value of a battery health feature dataset; and taking the initial characteristic larger than a preset correlation coefficient as the candidate battery health characteristic.
  5. 5. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition according to claim 4, wherein said initial features include time domain features, statistical features and cyclic features; wherein the time domain features include at least one of charge voltage plateau length, discharge voltage drop rate, and charge-discharge energy difference; The statistical features comprise at least one of voltage fluctuation variance, current peak duty ratio and temperature mean; The cycle characteristics include at least one of a unit cycle capacity delta and a cycle life ratio.
  6. 6. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition according to claim 4, wherein said preprocessing step of preprocessing a battery operation dataset comprises: Determining dynamic current data/dynamic voltage data in an initial battery operation data set, and eliminating peak noise in the dynamic current data/dynamic voltage data by adopting sliding window median filtering, wherein the length of a sliding window is adjusted based on the data fluctuation frequency of the dynamic current data/dynamic voltage data; determining abnormal data in the initial battery operation data set, and repairing the abnormal data by adopting a linear interpolation method; And carrying out data normalization on the initial battery operation data set repaired by the sliding window median filtering and linear interpolation method to obtain the preprocessed battery operation data set.
  7. 7. The method for predicting battery health based on adaptive feature fusion and dual-branch frequency domain decomposition according to any one of claims 4 to 6, wherein said step of generating the fusion key feature comprises: Determining individual candidate battery health characteristics Corresponding feature weights : ; In the formula, Is a weight adjustment coefficient; A correlation coefficient of the j-th candidate battery health characteristic and the battery health state, In order to be a function of the attention, A coefficient of correlation representing a kth candidate battery health characteristic and a battery state of health; Characterizing individual candidate battery health And corresponding sign weights thereof And (3) weighting and summing to obtain the fusion key characteristics: ; wherein m is the number of the health features of the candidate battery.
  8. 8. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition according to claim 1, wherein the battery health prediction result comprises a battery health prediction value; and determining the grade of the battery health of the battery to be predicted according to the interval of the battery health predicted value.
  9. 9. A computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of the battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition according to any one of claims 1 to 8.
  10. 10. A computer readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of the battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition according to any of claims 1 to 8.

Description

Battery health prediction method, system and storage medium based on self-adaptive feature fusion and double-branch frequency domain decomposition Technical Field The application relates to the technical field of new energy and artificial intelligence, in particular to a battery health prediction method, a system and a storage medium based on self-adaptive feature fusion and double-branch frequency domain decomposition. Background Along with the rapid development of new energy industry, the battery is used as a core energy storage component, and the health state of the battery directly determines the running reliability, the cruising ability and the service life of the equipment. The State of Health (SOH) of a battery is generally defined as the ratio of the current capacity to the rated capacity of the battery, and is a key indicator for measuring the performance degradation of the battery. However, in the long-term cyclic use process of the battery, SOH is affected by factors such as material aging (such as electrode active material falling and electrolyte decomposition), working condition fluctuation (such as dynamic charge and discharge current and temperature change), environmental interference (such as humidity and vibration) and the like, and is attenuated in a nonlinear trend, so that the traditional prediction method is difficult. In the related technical scheme, a time sequence model of time domain modeling is often adopted to predict the attenuation curve of the SOH of the battery. However, the SOH curve of the battery plotted in this way ignores two phenomena that are difficult to capture simultaneously, long-term decay trend and capacity regeneration phenomenon. The former means that the battery capacity increases with time and cycle number, and the trend is an irreversible slow decrease trend, and the trend is irregular increase along with the complexity of the battery working condition, and the latter means that the battery capacity is temporarily restored or fluctuated in a short term and small amplitude under the long-term attenuation trend, but the battery capacity is not truly increased. Both of these phenomena lead to a large error between the predicted value and the actual value of the battery SOH. In view of this, the application proposes a new battery state of health prediction method, which aims to realize feature capture of long-term attenuation trend and capacity regeneration phenomenon, thereby improving SOH prediction accuracy of the battery under complex working conditions. Disclosure of Invention The application mainly aims to provide a battery health prediction method based on self-adaptive feature fusion and double-branch frequency domain decomposition, which aims to solve the problem of capturing the features of long-term attenuation trend and capacity regeneration phenomenon of a battery. In order to achieve the above object, the present application provides a battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition, the method comprising: S10, acquiring fusion key characteristics of a battery to be predicted, wherein the fusion key characteristics are obtained by weighting and fusing candidate battery health characteristics of the battery to be predicted by adopting an attention mechanism; S20, acquiring fusion key features acquired in a preset period of time, sequencing the fusion key features according to a time sequence to form a fusion key feature sequence, determining a sequence representation corresponding to the fusion key feature sequence in a frequency domain, and dividing the sequence representation into a low-frequency feature sequence and a high-frequency feature sequence according to a preset frequency threshold; s30, extracting battery degradation trend characteristics in the low-frequency characteristic sequence based on a bidirectional gating circulating unit, and extracting nonlinear fluctuation characteristics in the high-frequency characteristic sequence based on a complex-valued neural network; And S40, generating a battery health prediction result of the battery to be predicted based on the battery degradation trend characteristic and the nonlinear fluctuation characteristic. Optionally, in step S30, the extracting step of the battery degradation trend feature includes: S31, defining a low-frequency signal sequence: In the formula, Length of sequence, for each time instantThe low frequency branch circuit carries out gate control state update through the GRU unit: ; ; ; ; In the formula, On behalf of the update gate,Representing a reset gate and,Representing a hidden state of the object,、AndRespectively representing the input weight matrix of the update gate, resetting the input weight matrix of the gate, and candidate hidden state;、 And Respectively represent the hidden state weight matrix of the update gate, the hidden state weight matrix of the reset gate, the hidden state weight matrix of the c