CN113125987A - Novel hybrid lithium ion battery health state prediction method - Google Patents

Abstract

The invention relates to a novel hybrid lithium ion battery health state prediction method, which comprises the steps of firstly collecting battery capacity data, analyzing the capacity data, then carrying out correlation analysis on an extracted charging and discharging sequence and a capacity data sequence to construct a health index, and then decomposing the health index sequence into a trend factor and a random factor by utilizing ensemble empirical mode decomposition (CEEMDAN) of adaptive noise. Wherein, the relatively smooth trend data sequence is predicted by adopting an autoregressive integrated moving average model (ARIMA); combining the residual error of ARIMA prediction and the non-trend item obtained by CEEMDAN decomposition into a new non-trend item; and then introducing a Least Square Support Vector Machine (LSSVM) to establish a nonlinear prediction model and predict. And finally, combining the prediction results of the trend item data sequence and the non-trend item data sequence to provide reference for the evaluation of the health state and the residual service life. The invention effectively improves the aspect of predicting the residual life of the lithium ion battery and greatly improves the prediction precision.

Description

Novel hybrid lithium ion battery health state prediction method

Technical Field

The invention relates to the technical field of lithium ion batteries, in particular to a novel hybrid lithium ion battery health state prediction method.

Background

The lithium ion battery has the outstanding characteristics of high capacity, high reliability, high safety and the like, and is widely applied to the intelligent manufacturing fields of computing engineering, logistics, aerospace and the like. However, degradation of the lithium battery may significantly lead to a reduction in the performance of the electrical device, thereby increasing the cost of accidental maintenance. As the discharge time increases, the battery failure due to the deterioration shortens the service life of the secondary battery, and even causes a serious accident. Therefore, the remaining service life (RUL) of the battery, i.e., the end time of the service life, needs to be accurately predicted. The method is also significant for avoiding and preventing major accidents.

The RUL of a battery is defined as the number of charge and discharge cycles remaining before its operating state deteriorates to a fault threshold. The state of degradation of the battery may be characterized by Health Indicators (HI), such as current, voltage, impedance, and capacity. Capacity is the most widespread indicator of battery health, and it is generally accepted that a battery reaches its end-of-life (EOL) threshold when its capacity degrades to 70% of its rated capacity. Currently, the RUL prediction methods are mainly classified into three major categories, namely, model-based methods, data-driven methods, and fusion methods. The method based on the model can well describe the physical and chemical change process in the battery, but is easily influenced by load and environment, and the modeling is complex. The data driving method is low in complexity, does not need to consider the electrochemical reaction inside the battery, and depends on a large amount of battery degradation data. The prediction method based on data driving and the prediction method based on the model have respective limitations when applied to the prediction of the lithium ion battery, so that the fusion prediction becomes a research hotspot for improving the prediction performance of the RUL.

The conventional CEEMDAN-ARIMA-LSSVM can obtain a relatively accurate long-term prediction result of the battery capacity as a data-driven prediction method, and the prediction effect when the battery monitoring data change is large can be effectively improved, such as sudden data drop or sudden data rise caused by energy regeneration.

Disclosure of Invention

Aiming at the problems that the prediction result in the prior art depends on an empirical degradation model excessively and the adaptability to different data is poor, the invention provides a lithium ion battery residual service life prediction method based on the integration of CEEMDAN-ARIMA and least squares support vector machine LSSVM algorithm.

In order to achieve the purpose, the invention adopts the following technical scheme:

a novel hybrid lithium ion battery health state prediction method comprises the following steps:

step S1: acquiring battery capacity data, analyzing the capacity data, and performing correlation analysis on the extracted charging and discharging sequence and the capacity data sequence to construct a health index; dividing the health index data into 2 groups, wherein one group is a training data set, the other group is a testing data set, and a battery capacity failure threshold CEOL is set;

step S2: performing signal decomposition on the training data set by adopting a CEEMDAN algorithm to obtain two data sequences of a trend item and a non-trend item;

step S3: predicting a relatively smooth trend item data sequence in the decomposition result obtained in the step 2 by using ARIMA;

step S4: combining the residual error of ARIMA prediction in the step 3 and the non-trend term obtained by CEEMDAN decomposition in the step 2 into a new non-trend term, introducing a Least Square Support Vector Machine (LSSVM) to establish a non-linear prediction model and estimating the system state;

step S5: and judging whether the predicted capacity value reaches a capacity failure threshold value, if so, calculating the prediction result of the RUL by combining the step 4 in the step 3, and if not, returning to the step 3.

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

the method comprises the steps of firstly carrying out multi-scale decomposition on a charging/discharging sequence and capacity by using a self-adaptive noise-based fully integrated empirical mode decomposition (CEEMDAN), then using a decomposed trend factor as an input of ARIMA, training a model, then combining a residual error of the ARIMA with a random factor of the CEEMDAN, and introducing a Least Square Support Vector Machine (LSSVM) for prediction. The invention can effectively predict the SOH and RUL of the battery, has better prediction efficiency and prediction precision, effectively judges the future working capacity, finds problems in time and avoids unnecessary troubles and loss.

Drawings

FIG. 1 is a schematic view of the overall process of the present invention

FIG. 2 is a graph showing the relationship between the capacity deterioration of a lithium ion No. 5 battery

FIG. 3 is a diagram of the prediction result of the remaining service life of a lithium ion battery

Detailed Description

The following takes the operational data of lithium ion battery No. B0005 in the NASA public data set shown in table 1 as an example, and combines with the drawings in the embodiment of the present invention, to clearly and completely describe the technical solution in the embodiment of the present invention.

TABLE 1B 0005 lithium ion batteries

Reference will now be made in detail to the embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the like or similar elements throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.

As shown in fig. 1, according to an embodiment of the present invention, a method for predicting remaining service life of a lithium ion battery based on CEEMDAN-ARIMA and LSSVM includes the following steps:

step S1: acquiring battery capacity data, analyzing the capacity data, and performing correlation analysis on the extracted charging and discharging sequence and the capacity data sequence to construct a health index; dividing the health index data into 2 groups, one group is a training data set, the other group is a testing data set, and setting a battery capacity failure threshold value C_EOL；

The specific implementation comprises the following substeps:

step S1.1: setting the charging voltage range of the battery to be 3.7-4.2 v, and setting the discharging voltage range to be 4-3 v;

step S1.2: because the closer the voltage is to the area near the charging and discharging voltage platform in each period, the higher the correlation coefficient between the time required by the voltage rising/falling and the battery capacity is, the charging and discharging data sequence is constructed based on the method and is compared with the capacity data sequence;

step S1.3: analyzing the correlation between the health index data sequence and the ability degradation data sequence by using a Pearson linear correlation coefficient, a Spearman rank correlation coefficient and a Kendall correlation coefficient, wherein the following table 2 is a relation between parameters and correlation coefficients;

step S1.4: a method for assessing the state of health of a lithium ion battery is determined.

TABLE 2 relationship of parameters to correlation coefficients

Step S2: performing signal decomposition on the training data set by adopting a CEEMDAN algorithm to obtain two data sequences of a trend item and a non-trend item;

step S2.1: the training data set is used as an original signal x (n), and then the original data is added with white noise, namely x (n) + epsilon₀ωⁱ(n) into a set of modal components IMF and residual signal, and finally IMF and original signal x (n) can be expressed as:

wherein ω isⁱ(n) is white Gaussian noise added at the i-th time, R (n) represents a residual component, E_kRefers to the k-th order mode produced by EMD;

step S2.2: after CEEMDAN decomposition is carried out on the signals, an endpoint mirror image method is adopted to improve the endpoint effect;

step S2.3: determining a trend item and a non-trend item by using a reverse combination method, wherein the specific process comprises the following steps:

(1) assuming that the trend term Res,

(2) determining whether a trend term is appropriate (e.g., mean calculation, correlation analysis, etc.);

(3) if the trend term and the non-trend term are not appropriate, the trend term Res + IMF is assumed₉，

(4) Continue until both terms are appropriate.

Step S3: predicting a relatively smooth trend item data sequence in the decomposition result obtained in the step 2 by using ARIMA;

firstly, the non-stationary time sequence stated in the S2 is converted into a stationary time sequence by adopting a d-order difference, and then an ARIMA (p, q) model is established, and the d-order difference of y (t) is expressed as

The ARIMA (p, d, q) model can therefore be described as:

w_t＝Φ₁w_t-1+Φ_2wt-2+...+Φ_pw_t-p+ε_t-θ₁ε_t-1-θ₂ε_t-2-...-θ_pε_t-p (6)

wherein, w_tIs a time sequence of ∈_tIs an allZero value variance σ²P is the order of the AR model, q is the order of the MA model, phi_iAnd theta_iThe parameters of the AR model and the MA model respectively,

d is the order of the difference.

Step S4: combining the residual error of ARIMA prediction in the step 3 and the non-trend term obtained by CEEMDAN decomposition in the step 2 into a new non-trend term, introducing a Least Square Support Vector Machine (LSSVM) to establish a non-linear prediction model and estimating the system state; the method specifically comprises the following substeps:

(1) the input data sequence is set as follows:

{(x₁,y₁)(x₂,y₂)...(x_l,y_l)} (7)

(2) if x is a non-linear mapping, w is a weight coefficient of the feature space, and b is an offset, then a linear equation for a high dimensional feature space can be used

To fit the training data, the LSSVM regression according to the principle of structure risk minimization can be expressed as:

in the formula: omega is a weight vector; b is a bias vector; e is a relaxation variable; λ is a penalty factor;

(3) solving the optimization problem by Lagrange method, wherein

The Lagrange multiplier is solved by a KKT condition, and according to a Mercer condition, the expression is as follows:

k(x_i,x_j)＝φ(x_i)^Tφ(x_i) (9)

(4) the final decision function of the LSSVM is:

because the selection of the parameters of the LSSVM can not easily influence the recognition rate and the generalization capability of the model, the invention adopts a cross validation method, namely, data is divided into a training set and a validation set, the model is trained by the training set, the model is tested and evaluated by the validation set, and then the model parameters are solved when the performance index of the model is the best.

Step S5: and judging whether the predicted capacity value reaches a capacity failure threshold value, if so, calculating the prediction result of the RUL by combining the step 4 in the step 3, and if not, returning to the step 3.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims (5)

1. A novel hybrid lithium ion battery health status prediction method is characterized in that the method is realized by the following steps:

step S1: collecting battery capacity data, analyzing the capacity data,performing correlation analysis on the extracted charging and discharging sequence and the capacity data sequence to construct a health index; dividing the health index data into 2 groups, one group is a training data set, the other group is a testing data set, and setting a battery capacity failure threshold value C_EOL；

Step S2: performing signal decomposition on the training data set by adopting a CEEMDAN algorithm to obtain two data sequences of a trend item and a non-trend item;

step S3: predicting a relatively smooth trend item data sequence in the decomposition result obtained in the step 2 by using ARIMA;

step S4: combining the residual error of ARIMA prediction in the step 3 and the non-trend term obtained by CEEMDAN decomposition in the step 2 into a new non-trend term, introducing a Least Square Support Vector Machine (LSSVM) to establish a non-linear prediction model and estimating the system state;

step S5: and judging whether the predicted capacity value reaches a capacity failure threshold value, if so, calculating the prediction result of the RUL by combining the step 4 in the step 3, and if not, returning to the step 3.

2. The method according to claim 1, wherein the step S1 specifically comprises:

step S1.1: setting the charging voltage range of the battery to be 3.7-4.0 v, and setting the discharging voltage range to be 4-3 v;

step S1.2: because the closer the voltage is to the area near the charging and discharging voltage platform in each period, the higher the correlation coefficient between the time required by the voltage rising/falling and the battery capacity is, the charging and discharging data sequence is constructed based on the method and is compared with the capacity data sequence;

step S1.3: and analyzing the correlation between the health index data sequence and the ability degradation data sequence by using a Pearson linear correlation coefficient, a Spearman rank correlation coefficient and a Kendall correlation coefficient:

step S1.4: a method for assessing the state of health of a lithium ion battery is determined.

3. The method according to claim 1, wherein the step S2 specifically comprises:

step S2.1: the training data set is used as an original signal x (n), and then the original data is added with white noise, namely x (n) + epsilon₀ωⁱ(n) into a set of modal components IMF and residual signal, and finally IMF and original signal x (n) can be expressed as:

wherein ω isⁱ(n) is white Gaussian noise added at the i-th time, R (n) represents a residual component, E_kRefers to the k-th order mode produced by EMD;

step S2.2: after CEEMDAN decomposition is carried out on the signals, an endpoint mirror image method is adopted to improve the endpoint effect;

step S2.3: and determining a trend item and a non-trend item by using a reverse combination method.

4. The method according to claim 1, wherein the ARIMA prediction in step S3 is specifically:

converting the non-stationary time series into stationary time series by using d-order difference in S2, and then establishing ARIMA (p, q) model, wherein the d-order difference of y (t) is expressed as

The ARIMA (p, d, q) model can therefore be described as:

w_t＝Φ₁w_t-1+Φ_2wt-2+...+Φ_pw_t-p+ε_t-θ₁ε_t-1-θ₂ε_t-2-...-θ_pε_t-p (6)

wherein, w_tIs a time sequence of ∈_tIs a mean value of zero and a variance of σ²P is the order of the AR model, q is the order of the MA model, phi_iAnd theta_iThe parameters of the AR model and the MA model respectively,

d is the order of the difference.

5. The method according to claim 1, wherein the LSSVM in step S4 is implemented by:

(1) the input data sequence is set as follows:

{(x₁,y₁)(x₂,y₂)...(x_l,y_l)} (7)

(2) if x is a non-linear mapping, w is a weight coefficient of the feature space, and b is an offset, then a linear equation for a high dimensional feature space can be used

To fit the training data, the LSSVM regression according to the principle of structure risk minimization can be expressed as:

in the formula: omega is a weight vector; b is a bias vector; e is a relaxation variable; λ is a penalty factor;

(3) solving the optimization problem by Lagrange method, wherein

The Lagrange multiplier is solved by a KKT condition, and according to a Mercer condition, the expression is as follows:

k(x_i,x_j)＝φ(x_i)^Tφ(x_i) (9)

(4) the final decision function of the LSSVM is:

because the selection of the parameters of the LSSVM can not easily influence the recognition rate and the generalization capability of the model, the invention adopts a cross validation method, namely, data is divided into a training set and a validation set, the model is trained by the training set, the model is tested and evaluated by the validation set, and then the model parameters are solved when the performance index of the model is the best.

Images