CN115684972A - Lithium ion battery SOH estimation method based on SSA-SVR model - Google Patents

Abstract

The invention discloses a lithium ion battery SOH estimation method based on an SSA-SVR model. The technical scheme adopted by the invention is as follows: collecting data; extracting health characteristics; constructing a characteristic processing model based on Savitzky-Golay filtering; taking a matrix F = [ F1, F2, F3 and F4] formed by the extracted and processed health characteristics as input, taking the SOH of the battery as output, and constructing a training set and a prediction set; and 5, constructing an SSA-SVR model: and constructing a reference SVR model, optimizing the hyperparameters of the SVR model by adopting a sparrow search algorithm based on the training set, constructing an SOH estimation model, and outputting the estimated SOH. The invention adopts a SVR model which considers 4 indexes of voltage and current segment data in the charging process, processes characteristics and adjusts parameters by combining a sparrow search algorithm, thereby realizing the SOH prediction of the lithium ion battery.

Description

Lithium ion battery SOH estimation method based on SSA-SVR model

Technical Field

The invention relates to the technical field of lithium ion battery state estimation, in particular to a lithium ion battery SOH estimation method based on an SSA-SVR model.

Background

The lithium ion battery has the advantage that good balance can be achieved among various performance parameters such as cost, service life, safety and environmental influence, and plays a vital role in the fields of power grid energy storage systems and new energy automobiles. However, the performance of the lithium ion battery gradually decreases with the use period, which causes a safety hazard to the battery system. In order to improve the capability of a system for resisting risk faults and reduce the maintenance cost, the state of health (SOH) of the lithium ion battery needs to be accurately estimated. However, the parameters directly related to the SOH of the battery are internal variables, are difficult to measure directly with sensors, and need to be estimated by measuring the characterization parameters. The SOH of a lithium ion battery is generally defined as the ratio of the measured capacity to the nominal capacity, and when the measured capacity is reduced to a certain extent, the battery cell should be replaced and maintained. Existing SOH estimation methods can be divided into three categories: model-based methods, data-driven based methods, and fusion-based methods. The model-based method requires a high-complexity model to be constructed in order to obtain a high-precision SOH estimation result, and is difficult to be practically applied. The fusion-based method aims at combining the advantages of different methods, but has not made great progress, has the problems of model compatibility and higher computational complexity. The data-driven method does not need to analyze a complex reaction mechanism, and can extract a hidden mapping relation from data. Therefore, in contrast, a data-driven method with more flexibility and stronger applicability is a research hotspot at present. The data-driven-based method is mainly influenced by two aspects of input features and a machine learning algorithm.

The high-quality input features have the characteristics of stable required data, easy extraction and high correlation. In some researches, data in the discharging process is adopted for feature extraction, but the discharging process is greatly influenced by working conditions and loads, so that the data is not as stable as the data in the charging process, and the use of the features of the discharging data is not beneficial to adapting the model to complex working conditions. The high-correlation feature plays an important role in improving the modeling accuracy, and some processing can be performed after the feature is extracted to improve the correlation between the feature and the SOH. The existing machine learning models are various in types, different in principle and respectively advantageous; and the machine learning method is more dependent on the setting of model hyper-parameters, and improper parameters can cause insufficient generalization capability of the model. Therefore, the problem that the correlation between the extracted features and the target quantity is further improved by performing feature extraction based on the segment charging data and combining with feature processing, and meanwhile, the accuracy of SOH estimation is improved by performing parameter adjustment on a machine learning model by adopting a high-performance search algorithm is an urgent need to be solved.

Disclosure of Invention

In order to overcome the defects in the prior art, the invention provides a lithium ion battery SOH estimation method based on an SSA-SVR model, so that the characteristics highly related to SOH can be mined from a charging data segment, the characteristics with poor correlation are processed by adopting a Savitzky-Golay filtering algorithm, the correlation of the existing characteristics is further improved, and meanwhile, the hyper-parameters of the SVR are optimized by adopting a sparrow searching algorithm, so that the prediction accuracy of the SOH estimation model is improved.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows: a lithium ion battery SOH estimation method based on an SSA-SVR model comprises the following steps:

step 1, data acquisition: carrying out charging and discharging on the lithium ion battery for multiple times, and recording current, voltage and time data of the lithium ion battery in the charging and discharging process and the capacity of complete discharging each time;

step 2, health feature extraction: extracting four health characteristics from the voltage and current data segments in the charging process, namely a charged voltage value, a charged current value, the charging duration of the voltage data segment and the charging duration of the current data segment;

step 3, constructing a feature processing model based on Savitzky-Golay filtering: selecting a Savitzky-Golay filtering algorithm to process the extracted health characteristics, eliminating disorder fluctuation of the health characteristics caused by the influence of the charging data on the drift noise of the sensor, and constructing a characteristic processing model;

step 4, constructing a training set and a prediction set by taking a matrix F = [ F1, F2, F3, F4] formed by the extracted and processed health characteristics as input and the SOH of the battery as output;

and 5, constructing an SSA-SVR model: and constructing a reference SVR model, optimizing the hyper-parameters of the SVR model by adopting a sparrow search algorithm based on the training set, constructing an SOH estimation model, and outputting the estimated SOH.

Further, the step 2 comprises:

step 2.1, extracting a voltage data segment [ V ] in the charging process _a ,V _b ]Is the first healthy feature F1, the feature F1 is calculated using equation (1):

F1＝t _b -t _a (1)

in the formula, t _a For voltage rising to V _a Corresponding charging time, t _b For voltage rising to V _b Corresponding charging time.

Step 2.2, extracting a current data segment [ I ] in the charging process _c ,I _d ]Is the second healthy feature F2, the feature F2 is calculated using equation (2):

F2＝t _d -t _c (2)

in the formula, t _c For the current to drop to I _c Corresponding charging time, t _d For the current to drop to I _d Corresponding charging time.

Step 2.3, extracting the secondary voltage V in the constant current charging process _e As a starting point of the time counting, charging t _e The voltage value after the time is taken as a third healthy feature F3, and the feature F3 is calculated by using equation (3):

F3＝V _e +ΔV _e (3)

wherein, is Δ V _e Is t _e The voltage value that changes during the charging time.

Step 2.4, extracting the secondary current I in the constant voltage charging process _f As a starting point of time counting, charging t _f The current value after the time is taken as a fourth healthy feature F4, and the feature F4 is calculated by equation (4):

F4＝I _f +ΔI _f (4)

wherein, delta I _f Is the current value that varies over the tf charge time.

Further, the step 3 comprises:

step 3.1, calculating the correlation between the extracted 4 kinds of health characteristic quantities and the target quantity SOH by using the formula (5):

in the formula, X and Y are respectively a health characteristic and an SOH sample.

Step 3.2, constructing a feature processing model by adopting a Savitzky-Golay filter, wherein the feature processing model is determined by a polynomial order N and a window length L =2M +1, and M represents the absolute value of the maximum and minimum measurement point coordinates in a measurement window; for data for each measurement point of (-M, -M + 1.., 0,1.., M-1,M), processing is performed using equations (6), (7), and (8):

in which p (n) is a limiting polynomial, a _k Representing a plurality of itemsCoefficient of the formula epsilon _N Residual error for least squares fitting, x [ n ]]And (4) solving the equation set obtained in the step (8) to obtain the optimal polynomial coefficient.

Step 3.3, setting a threshold value to be 0.97, and when the Pearson correlation coefficient between the health characteristic quantity and the SOH is larger than the set threshold value, not activating the characteristic processing model, and directly estimating the SOH by using the original aging characteristic; and when the Pearson correlation coefficient between the health characteristic quantity and the SOH is smaller than a set threshold value, the health characteristic is greatly influenced by the drift noise of the sensor, and the health characteristic is processed to improve the correlation between the health characteristic and the target quantity.

Further, the step 5 comprises:

step 5.1, constructing a reference SVR model

S＝{x _i ,y _i |x _i ∈R ^m ,y _i ∈R}；i＝1,2,…,T (9)

In the formula, x _i Is the feature vector of the ith sample, y _i Is the corresponding regression value, T is the number of samples, and m is the dimension of the feature vector;

the SVR function is defined as:

f(x)＝ωΦ(x)+b (10)

in the formula, f (x) is output, Φ (x) is a nonlinear mapping function, ω, b are parameters to be determined, and the following objective function is minimized to solve ω and b:

wherein C is a penalty coefficient, f (x) _i ) Is the predicted value of the ith sample, and epsilon represents the maximum error allowed by regression and is defined as:

|y-f(x)| _ε = max {0, | y-f (x) | - ε } (12) introduces a relaxation variable ξ _i And

then, it is transformed into the following objective function:

and (3) constraint:

equation (10) is converted to solve the dual problem:

in the formula, beta _i And

for Lagrangian, K (x) _i ,x _j ) Selecting an RBF kernel function with strong linear approximation capability as a kernel function, and defining the RBF kernel function as follows:

where σ is the width of the kernel function;

and 5.2, optimizing the parameters of the SVR model by adopting a sparrow search algorithm, constructing an SOH estimation model, and outputting the estimated SOH.

Further, the step 5.2 comprises:

step 5.2.1, normalizing the sample data by using the formula (17), wherein the normalized data is divided into a training set and a testing set,

in the formula, Y and Y' are sample characteristic values before and after the normalization of the training data respectively, Y _min And Y _max Respectively representing the minimum value and the maximum value before normalization in the sample characteristic data;

step 5.2.2, setting parameters of a sparrow searching algorithm;

step 5.2.3, setting the value ranges of the penalty coefficient C and the kernel function parameter sigma, and initializing the population;

and 5.2.4, training an SVR (singular value regression) model by using the training set, and calculating the fitness value of each sparrow by using a formula (18):

in the formula (I), the compound is shown in the specification,

for the predicted value of the nth training sample, Y _n Is the true value of the nth training sample, N _tr The number of training samples;

step 5.2.5, calculating and updating the positions of the finder, the joiner and the alerter according to the formulas (19) - (21) respectively;

the finder location update formula is:

in the formula (I), the compound is shown in the specification,

representing the position of a sparrow monomer, i is a positive integer from 1 to P, j is a positive integer from 1 to d, P is the total number of the population, d is the dimension of characteristic data, alpha is a random number between 0 and 1, it _max For maximum number of iterations, R ₂ Is an early warning value, ST is a safety value, Q is a random number, and E is a 1 × d all-1 matrix; when R is ₂ ST ≧ represents that the risk of being prey is greater, and the sparrow at that position will be rapidly transferred; otherwise, the searching range can be expanded;

the update formula of the subscriber location is:

in the formula (I), the compound is shown in the specification,

and

representing the globally least favorable and most favorable positions, respectively; a. The ⁺ Is the generalized inverse of A, A representing a 1 × d matrix with elements randomly preset to 1 or-1; when i is more than n/2, the follower fails to grab food and needs to be transferred;

the updating formula of the position of the warner is as follows:

in the formula (I), the compound is shown in the specification,

representing the most favorable position of the authorities, wherein gamma is a standard normal random number and represents a step size control coefficient; k is a random number from-1 to 1, representing the direction of a sparrow; f. of _g Represents the maximum value of the fitness of the authority, f _w And f _g Conversely, ε is a small constant to ensure that the denominator is not 0; f. of _i Representing the fitness value of the ith sparrow;

step 5.2.6, obtaining the current updated position, and calculating to obtain the optimal individual and the optimal fitness value;

5.2.7, when the result after training reaches the set parameter, stopping calculation, and executing output parameter (C, σ), otherwise, returning to 5.2.4 to recalculate the fitness value;

and 5.2.8, establishing an estimation model according to the parameters (C, sigma) and outputting the estimated SOH.

Compared with the prior art, the invention has the beneficial effects that:

the method is used for extracting the characteristics based on the voltage and current data segments in the charging process and extracting four health characteristics with high correlation. The health characteristics have strong correlation, high extraction efficiency and proper quantity, and can be applied to the online high-precision estimation of the SOH of the lithium ion battery.

Aiming at the characteristic that a sensor is susceptible to noise in the data acquisition process, the invention adopts the Savitzky-Golay filter to process the features with relatively low correlation, eliminates the noise influence, further improves the correlation between the provided features and the target quantity, and improves the prediction accuracy and stability of the model for SOH estimation.

Aiming at the characteristic that a machine learning method is more dependent on the setting of model hyperparameters, and inappropriate parameters can cause insufficient generalization capability of the model, the support vector regression with less requirements on data volume and stronger nonlinear regression capability is adopted as a reference model, and the hyperparameters of the support vector regression are optimized through a sparrow search algorithm, so that the effectiveness of the SOH estimation model is ensured.

Drawings

FIG. 1 is a graph of the trend of the normalized four initial extracted features in an embodiment of the present invention;

FIG. 2 (a) is a graph comparing the trend of F2 characteristic changes before and after Savitzky-Golay filtering in the embodiment of the present invention;

FIG. 2 (b) is a graph comparing the trend of F4 characteristic changes before and after Savitzky-Golay filtering in the embodiment of the present invention;

FIG. 3 is a flow chart of modeling of an SSA-SVR model in accordance with an embodiment of the present invention;

FIG. 4 is a comparison of predicted results for different methods of B0005 battery implementation of the invention;

fig. 5 is a graph comparing predicted results of different methods for a B0007 battery according to an embodiment of the invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings in conjunction with the following detailed description. It should be understood that the description is intended to be exemplary only, and is not intended to limit the scope of the present invention. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present invention.

The embodiment is a lithium ion battery SOH estimation method based on an SSA-SVR model, which can realize the SOH on-line estimation with high feature extraction efficiency and high estimation precision as a target, mine information related to battery degradation in different data types of voltage and current by combining with the feature extraction of charging process segment data, optimize related features by combining with feature processing, enhance the correlation between the related features and a target quantity, optimize the SVR model by a sparrow search algorithm after obtaining input features, and improve the prediction precision of an overall framework. Specifically, the estimation method is performed as follows:

step one, data acquisition. And carrying out multiple charging and discharging on the lithium ion battery, and recording current, voltage and time data of the lithium ion battery in the charging and discharging process and the capacity of complete discharging in each time.

The lithium ion battery degradation data used in this example was from the NASA PCOE research center, which included charge and discharge test data for several commercial lithium ion 18650 batteries. The battery data set records detailed data of voltage, current and temperature during the battery cycle aging experiment. Batteries B05 and B07 were selected as experimental subjects and their detailed data were recorded at a temperature of 24 ℃ in three operating modes (charge, discharge and impedance test). The rated capacity of batteries B05 and B07 is 2Ah. The charging process was carried out at 1.5A in a Constant Current (CC) mode until the battery voltage reached 4.2V, and then charged in a Constant Voltage (CV) mode until the charging current dropped to 20mA. The discharge process is carried out in 2A Constant Current (CC) mode until the cell voltages B05, B07 drop to 2.7V, 2.2V, respectively.

Step two, feature extraction

The voltage and current curves change regularly as the number of cycles increases. As the battery ages, the duration within the same voltage interval becomes progressively shorter; the duration time in the same current interval is gradually prolonged; the higher the voltage and current values are after the same time as the charging is started from a certain voltage and current value. To this end, the present embodiment selects the following four health features: the charged voltage value, the charged current value, the charging duration of the voltage data segment and the charging duration of the current data segment are used for SOH estimation.

Step 2.1, extracting a voltage data segment [ V ] in the charging process _a ,V _b ]Is the first health characteristic F1. Feature F1 is calculated using equation (1):

F1＝t _b -t _a (1)

in the formula, t _a For voltage rising to V _a Corresponding charging time, t _b For voltage rising to V _b Corresponding charging time.

The optimal segment of F1 is selected by a traversal method, 3.6V is used as a lower voltage limit, 4.2V is used as an upper voltage limit, and 0.01V is used as a change interval to perform traversal, wherein V can be obtained by traversal _a ＝3.7V，V _b ＝4.2V。

Step 2.2, extracting a current data segment [ I ] in the charging process _c ,I _d ]Is the second health characteristic F2. Feature F2 is calculated using equation (2):

F2＝t _d -t _c (2)

in the formula, t _c For the current to drop to I _c Corresponding charging time, t _d For the current to drop to I _d Corresponding charging time.

The optimal segment of F2 is selected by a traversal method, traversal is carried out by taking 0.3A as a current lower limit, 1.5A as a current upper limit and 0.01A as a change interval, and the optimal segment can be obtained by traversal I _c ＝1.49A，I _d ＝0.42A。

Step 2.3, extracting the secondary voltage V in the Constant Current (CC) charging process _e As a starting point of time counting, charging t _e The voltage value after the time is taken as a third health feature F3. Feature F3 is calculated using equation (3):

F3＝V _e +ΔV _e (3)

wherein, is Δ V _e Is t _e The voltage value that changes during the charging time.

The optimal segment of F3 is selected by a traversal method, 3.6V is used as a lower voltage limit, 4.2V is used as an upper voltage limit, 0.01V is used as a voltage change interval, 50s is used as a time change interval for traversal, and V can be obtained by traversal _e ＝3.72V，t _e ＝400s。

The normalized variation trend graph of the four initial extracted features is shown in fig. 1.

Step 2.4, extracting the secondary current I in the Constant Voltage (CV) charging process _f As a starting point of the time counting, charging t _f The current value after the time is taken as the fourth health feature F4. Feature F4 is calculated using equation (4):

F4＝I _f +ΔI _f (4)

wherein, delta I _f Is t _f The current value that changes during the charging time.

The optimal segment of F4 is selected by a traversal method, 0.3A is used as a current lower limit, 1.5A is used as a current upper limit, 0.01A is used as a current change interval, 50s is used as a time change interval for traversal, and I can be obtained by traversal _f ＝1.41V，t _f ＝1800s。

Step three, feature processing

Step 3.1, calculating the correlation between the extracted 4 characteristic quantities and the target quantity SOH by using the formula (5):

in the formula, X and Y are respectively a health characteristic and an SOH sample. The larger the absolute value of the correlation coefficient, the stronger the correlation between the feature and the SOH, and the higher the accuracy in estimating the SOH using the feature.

Step 3.2, constructing a feature processing model by adopting a Savitzky-Golay filter, wherein the feature processing model is determined by a polynomial order N and a window length L =2M +1, and M represents the absolute value of the maximum and minimum measurement point coordinates in a measurement window; for data for each measurement point of (-M, -M + 1.., 0,1.., M-1,M), processing is performed using equations (6), (7), and (8):

in which p (n) is a limiting polynomial, a _k Coefficient, epsilon, representing a polynomial term _N Residual error for least squares fitting, x [ n ]]And (4) solving the equation set obtained in the step (8) to obtain the optimal polynomial coefficient.

Step 3.3, setting a threshold value to be 0.97, and when the Pearson correlation coefficient between the characteristic quantity and the SOH is larger than the set threshold value, not activating the characteristic processing model, and directly estimating the SOH by using the original aging characteristic; and when the Pearson correlation coefficient between the characteristic quantity and the SOH is smaller than a set threshold value, the characteristic is greatly influenced by the drift noise of the sensor, and the characteristic is processed to improve the correlation between the characteristic and the target quantity.

The features F2 and F4 extracted from the charging current data have a lower correlation with the features extracted from the voltage data, and the variation tendency of the features fluctuates greatly. This is because the charging current data is susceptible to current sensor drift noise, and such fluctuations are unavoidable. Therefore, the invention adopts Savitzky-Golay (SG) filtering algorithm to remove the noise with the characteristics of F2 and F4. To visually demonstrate the effect of feature processing, a comparison graph of feature variation trend before and after Savitzky-Golay filtering is shown in fig. 2 (a) and 2 (b).

And step four, establishing a training set and a prediction set by taking the four health characteristics obtained in the step as input and the SOH of the battery as output.

And step five, constructing an SSA-SVR model.

Step 5.1, constructing a reference SVR model

S＝{x _i ,y _i |x _i ∈R ⁿ ,y _i ∈R}；(i＝1,2,…T) (9)

In the formula, x _i Is the feature vector of the i-th sample, y _i And T is the number of samples and n is the dimension of the feature vector for the corresponding regression value.

The SVR function is defined as:

f(x)＝ωΦ(x)+b (10)

in the formula, f (x) is output, phi (x) is a nonlinear mapping function, and omega and b are parameters to be determined. The following objective function is minimized to solve ω and b:

wherein C is a penalty coefficient, f (x) _i ) Is the predicted value of the ith sample, and epsilon represents the maximum error allowed by regression and is defined as:

|y-f(x)| _ε ＝max{0,|y-f(x)|-ε} (12)

introducing a relaxation variable xi _i And

then, it can be modeled as the following objective function:

and (3) constraint:

equation (10) is converted to solve the dual problem:

in the formula, beta _i And

for Lagrangian, K (x) _i ,x _j ) As a kernel function, the RBF kernel function with strong linear approximation capability is selected and defined as follows:

where σ is the width of the kernel function.

And 5.2, optimizing the parameters of the SVR model by adopting a sparrow search algorithm.

The penalty factor C and the kernel function parameter sigma are key parameters of the SVR model, and determine the estimation accuracy and the fitting capability of the estimation model. SSA is a novel natural heuristic algorithm proposed according to behaviors of sparrows for foraging and avoiding predators. Sparrows in the SSA can be converged to the current optimal solution by jumping to the vicinity of the current optimal solution, and the performance in the aspects of precision, convergence speed and the like is excellent. The estimation performance of the model can be effectively improved by optimizing the hyper-parameters by adopting a sparrow search algorithm. The steps of the SSA-SVR are briefly described as follows:

and 5.2.1, normalizing the sample data by using the formula (17). The normalized data can be divided into a training set and a test set.

In the formula, Y and Y' are sample characteristic values before and after the training data normalization, respectively, Y _min And Y _max Respectively representing the minimum value and the maximum value before normalization in the sample characteristic data.

Step 5.2.2, setting parameters of a sparrow search algorithm, wherein the maximum iteration number N =150, the population size N =50, the number of discoverers PD =0.6, the number of cautioners SD =0.3, the safety value ST =0.5, and the upper and lower limits DL of an independent variable = [ -7,7];

step 5.2.3, setting the value ranges of the penalty coefficient C and the kernel function parameter sigma, and initializing a population;

and 5.2.4, training an SVR model by using the training set, and calculating the fitness value of each sparrow by using a formula (18):

in the formula (I), the compound is shown in the specification,

for the predicted value of the nth training sample, Y _n Is the true value, N, of the nth training sample _tr The number of training samples.

And 5.2.5, calculating and updating the positions of the finder, the joiner and the alertness respectively according to the formulas (19) to (21).

The finder location update formula is:

in the formula (I), the compound is shown in the specification,

representing a sparrow unit position, i is a positive integer from 1 to P, j is a positive integer from 1 to d, P is the total number of the population, d is a characteristic data dimension, alpha is a random number between 0 and 1, it _max For maximum number of iterations, R ₂ Is an early warning value, ST is a safety value, Q is a random number, and E is a 1 × d all-1 matrix; when R is ₂ ST ≧ represents that the risk of being prey is greater, and the sparrow at that position will be rapidly transferred; otherwise the search range may be expanded.

When the finder finds better food, the joiner can rob the food. If successful, the location of the finder is changed, and the location of the negative subscriber is changed.

The update formula of the subscriber location is:

in the formula (I), the compound is shown in the specification,

and

representing the global least favorable and most favorable bits, respectively，A ⁺ Is the generalized inverse of A, A representing a 1 × d matrix with elements randomly preset to 1 or-1; when i is more than n/2, it means that the follower fails to grab food and needs to transfer.

The updating formula of the position of the warner is as follows:

in the formula (I), the compound is shown in the specification,

indicating the most favorable position of the authorities; gamma is a standard normal random number and represents a step control coefficient; k is a random number from-1 to 1, representing the direction of a sparrow; f. of _g Represents the maximum value of the fitness of the authority, f _w And f _g Conversely, ε is a constant to ensure that the denominator is not 0; f. of _i The fitness value of the ith sparrow is shown.

And 5.2.6, obtaining the current updated position, and calculating to obtain the optimal individual and the optimal fitness value.

And 5.2.7, when the trained result reaches the set parameter, stopping calculation, and executing the output parameter (C, σ), otherwise, returning to step 5.2.4 to recalculate the fitness value.

And 5.2.8, establishing an estimation model according to the parameters (C, sigma) and outputting the estimated SOH.

A modeling flow chart of the SSA-SVR model is shown in FIG. 3.

In order to verify the superiority of the method provided by the invention, the following three methods are selected to be combined with the actual SOH for comparison, wherein the method I comprises the following steps: the method provided by the invention comprises SG filtering optimization characteristics and adopts SSA to optimize the hyperparameter of the SVR. The second method comprises the following steps: SG filtering is included but SSA is not used to optimize the hyperparameters of SVR. The third method comprises the following steps: SSA was used to optimize the hyper-parameters, but SG filtering was not included. The second method is used for verifying the superiority of optimizing the hyper-parameters by utilizing SSA, and the third method is used for verifying the effectiveness of the SG filtering optimization characteristic method. The invention utilizes two batteries B0005 and B0007 in NASA data set to carry out experiments. With 60% of the full life cycle of the battery as the training set and the rest for testing, the predicted starting point for the B0005, B0007 batteries is 101. Fig. 4 and 5 are graphs comparing the predicted results of different methods for B0005 and B0007 batteries.

In conclusion, the health state of the lithium ion battery is estimated based on the SSA-SVR model, 4 characteristics extracted from the charging voltage and current curve segments have extremely high correlation with the SOH, and the basis of high-precision prediction of the model is laid. SG filtering is used for optimization for two features with noise influence extracted from the charging current segment. Experimental results show that the method can reduce the model error by more than 50%. And the SSA is adopted to carry out global optimization on the parameters of the SVR model, so that the error can be reduced by over 50 percent, and the precision and the generalization capability of the model are effectively improved.

It is to be understood that the above-described embodiments of the present invention are merely illustrative of or explaining the principles of the invention and are not to be construed as limiting the invention. Therefore, any modification, equivalent replacement, improvement and the like made without departing from the spirit and scope of the present invention should be included in the protection scope of the present invention. Further, it is intended that the appended claims cover all such variations and modifications as fall within the scope and boundaries of the appended claims or the equivalents of such scope and boundaries.

Claims (10)

1. A lithium ion battery SOH estimation method based on an SSA-SVR model is characterized by comprising the following steps:

step 1, data acquisition: carrying out charging and discharging on the lithium ion battery for multiple times, and recording current, voltage and time data of the lithium ion battery in the charging and discharging process and the capacity of complete discharging in each time;

step 2, health feature extraction: extracting four health characteristics from the voltage and current data segments in the charging process, namely a charged voltage value, a charged current value, the charging duration of the voltage data segment and the charging duration of the current data segment;

step 3, constructing a feature processing model based on Savitzky-Golay filtering: selecting a Savitzky-Golay filtering algorithm to process the extracted health characteristics, eliminating messy fluctuation of the health characteristics caused by the influence of the charging data on the drift noise of the sensor, and constructing a characteristic processing model;

step 4, constructing a training set and a prediction set by taking a matrix F = [ F1, F2, F3, F4] formed by the extracted and processed health characteristics as input and the SOH of the battery as output;

and 5, constructing an SSA-SVR model: and constructing a reference SVR model, optimizing the hyper-parameters of the SVR model by adopting a sparrow search algorithm based on the training set, constructing an SOH estimation model, and outputting the estimated SOH.

2. The method for estimating the SOH of the lithium ion battery based on the SSA-SVR model according to claim 1, wherein the step 2 comprises:

step 2.1, extracting a voltage data segment [ V ] in the charging process _a ,V _b ]Is the first healthy feature F1, the feature F1 is calculated using equation (1):

F1＝t _b -t _a (1)

in the formula, t _a For voltage rising to V _a Corresponding charging time, t _b For voltage rising to V _b Corresponding charging time.

3. The method for estimating the SOH of the lithium ion battery based on the SSA-SVR model according to claim 2, wherein the step 2 further comprises:

step 2.2, extracting a current data segment [ I ] in the charging process _c ,I _d ]Is the second healthy feature F2, the feature F2 is calculated using equation (2):

F2＝t _d -t _c (2)

in the formula, t _c For the current to drop to I _c Corresponding charging time, t _d For the current to drop to I _d Corresponding charging time.

4. The method for estimating the SOH of the lithium ion battery based on the SSA-SVR model according to claim 3, wherein said step 2 further comprises:

step 2.3, extracting the secondary voltage V in the constant current charging process _e As a starting point of the time counting, charging t _e The voltage value after the time is taken as a third healthy feature F3, and the feature F3 is calculated by using equation (3):

F3＝V _e +ΔV _e (3)

wherein, is Δ V _e Is t _e The voltage value that changes during the charging time.

5. The method for estimating SOH of a lithium ion battery based on SSA-SVR model according to claim 4, wherein the step 2 further comprises:

step 2.4, extracting the secondary current I in the constant voltage charging process _f As a starting point of time counting, charging t _f The current value after the time is taken as a fourth healthy feature F4, and the feature F4 is calculated by equation (4):

F4＝I _f +ΔI _f (4)

wherein, delta I _f Is t _f The current value that varies during the charging time.

6. The method for estimating the SOH of the lithium ion battery based on the SSA-SVR model according to claim 1, wherein said step 3 comprises:

step 3.1, calculating the correlation between the extracted 4 health characteristic quantities and the target quantity SOH by using the formula (5):

in the formula, X and Y are respectively a health characteristic and an SOH sample.

7. The method for estimating SOH of a lithium ion battery based on SSA-SVR model according to claim 6, wherein the step 3 further comprises:

step 3.2, constructing a feature processing model by adopting a Savitzky-Golay filter, wherein the feature processing model is determined by a polynomial order N and a window length L =2M +1, and M represents the absolute value of the maximum and minimum measurement point coordinates in a measurement window; for data for each measurement point of (-M, -M + 1.., 0,1.., M-1,M), processing is performed using equations (6), (7), and (8):

in which p (n) is a limiting polynomial, a _k Coefficient, epsilon, representing a polynomial term _N For the residual of the least squares fit, x [ n ]]And (4) solving the equation set obtained in the step (8) to obtain the optimal polynomial coefficient.

8. The method for estimating SOH of a lithium ion battery based on SSA-SVR model according to claim 7, wherein the step 3 further comprises:

step 3.3, setting a threshold value to be 0.97, and when the Pearson correlation coefficient between the health characteristic quantity and the SOH is larger than the set threshold value, not activating the characteristic processing model, and directly estimating the SOH by using the original aging characteristic; and when the Pearson correlation coefficient between the health characteristic quantity and the SOH is smaller than a set threshold value, the health characteristic is greatly influenced by the drift noise of the sensor, and the health characteristic is processed to improve the correlation between the health characteristic and the target quantity.

9. The method for estimating SOH of a lithium ion battery based on SSA-SVR model according to claim 1, wherein the step 5 comprises:

step 5.1, constructing a reference SVR model

S＝{x _i ,y _i |x _i ∈R ^m ,y _i ∈R}；i＝1,2,…,T (9)

In the formula, x _i Is the feature vector of the ith sample, y _i Is the corresponding regression value, T is the number of samples, and m is the dimension of the feature vector;

the SVR function is defined as:

f(x)＝ωΦ(x)+b (10)

in the formula, f (x) is output, Φ (x) is a nonlinear mapping function, ω, b are parameters to be determined, and the following objective function is minimized to solve ω and b:

wherein C is a penalty coefficient, f (x) _i ) Is the predicted value of the ith sample, and epsilon represents the maximum error allowed by regression and is defined as:

|y-f(x)| _ε ＝max{0,|y-f(x)|-ε} (12)

introducing a relaxation variable xi _i And

then, it is transformed into the following objective function:

and (3) constraint:

equation (10) is converted to solve the dual problem:

in the formula, beta _i And

for Lagrangian, K (x) _i ,x _j ) Selecting an RBF kernel function with strong linear approximation capability as a kernel function, and defining the RBF kernel function as follows:

where σ is the width of the kernel function;

and 5.2, optimizing the parameters of the SVR model by adopting a sparrow search algorithm, constructing an SOH estimation model, and outputting the estimated SOH.

10. The method for estimating SOH of a lithium ion battery based on SSA-SVR model according to claim 9, wherein the step 5.2 comprises:

step 5.2.1, normalizing the sample data by using the formula (17), wherein the normalized data is divided into a training set and a testing set,

in the formula, Y and Y' are sample characteristic values before and after the normalization of the training data respectively, Y _min And Y _max Respectively representing the minimum value and the maximum value before normalization in the sample characteristic data;

step 5.2.2, setting parameters of a sparrow searching algorithm;

step 5.2.3, setting the value ranges of the penalty coefficient C and the kernel function parameter sigma, and initializing a population;

and 5.2.4, training an SVR model by using the training set, and calculating the fitness value of each sparrow by using a formula (18):

in the formula (I), the compound is shown in the specification,

is the predicted value of the nth training sample, Y _n Is the true value of the nth training sample, N _tr The number of training samples;

step 5.2.5, calculating and respectively updating the positions of the finder, the joiner and the alertness according to the formulas (19) - (21);

the finder location update formula is:

in the formula (I), the compound is shown in the specification,

representing the position of a sparrow monomer, i is a positive integer from 1 to P, j is a positive integer from 1 to d, P is the total number of the population, d is the dimension of characteristic data, alpha is a random number between 0 and 1, it _max For maximum number of iterations, R ₂ Is an early warning value, ST is a safety value, Q is a random number, and E is a 1 × d all-1 matrix; when R is ₂ ST ≧ represents that the risk of being prey is greater, and the sparrow at that position will be rapidly transferred; otherwise, the searching range can be expanded;

the update formula of the subscriber location is:

in the formula (I), the compound is shown in the specification,

and

representing the global least favorable and most favorable bits, A, respectively ⁺ Is a generalized inverse of A, A representing a 1 × d matrix, where the elements are randomly preset to 1 or-1; when i is more than n/2, the follow-up person fails to snatch food and needs to transfer;

the updated formula for the position of the alert is:

in the formula (I), the compound is shown in the specification,

indicating the most favorable position of the authorities; gamma is a standard normal random number and represents a step length control coefficient; k is a random number from-1 to 1, representing the orientation of a sparrow; f. of _g Represents the maximum value of the fitness of the authority, f _w And f _g Conversely, ε is a constant to ensure that the denominator is not 0; f. of _i Representing the fitness value of the ith sparrow;

step 5.2.6, obtaining the current updated position, and calculating to obtain the optimal individual and the optimal fitness value;

5.2.7, when the result after training reaches the set parameter, stopping calculation, and executing output parameter (C, σ), otherwise, returning to 5.2.4 to recalculate the fitness value;

and 5.2.8, establishing an estimation model according to the parameters (C, sigma) and outputting the estimated SOH.

Images