Search

CN-122017649-A - On-line evaluation method and system for health state of sodium ion battery Pack

CN122017649ACN 122017649 ACN122017649 ACN 122017649ACN-122017649-A

Abstract

The invention discloses a method and a system for on-line evaluation of the health state of a sodium ion battery Pack, wherein the method comprises the following steps of S1, constructing a sodium-electricity aging model, collecting and preprocessing Pack operation data to generate a data sequence, S2, calculating a characteristic entropy value, adjusting the state of a liquid neural network, extracting a fusion characteristic sequence, S3, carrying out path signature transformation on the fusion characteristic and a simulation sequence, calculating structural deviation to generate a deviation sequence, S4, utilizing particle filtering to update model parameters, reasoning capacity estimation values to form a capacity sequence, S5, carrying out Loess fitting and smoothing filtering on the capacity sequence to generate a capacity estimation curve, S6, evaluating the health state of the Pack according to the capacity estimation curve, and outputting an on-line result. The invention integrates digital twin modeling, a liquid neural network and the like, and has the advantages of high evaluation accuracy, high response speed and strong adaptability.

Inventors

  • YE YIN

Assignees

  • 四川储元世纪新能源科技有限公司

Dates

Publication Date
20260512
Application Date
20260228

Claims (10)

  1. 1. The on-line evaluation method for the health state of the sodium ion battery Pack is characterized by comprising the following steps of: s1, constructing a battery aging model based on a preset sodium-electricity aging mechanism, collecting multidimensional operation data of a sodium ion battery Pack according to a preset frequency, and preprocessing the multidimensional operation data to generate an operation data sequence; S2, extracting characteristic entropy values of each time step of the operation data sequence, dynamically adjusting the neural state of the liquid neural network in the liquid neural network according to the characteristic entropy values, and executing characteristic extraction and fusion operation on the operation data sequence to obtain a fusion characteristic sequence; s3, extracting a simulation fusion sequence generated in the battery aging model, respectively executing path signature transformation operation on the simulation fusion sequence and the fusion characteristic sequence, executing structure error analysis operation on two transformation results, and calculating a structure deviation value to construct a deviation sequence; S4, performing parameter self-adaptive updating operation on the battery aging model according to the deviation sequence by adopting a particle filtering algorithm, and performing reasoning calculation operation on each time step by utilizing the updated battery aging model to obtain a corresponding capacity estimation value so as to form a capacity estimation sequence; s5, performing curve fitting on the capacity estimation sequence by adopting a Loess algorithm, and performing smooth filtering operation on a fitting result to generate a capacity estimation curve; and S6, performing health state evaluation operation on the sodium ion battery Pack according to the capacity estimation curve, and generating a health state online evaluation result.
  2. 2. The method of claim 1, wherein the sodium-electrical aging model is a digital twin model for simulating the operational state of the sodium-ion battery Pack, the multidimensional operational data includes voltage, current, temperature and environmental parameters of each cell in the sodium-ion battery Pack, the preprocessing includes deletion filling, time alignment and abnormal rejection operations, and the neural state represents an internal response vector of the liquid neural network driving evolution in each time step according to the operational data sequence and the corresponding characteristic entropy value.
  3. 3. The method for online evaluation of the health status of a sodium ion battery Pack according to claim 1, wherein the internal calculation process of the battery aging model specifically comprises: Collecting accumulated use time length, accumulated charge and discharge times, historical highest operation temperature and average discharge multiplying power of a sodium ion battery Pack, carrying out normalization treatment, linearly mapping an original numerical value to a [0,1] interval, and constructing a battery characteristic vector comprising a long-time use factor, a temperature factor, a circulation factor and a multiplying power factor, wherein the long-time factor represents the ratio of the accumulated use time length to the rated service life, the temperature factor represents the ratio of the historical highest operation temperature to the rated safe working temperature set by a manufacturer, the circulation factor represents the ratio of the accumulated charge and discharge times to the designed cycle life time of the battery, and the multiplying power factor represents the ratio of the average discharge multiplying power to the standard rated multiplying power; setting the total simulation period length according to the duration factor in the battery characteristic vector, and setting the capacity attenuation coefficient of each simulation time step according to the circulation factor; In the first time step, setting the battery capacity of the sodium ion battery Pack as an analog capacity value of the last time step; at each time step, taking the analog capacity value of the last time step as a starting capacity, and executing capacity fading calculation, wherein the method specifically comprises the following steps: multiplying the current temperature factor by a set temperature sensitivity coefficient, and calculating to obtain a temperature adjustment quantity; Multiplying the product of the current multiplying factor and the circulating factor by a preset multiplying factor sensitivity coefficient, and calculating to obtain multiplying factor influence; according to a preset weight proportion, carrying out weighted superposition on the temperature adjustment quantity and the multiplying power influence quantity, multiplying the superposition result by a preset capacity attenuation coefficient to obtain a capacity attenuation value of the current time step, and subtracting the capacity attenuation value from the initial capacity to obtain a current analog capacity value; Arranging all analog capacity values according to a time sequence to construct an analog capacity sequence; In each time step, respectively deducing an analog voltage value, an analog current value and an analog temperature value of the current time step according to the battery characteristic vector and the current time step index, and splicing the analog voltage value, the analog current value and the analog temperature value with the environmental parameters acquired in real time to form an analog operation sequence; And performing feature extraction and fusion operation on the simulation run sequence through the liquid neural network to obtain a simulation fusion sequence consistent with the fusion feature sequence in time dimension, feature dimension and arrangement sequence.
  4. 4. The method for online evaluation of health status of a sodium ion battery Pack according to claim 1, wherein S2 specifically comprises: S21, expanding an operation data sequence according to time steps, and respectively calculating the maximum value, the minimum value, the mean value and the change rate of four types of data in a sliding window according to the voltage, the current, the temperature and the environmental parameters of each battery cell in each time step to form a characteristic statistical set of the current time step; S22, calculating the characteristic entropy value of each time step based on the characteristic statistics set of the time step, arranging all the characteristic entropy values to form a characteristic entropy sequence, and performing time alignment with the operation data sequence, wherein the calculation process of the characteristic entropy value specifically comprises the following steps: dividing the characteristic statistical set of the current time step into a voltage subset, a current subset, a temperature subset and an environment parameter subset according to data types, and respectively executing Min-Max normalization operation on the maximum value, the minimum value, the mean value and the change rate in each subset; In each subset, dividing the characteristic value into a plurality of continuous non-overlapping value range intervals according to a preset interval dividing rule based on the normalized characteristic value of the corresponding characteristic of the subset in the current sliding window; Counting the number of times that all normalized characteristic values in the current sliding window fall into each value range interval as an interval frequency, dividing the interval frequency by the number of the normalized characteristic values in the sliding window, and obtaining a normalized probability value corresponding to each value range interval; combining the normalized probability values corresponding to all the value range intervals into probability distribution vectors of the current features in the sliding window; Calculating local entropy values corresponding to each subset according to probability distribution vectors of the normalized features, and carrying out weighted summation on the local entropy values of the four subsets according to a preset weight coefficient set to obtain a feature entropy value of the current time step; s23, inputting a time alignment result into the liquid neural network, dynamically adjusting the neural state of the liquid neural network according to the characteristic entropy value in each time step, and updating the internal response vector of the time step; s24, based on the internal response vector and the operation data of each time step, performing feature enhancement operation in the liquid neural network, extracting fusion feature vectors, and generating fusion feature sequences according to time sequence.
  5. 5. The method for online evaluation of health status of a sodium ion battery Pack according to claim 4, wherein S23 specifically comprises: s231, extracting a characteristic entropy value of the current time step in each time step, matching the characteristic entropy value with a preset adjusting function, and calculating to obtain an adjusting weight; S232, dynamically scaling the time constant of each neuron in the liquid neural network according to the adjustment weight, and taking the product of the original time constant and the adjustment weight as the updated time constant of the current time step; S233, according to the updated time constant, carrying out weighted fusion on the operation data of the current time step and the internal response vector of the previous time step to generate the internal response vector of the current time step; S234, arranging the internal response vectors under all time steps in time sequence, constructing an internal response sequence consistent with the time dimension of the operation data sequence, and keeping consistency of the characteristic dimension.
  6. 6. The method for online evaluation of health status of a sodium ion battery Pack according to claim 1, wherein the step S3 specifically comprises: S31, according to the time dimension of the fusion characteristic sequence, reading a simulation fusion sequence corresponding to the time step in a battery aging model; S32, respectively expanding the fusion characteristic sequence and the simulation fusion sequence according to time steps, and respectively constructing corresponding path sequences according to a fixed arrangement sequence; S33, taking the path vector difference value between adjacent time steps in each path sequence as a path increment, and gradually accumulating the path increment according to the time sequence to generate a path representation; S34, in each time step, path signature transformation operation is carried out on the path representation, path integral values under different orders are extracted, and a fusion signature sequence and a simulation signature sequence are formed; S35, carrying out dimension-by-dimension difference value calculation on two path integral values of corresponding time steps in the fused signature sequence and the simulated signature sequence, summing absolute values of the dimension difference values, and carrying out Z-Score standardization operation to obtain a structural deviation value of the current time step; S36, arranging all the structural deviation values according to a time sequence, and constructing a deviation sequence.
  7. 7. The method for online evaluation of health status of a sodium ion battery Pack according to claim 6, wherein S34 specifically comprises: S341, in each time step, taking the current time step as a cut-off point according to the path increment arrangement sequence in the fused path representation and the simulated path representation, and intercepting a path increment sequence from the first time step to the current time step as a path representation fragment of the current time step; S342, performing multi-order path integral calculation operation on the path representation segment, sequentially calculating path integral values of a first order, a second order and a preset highest order, and constructing a path integral set corresponding to the current time step, wherein the path integral value of each order is accumulated and calculated according to a fixed arrangement sequence by multi-dimensional feature values in a path increment sequence; S343, executing feature mapping operation on the path integral set, and respectively arranging path integral values under different orders according to feature types by adopting a dimension unfolding mode to generate a path signature vector matched with the path representation; s344, the path signature vectors generated in all time steps are arranged in time sequence, and a fusion signature sequence and a simulation signature sequence are respectively formed according to the sources.
  8. 8. The method for online evaluation of health status of a sodium ion battery Pack according to claim 1, wherein S4 specifically comprises: S41, setting a capacity attenuation coefficient, a temperature sensitivity coefficient and a multiplying power sensitivity coefficient in a battery aging model as parameter sets to be estimated of a particle filtering algorithm, taking fusion characteristic vectors of corresponding time steps in a fusion characteristic sequence as state driving, and initializing a particle set containing corresponding weights of all particles; s42, in each time step, based on the particle set of the previous time step, performing state prediction updating operation on the parameter set to be estimated, substituting a prediction result into a battery aging model, calculating a prediction capacity value of the current time step, and matching the prediction capacity value with a structural deviation value of the corresponding time step in a deviation sequence to construct an observation error of the current time step; S43, performing weight updating operation on each particle in the particle set according to the observation error of the current time step, wherein the particle weight with smaller observation error is higher, and performing Min-Max normalization processing on the weights updated by all the particles to generate an updated particle set, and specifically comprising the following steps: Performing a Z-Score normalization process on the observed error of each particle, and calculating an error Score value, wherein the lower the error Score value is, the closer the particle is to the whole error distribution center; performing mapping operation according to the error grading value, mapping the error grading value into non-negative weight by adopting a monotonically decreasing preset weight function, and constructing an initial weight set under the current time step; Performing Min-Max normalization operation on the initial weight set, and linearly compressing the weights of all particles to a set interval range to form a normalized weight set; Mapping the normalized weight set back to the particle set, and updating the weights of all particles in the current time step; s44, performing resampling operation on the updated particle set, removing particles with weights lower than a preset threshold, sorting the particles according to the weights, selecting particles with weights with a preset proportion before the ranking, and performing resampling replication operation to form a resampled particle set in the current time step; S45, taking a particle parameter mean value in the resampled particle set as an updating parameter of the battery aging model in the current time step, substituting the updating parameter into the battery aging model, and executing capacity reasoning calculation in the current time step to obtain a corresponding capacity estimation value; s46, arranging the capacity estimation values obtained in each time step according to the time sequence, and constructing a capacity estimation sequence.
  9. 9. The method for online evaluation of health status of a sodium ion battery Pack according to claim 1, wherein S5 specifically comprises: S51, setting sliding windows with fixed lengths for capacity estimation sequences according to time steps, and adopting a weighted polynomial regression method to perform local fitting on capacity estimation values in each sliding window to generate local fitting values corresponding to the time steps; s52, arranging the local fitting values under all time steps according to the time step sequence, constructing an initial capacity curve, and executing continuity detection operation on the initial capacity curve to identify mutation points; S53, performing curve splicing adjustment operation on local fitting values before and after the mutation points, and performing Savitzky-Golay filtering operation on the adjustment result to generate an intermediate capacity curve, wherein the method specifically comprises the following steps of: setting symmetrical splicing intervals at two sides of the mutation point, extracting local fitting values in corresponding time periods, and constructing fitting curve segments; setting a smooth weight according to the endpoint slope change trend and the numerical value difference of the fitting curve segment, and performing weighted superposition operation on the local fitting value in the splicing section to generate a smooth transition curve; Splicing the smooth transition curve and the fitted curve section according to time sequence, and constructing a corrected initial capacity curve as an adjustment result; on the adjustment result, setting the length of a sliding window and the order of a polynomial, smoothing the adjustment result by adopting a Savitzky-Golay filtering algorithm, and extracting a filtering value of each time step; all the filtering values are arranged in time sequence to generate an intermediate capacity curve; And S54, executing monotonicity judging processing on the middle capacity curve, and executing curve segment replacing operation to form a capacity estimation curve if the local fluctuation amplitude is found to exceed the abnormal interval of the set threshold value.
  10. 10. An online evaluation system for the health status of a sodium ion battery Pack, which performs the online evaluation method for the health status of a sodium ion battery Pack according to any one of claims 1 to 9, comprising: the data acquisition module is used for constructing a battery aging model based on a preset sodium-electricity aging mechanism, acquiring multidimensional operation data of the sodium ion battery Pack according to a preset frequency, and preprocessing the multidimensional operation data to generate an operation data sequence; The feature fusion module is used for extracting the feature entropy value of each time step of the operation data sequence, dynamically adjusting the neural state of the liquid neural network according to the feature entropy value in the liquid neural network, and executing feature extraction and fusion operation on the operation data sequence to obtain a fusion feature sequence; The deviation analysis module is used for extracting a simulation fusion sequence generated in the battery aging model, respectively executing path signature transformation operation on the simulation fusion sequence and the fusion characteristic sequence, executing structure error analysis operation on two transformation results, and calculating a structure deviation value to construct a deviation sequence; The capacity estimation module is used for performing parameter self-adaptive updating operation on the battery aging model according to the deviation sequence by adopting a particle filtering algorithm, performing reasoning calculation operation on each time step by utilizing the updated battery aging model to obtain a corresponding capacity estimation value, and forming a capacity estimation sequence; The curve fitting module is used for performing curve fitting on the capacity estimation sequence by adopting a Loess algorithm, and performing smooth filtering operation on the fitting result to generate a capacity estimation curve; and the state evaluation module is used for executing the health state evaluation operation on the sodium ion battery Pack according to the capacity evaluation curve to generate an on-line health state evaluation result.

Description

On-line evaluation method and system for health state of sodium ion battery Pack Technical Field The invention relates to the technical field of battery management, in particular to a method and a system for on-line evaluation of the health state of a sodium ion battery Pack. Background As a new energy storage battery system which has been paid attention in recent years, sodium ion batteries are becoming one of potential schemes for replacing lithium batteries due to the advantages of abundant resource reserves, low raw material cost, strong high-temperature adaptability and the like. In the fields of electric power energy storage, low-speed electric vehicles and the like, the practical application scene of the sodium ion battery Pack system is continuously expanded, and the requirements on the operation reliability and the safety of the sodium ion battery Pack system are increasingly improved. In this context, how to accurately evaluate the state of health (SOH) of the sodium ion battery Pack becomes a key factor affecting system operation and maintenance and life management. In the prior art, the battery state of health evaluation method is mostly based on an empirical model or an electrochemical modeling structure of a lithium ion battery, and common means comprise static internal resistance inference, capacity curve comparison, characteristic fitting regression and the like. The method has a certain effect under experimental conditions, but has the defects that firstly, most of traditional models are based on single cells, the influence of inconsistency among the cells in a Pack level on the whole health state is difficult to effectively reflect, secondly, the lithium ion battery and the sodium ion battery have obvious differences in aging mechanism, capacity attenuation characteristic and electrochemical response, the lithium ion battery model is directly migrated to a sodium electric scene, the evaluation accuracy is low, thirdly, most of the traditional methods adopt static models or offline evaluation means, the real-time dynamic monitoring in the running process is difficult to realize, and a model updating mechanism synchronous with the actual working condition is lacked. In addition, few researches try to introduce data driving models such as a neural network and an LSTM (least squares) for SOH estimation, but the problems of weak generalization capability, characteristic response lag, difficulty in expression of a high-data structure and the like still exist, and especially in a scene that the operation characteristics of a sodium ion battery show nonlinear fluctuation and long-period change, the existing network structure is difficult to adapt to double challenges of time sequence dynamic and structure disturbance. Therefore, how to provide a method and a system for online evaluation of the health status of a sodium ion battery Pack is a problem that needs to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide a method and a system for on-line evaluation of the health state of a sodium ion battery Pack, which are used for integrating digital twin modeling, a liquid neural network, path signature transformation and a particle filtering algorithm, constructing a special dynamic model aiming at the aging characteristics of the sodium ion battery, extracting the characteristics of a multidimensional sensor in real time and adaptively correcting an SOH estimation result in the running process, and describing the whole process of realizing the Pack-level health state accurate evaluation under the non-stable time sequence working condition in detail. According to the embodiment of the invention, the on-line evaluation method for the health state of the sodium ion battery Pack comprises the following steps: s1, constructing a battery aging model based on a preset sodium-electricity aging mechanism, collecting multidimensional operation data of a sodium ion battery Pack according to a preset frequency, and preprocessing the multidimensional operation data to generate an operation data sequence; S2, extracting characteristic entropy values of each time step of the operation data sequence, dynamically adjusting the neural state of the liquid neural network in the liquid neural network according to the characteristic entropy values, and executing characteristic extraction and fusion operation on the operation data sequence to obtain a fusion characteristic sequence; s3, extracting a simulation fusion sequence generated in the battery aging model, respectively executing path signature transformation operation on the simulation fusion sequence and the fusion characteristic sequence, executing structure error analysis operation on two transformation results, and calculating a structure deviation value to construct a deviation sequence; S4, performing parameter self-adaptive updating operation on the battery aging model according to the deviation sequence by adopting a