1. A lithium battery health state online estimation method based on feature analysis is characterized by comprising the following steps:

selecting charging and discharging characteristics by utilizing a principal component analysis strategy;

decomposing the charge and discharge characteristics into a main trend part and a secondary fluctuation part based on a complete empirical mode decomposition strategy of adaptive noise;

predicting the main trend part based on a logistic regression strategy of a sliding time window, and predicting the secondary fluctuation part by adopting a Kalman filtering strategy;

and combining the predicted main trend part with the secondary fluctuation part to obtain predicted characteristic data, and substituting the predicted characteristic data into the radial basis function neural network to realize the online prediction of the health state of the lithium battery.

2. The feature analysis-based lithium battery state of health online estimation method of claim 1, characterized in that: the process of selecting the charge-discharge characteristics by utilizing the principal component analysis strategy comprises the following steps,

converting the original random vector of the charge and discharge characteristics into a new random vector with irrelevant components by utilizing orthogonal transformation;

pointing the new random vector to a plurality of orthogonal directions with most spread sample points, and performing dimensionality reduction processing on a multi-dimensional variable system to obtain a low-dimensional variable system;

and converting the original random vector into a group of representations which are linearly independent of each dimension, and screening out main characteristic components of the charge and discharge characteristic data.

3. The feature analysis-based lithium battery state of health online estimation method of claim 2, characterized in that: the empirical mode decomposition strategy may include,

decomposing the signal into k eigenmode functions, each eigenmode function being represented by an IMF_kRepresents;

for each IMF_kThe jth IMF component of which is decomposed by empirical mode is E_j() It is shown that the original signal is defined as s (n), and white gaussian noise ω (n) is added to s (n).

4. The feature analysis-based lithium battery state of health online estimation method of claim 3, characterized in that: the empirical mode decomposition strategy may further comprise,

computing IMF₁：

Wherein I represents the signal s (n) +. epsilon₀ω_i(n) is decomposed I times, the parameter epsilon represents the signal-to-noise ratio of the control additive noise to the original signal;

and residual calculation: when k is 1, the residual error is calculated as follows:

r₁(n)＝s(n)-IMF₁(n)

before the next decomposition, the empirical mode decomposition is carried out on the Gaussian white noise to obtain a value of a first component,

adding the value of the first component to the residual signal, and updating the signal to be decomposed into r₁(n)+ε₁E₁(ω_i(n)) (I ═ 1,2, …, I), and the residual calculation was performed again to obtain IMF₂；

The IMF₂The calculation method of (a) is as follows:

when K is 2,3, …, K, the K-th residual is:

r_k(n)＝r_k-1(n)-IMF_k(n)

to r_k(n)+ε_kE_k(ω_i(n)) (I ═ 1,2, …, I) are decomposed until the first empirical mode component, IMF, is obtained_k+1The calculation method of (a) is as follows:

repeatedly calculating the k-th residual sum IMF_k+1Until the residual signal cannot be resolvable, the original signal s (n) is expressed as a combination of k IMFs and one residual r (n), as follows:

5. the feature analysis-based lithium battery state of health online estimation method of claim 4, characterized in that: the logistic function prob (event) of the logistic regression strategy for the sliding time window includes,

wherein, x (x)₁,x₂,…,x_d) Representing the input vectors corresponding to the d arguments, g (x) representing the logical model.

6. The feature analysis-based lithium battery state of health online estimation method of claim 5, characterized in that: defining the logical model g (x) comprises,

wherein g (x) represents all input vectors x (x)₁,x₂,…,x_d) α, β represent polynomial coefficients of g (x).

7. The feature analysis-based lithium battery state of health online estimation method of claim 6, characterized in that: the kalman filtering strategy includes a time update and a measurement update.

8. The feature analysis-based lithium battery state of health online estimation method of claim 7, characterized in that: the time update includes the time of day that the user is,

discrete kalman filter time update equation:

wherein the content of the first and second substances,representing the a priori state estimate at time k, a representing the state transition matrix,representing the estimate of the posterior state at time k-1, B representing the matrix that transforms the input into state, u_k-1Representing the input at time k-1,representing the prior estimated covariance, P, of time k_k-1Represents the a posteriori estimated covariance at time k-1 and Q represents the covariance of the system process.

9. The feature analysis-based lithium battery state of health online estimation method of claim 7, characterized in that: the measurement update includes a list of measurements of the measurement,

discrete kalman filter measurement update equation:

wherein, K_kRepresenting the Kalman gain, H representing the state variable to measurement transformation matrix, R representing the measurement noise covariance, z_kRepresenting the observed value.

10. The feature analysis-based lithium battery state of health online estimation method of claim 7, characterized in that: the radial basis function neural network includes a plurality of radial basis function neural networks,

activation function F_activationExpressed as:

wherein | x_p-c_iII denotes the Euclidean norm, c denotes the Gaussian functionThe center of the number, σ, represents the variance of the gaussian function;

according to the structure of the radial basis function neural network, the output of the network is obtained as follows:

wherein the content of the first and second substances,denotes the pth input sample, P1, 2,3, …, P denotes the total number of samples, c_iRepresenting the centres of nodes of the hidden layers of the network, w_nDenotes the connection weight from hidden layer to output layer, i ═ 1,2,3, …, h, i denotes the number of hidden layer nodes, y denotes the number of hidden layer nodes_jRepresenting the actual output of the jth output node of the network corresponding to the input sample;

defining d to be the expected output value of the sample, the variance σ of the radial basis function is then expressed as:

Background

In recent years, lithium ion batteries have increasingly demanded in the fields of smart power grids, electric vehicles, aerospace and the like due to the advantages of low cost, high energy density, long service life and the like. With the increase of the number of charge and discharge cycles, the lithium ion battery can generate complex physical and chemical changes, so that the electrochemical components are subjected to nonlinear degradation, thereby causing capacity and power attenuation and even failure, and affecting the safety and reliability of the battery in the use process. Therefore, it is necessary to monitor the health of the lithium ion battery over its entire lifetime to ensure safe and reliable operation. Battery state of health (SOH) is often defined as a percentage between the current capacity and the initial capacity, reflecting the current capacity of the battery to store and supply energy relative to the beginning of its life, where 100% represents a new battery, and when the SOH reaches 70-80% of the initial capacity, the battery life is considered to have reached the end.

At present, researchers at home and abroad are increasingly researching on the estimation of the health state of the lithium ion battery. In some methods, four values of constant-current charging time, constant-current charging capacity, constant-voltage charging time and constant-current charging capacity in the charging process are used as characteristics to train a support vector regression model to predict SOH. Some PKNN models are combined with Markov chain models, input quantity is reduced through feature extraction, complex nonlinear problems are effectively fitted, and high SOH estimation accuracy is obtained. Some EEMD methods are adopted to decompose battery capacity data into components of multiple frequencies, then LSTM-STW and GS-LM are adopted to predict each component, a final prediction result is obtained through integration, and the EEMD decomposition can effectively improve prediction accuracy on the noise reduction decomposition of an original signal.

The method for describing the battery aging phenomenon by selecting different characteristic quantities by using a mathematical statistical method can obviously reduce the calculation complexity in the model prediction process and realize the online prediction of the health state of the lithium battery by adopting a simple neural network. However, the current related approaches still have the following disadvantages: in most of the existing characteristic quantity selection methods, information is overlapped more, certain redundancy exists, and a proper characteristic quantity is required to be selected to describe the aging process of the lithium ion battery; the complexity of the neural network is too high, which has great influence on the overall performance of the model; the practicability of the model needs to be improved, the internal parameters of the model are updated dynamically by rarely using the latest data, and the fast SOH online prediction is realized.

Disclosure of Invention

This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.

The present invention has been made in view of the above-mentioned conventional problems.

Therefore, the technical problem solved by the invention is as follows: in most of the existing characteristic quantity selection methods, information is overlapped more, certain redundancy exists, and a proper characteristic quantity is required to be selected to describe the aging process of the lithium ion battery; the complexity of the neural network is too high, and the overall performance of the model is greatly influenced; the utility performance of the model is not high, and the internal parameters of the model are rarely dynamically updated by using the latest data.

In order to solve the technical problems, the invention provides the following technical scheme: selecting charging and discharging characteristics by utilizing a principal component analysis strategy; decomposing the charge and discharge characteristics into a main trend part and a secondary fluctuation part based on a complete empirical mode decomposition strategy of adaptive noise; predicting the main trend part based on a logistic regression strategy of a sliding time window, and predicting the secondary fluctuation part by adopting a Kalman filtering strategy; and combining the predicted main trend part with the secondary fluctuation part to obtain predicted characteristic data, and substituting the predicted characteristic data into the radial basis function neural network to realize the online prediction of the health state of the lithium battery.

As a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: the process of selecting the charge and discharge characteristics by using the principal component analysis strategy comprises the steps of converting original random vectors of the charge and discharge characteristics into new random vectors with irrelevant components by using orthogonal transformation; pointing the new random vector to a plurality of orthogonal directions in which the sample points are most spread, performing dimensionality reduction on a multi-dimensional variable system to obtain a plurality of orthogonal directions in which the new random vector is most spread by the low-dimensional variable system, and performing dimensionality reduction on the multi-dimensional variable system to obtain the low-dimensional variable system; and converting the original random vector into a group of representations which are linearly independent of each dimension, and screening out main characteristic components of the charge and discharge characteristic data.

As a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: the empirical mode decomposition strategy comprises decomposing a signal into k intrinsic mode functions, wherein each intrinsic mode function uses IMF_kRepresents; for each IMF_kThe jth IMF component of which is decomposed by empirical mode is E_j() It is shown that the original signal is defined as s (n), and white gaussian noise ω (n) is added to s (n).

As a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: the empirical mode decomposition strategy may further comprise,

computing IMF₁：

Wherein I represents the signal s (n) +. epsilon₀ω_i(n) is decomposed I times, the parameter epsilon represents the signal-to-noise ratio of the control additive noise to the original signal;

and residual calculation: when k is 1, the residual error is calculated as follows:

r₁(n)＝s(n)-IMF₁(n)

before next decomposition, performing empirical mode decomposition on the Gaussian white noise to obtain a value of a first component;

adding the value of the first component to the residual signal, and updating the signal to be decomposed into r₁(n)+ε₁E₁(ω_i(n)) (I ═ 1,2, …, I), and the residual calculation was performed again to obtain IMF₂；

The IMF₂The calculation method of (a) is as follows:

when K is 2,3, …, K, the K-th residual is:

r_k(n)＝r_k-1(n)-IMF_k(n)

to r_k(n)+ε_kE_k(ω_i(n)) (I ═ 1,2, …, I) are decomposed until the first empirical mode component, IMF, is obtained_k+1The calculation method of (a) is as follows:

repeatedly calculating the k-th residual sum IMF_k+1Until the residual signal cannot be resolvable, the original signal s (n) is expressed as a combination of k IMFs and one residual r (n), as follows:

as a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: the logistic function prob (event) of the logistic regression strategy for the sliding time window includes,

wherein, x (x)₁,x₂,…,x_d) Representing the input vectors corresponding to the d arguments, g (x) representing the logical model.

As a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: defining the logical model g (x) comprises,

wherein g (x) represents all input vectors x (x)₁,x₂,…,x_d) α, β represent polynomial coefficients of g (x).

As a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: the kalman filtering strategy includes a time update and a measurement update.

As a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: the time update includes the time of day that the user is,

discrete kalman filter time update equation:

wherein the content of the first and second substances,representing the a priori state estimate at time k, a representing the state transition matrix,representing the estimate of the posterior state at time k-1, B representing the matrix that transforms the input into state, u_k-1Representing the input at time k-1,representing the prior estimated covariance, P, of time k_k-1Represents the a posteriori estimated covariance at time k-1 and Q represents the covariance of the system process.

As a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: the measurement update includes a list of measurements of the measurement,

discrete kalman filter measurement update equation:

wherein, K_kRepresenting the Kalman gain, H representing the state variable to measurement transformation matrix, R representing the measurement noise covariance, z_kRepresenting the observed value.

As a preferred embodiment of the online estimation method for the health state of the lithium battery based on the feature analysis, the method comprises the following steps: the radial basis function neural network includes a plurality of radial basis function neural networks,

activation function F_activationCan be expressed as:

wherein | x_p-c_i|' represents the euclidean norm, c represents the center of the gaussian function, and σ represents the variance of the gaussian function;

according to the structure of the radial basis function neural network, the output of the network is obtained as follows:

wherein the content of the first and second substances,denotes the pth input sample, P1, 2,3, …, P denotes the total number of samples, c_iRepresenting the centres of nodes of the hidden layers of the network, w_nDenotes the connection weight from hidden layer to output layer, i ═ 1,2,3, …, h, i denotes the number of hidden layer nodes, y denotes the number of hidden layer nodes_jRepresenting the actual output of the jth output node of the network corresponding to the input sample,

defining d to be the expected output value of the sample, the variance σ of the radial basis function is then expressed as:

the invention has the beneficial effects that: proper charging and discharging characteristics are selected by utilizing a principal component analysis method to describe the battery aging process, original charging and discharging characteristic data are subjected to noise reduction treatment by utilizing CEEMDAN, the original charging and discharging characteristic data are divided into a main trend part and a secondary fluctuation part and are respectively predicted, and the prediction precision is good; predicting the main trend data and the secondary fluctuation data of the processed charge and discharge characteristic quantity respectively by using a logistic regression and Kalman filtering method based on a sliding time window, wherein the prediction result of the charge and discharge characteristic quantity is good; the degradation process of the lithium ion battery can be well described, and the SOH prediction precision is high.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise. Wherein:

fig. 1 is a schematic basic flow chart of a method for online estimation of a state of health of a lithium battery based on feature analysis according to an embodiment of the present invention;

fig. 2 is another schematic basic flow chart of a method for online estimation of a state of health of a lithium battery based on feature analysis according to an embodiment of the present invention;

fig. 3 is a partial schematic diagram of a main trend prediction method of an online estimation method of a lithium battery state of health based on feature analysis, which is based on a logistic regression based on a sliding time window according to an embodiment of the present invention;

fig. 4 is a schematic diagram illustrating a part of fluctuation prediction by using kalman filtering in an online estimation method for a lithium battery state of health based on feature analysis according to an embodiment of the present invention;

fig. 5 is a schematic diagram illustrating an online prediction result of a selected characteristic quantity of an online estimation method for a lithium battery state of health based on feature analysis according to an embodiment of the present invention;

fig. 6 is a schematic diagram of SOH online prediction results of B0005, B0006 and B0007 of a lithium battery state of health online estimation method based on feature analysis according to an embodiment of the present invention.

Detailed Description

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, specific embodiments accompanied with figures are described in detail below, and it is apparent that the described embodiments are a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present invention, shall fall within the protection scope of the present invention.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.

Furthermore, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.

The present invention will be described in detail with reference to the drawings, wherein the cross-sectional views illustrating the structure of the device are not enlarged partially in general scale for convenience of illustration, and the drawings are only exemplary and should not be construed as limiting the scope of the present invention. In addition, the three-dimensional dimensions of length, width and depth should be included in the actual fabrication.

Meanwhile, in the description of the present invention, it should be noted that the terms "upper, lower, inner and outer" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of describing the present invention and simplifying the description, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation and operate, and thus, cannot be construed as limiting the present invention. Furthermore, the terms first, second, or third are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.

The terms "mounted, connected and connected" in the present invention are to be understood broadly, unless otherwise explicitly specified or limited, for example: can be fixedly connected, detachably connected or integrally connected; they may be mechanically, electrically, or directly connected, or indirectly connected through intervening media, or may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.

Example 1

Referring to fig. 1 to 5, an embodiment of the present invention provides a method for online estimating a health state of a lithium battery based on feature analysis, including:

s1, selecting charging and discharging characteristics by utilizing a principal component analysis strategy; it should be noted that, in the following description,

the principal Component analysis strategy is a commonly used data analysis method pca (principal Component analysis), and the main idea is to transform raw data into a set of representations linearly independent of each dimension in a linear transformation manner, so as to extract the principal characteristic components of the data. Principal component analysis method for selecting U_rec、U_cc、t_chgThe aging process of the battery is described for the charge and discharge characteristics, and the specific process comprises the following steps:

(1) converting the original random vector of the charge-discharge characteristics into a new random vector with irrelevant components by utilizing orthogonal transformation;

(2) pointing the new random vector to a plurality of orthogonal directions with most spread sample points, and performing dimensionality reduction on the multidimensional variable system to obtain a low-dimensional variable system, so that the data input quantity of the model can be reduced, and the calculation complexity is reduced;

(3) and converting the original random vector into a group of expressions which are linearly independent of each dimension, and screening out main characteristic components of the charge and discharge characteristic data.

The aging of a lithium ion battery is generally reflected by the change of charging and discharging, and the reflected characteristics are as follows:

(1) charging time t_chg: as the number of battery cycles increases, the SOH will gradually decrease, and the time for the CC charging process will also decrease;

(2) constant current charging voltage U_cc: because the polarization characteristic inside the battery is different along with different aging stages of the battery, the change of the polarization characteristic can cause the change of voltage when the battery is placed after constant-current constant-voltage charging every time, and the aging process of the battery can be reflected;

(3) minimum voltage U_min: as the aging degree of the battery is increased, the voltage of the battery drops to the lowest point after constant current dischargeThe change is easy to identify;

(4) recovery voltage U_rec: after the battery is discharged at constant current and constant voltage, the battery is kept stand for a period of time, and the terminal voltage is finally recovered to a certain steady state value U_rec。

Too much feature data can cause the complexity of the algorithm to be improved, and the calculation time is increased, so that the feature quantity is selected by selecting a Principal Component Analysis (PCA) method, and the calculation complexity is reduced; the principal component analysis result shows that two principal components are extracted in total, the two principal components are also called factor loads, linear combination of original index variables can be obtained through a factor load matrix, and the result is expressed as U_recAnd t_chgThe coefficient is larger in principal component 1, U_ccThe coefficient is larger in principal component 2, indicating that it accounts for the most of the original variables, as shown in table 1.

Table 1: table of principal component analysis results.

S2, decomposing the charge and discharge characteristics into a main trend part and a secondary fluctuation part based on a complete empirical mode decomposition (CEEMDAN) method of the adaptive noise; it should be noted that, in the following description,

the empirical mode decomposition method is a non-stationary signal analysis method and comprises the following steps:

decomposing the signal into k eigenmode functions, each eigenmode function being represented by an IMF_kRepresents;

for each IMF_kThe jth IMF component of which is decomposed by empirical mode is E_j() It is shown that the original signal is defined as s (n), and white gaussian noise ω (n) is added to s (n).

Further, the empirical mode decomposition strategy may further include,

computing IMF₁：

Wherein I represents the signal s (n) +. epsilon₀ω_i(n) is decomposed I times, the parameter epsilon represents the signal-to-noise ratio of the control additive noise to the original signal;

and residual calculation: when k is 1, the residual error is calculated as follows:

r₁(n)＝s(n)-IMF₁(n)

before next decomposition, empirical mode decomposition is carried out on white noise to obtain a value of a first component, and then the value of the first component is added into a residual signal to eliminate an error caused by the noise on an original signal; the signal to be decomposed is updated to r₁(n)+ε₁E₁(ω_i(n)) (I ═ 1,2, …, I), and the residual calculation was performed again to obtain IMF₂；

IMF₂The calculation method of (a) is as follows:

when K is 2,3, …, K, the K-th residual is:

r_k(n)＝r_k-1(n)-IMF_k(n)

to r_k(n)+ε_kE_k(ω_i(n)) (I ═ 1,2, …, I) are decomposed until the first empirical mode component, IMF, is obtained_k+1The calculation method of (a) is as follows:

repeatedly calculating the k-th residual sum IMF_k+1Until the residual signal cannot be resolvable again, the original signal s (n) can be expressed as a combination of k IMFs and one residual r (n), as follows:

the CEEMDAN is adopted to decompose and reduce noise of data, after charge and discharge characteristic data of the battery are decomposed, the Pearson correlation coefficient between each component and original data is calculated, the Pearson correlation coefficient reflects the degree of closeness of correlation between variables, and the larger the Pearson correlation coefficient is, the larger the correlation between the variables is, and vice versa.

The results show that IMF₆The Pearson correlation coefficient is far greater than the IMF₁-IMF₅Selecting the main trend as the main trend of the charge and discharge characteristic data, predicting the main trend by using a logistic regression (w + LR) method, and simultaneously carrying out IMF (intrinsic mode function) on the rest components₁-IMF₅The sum of (a) is taken as a minor trend, and the minor trend is predicted by using a Kalman Filtering (KF) method. The calculation results are shown in table 2:

table 2: pearson correlation Table between decomposition results and raw data.

IMF	Urec	Ucc	tchg
				1	0.082	0.422	0.176
2	0.102	0.275	0.277
				3	0.085	0.220	0.222
4	0.065	0.111	0.214
				5	0.047	0.092	0.227
6	0.987	0.702	0.951

S3, predicting the main trend part application based on the logistic regression strategy (w + LR) of the sliding time window, and predicting the secondary fluctuation part by adopting a Kalman filtering strategy (KF); it should be noted that, in the following description,

sliding time window based logistic regression (w + LR) is a model that predicts the probability of an event occurring by fitting data to a logistic regression curve; its logic function prob (event) is as follows:

wherein, x (x)₁,x₂,…,x_d) Representing the input vector for d arguments, g (x) representing the logical method, the logical function can be used as the probability distribution function.

Further, defining the logic model g (x) includes:

wherein g (x) represents all input vectors x (x)₁,x₂,…,x_d) Alpha and beta represent polynomial coefficients of g (x), and P (x) is calculated on the premise that parameters alpha and beta are determined_i(i-1, 2, …, d) using maximum likelihood estimation for parameter determination;

along with the increase of the charge-discharge cycle of the battery, the window boundary moves along with the increase of the charge-discharge cycle of the battery, the window forgets part of historical data, the LR model is reconstructed by adopting a data sequence in a new window, and then the LR model is used for real-time prediction.

Kalman Filtering (KF) combines the initial state estimation of the system at the next moment and the feedback obtained by measurement to finally obtain the more accurate state estimation at the moment;

the discrete Kalman filtering time updating equation is as follows:

wherein the content of the first and second substances,representing the a priori state estimate at time k, a representing the state transition matrix,representing the estimate of the posterior state at time k-1, B representing the matrix that transforms the input into state, u_k-1Representing the input at time k-1,representing the prior estimated covariance, P, of time k_k-1Representing the posterior estimated covariance at time k-1, and Q representing the covariance of the system process;

discrete kalman filter measurement update equation:

wherein, K_kRepresenting the Kalman gain, H representing the state variable to measurement transformation matrix, R representing the measurement noise covariance, z_kRepresenting the observed value.

Predicting the main trend part by using logistic regression (w + LR), and starting predicting on the basis of historical data of the previous two characteristic quantities; as can be seen from FIG. 3, the prediction results of the logistic regression (w + LR) are very accurate; the secondary fluctuation part is predicted by Kalman Filtering (KF), and as can be seen from the prediction result in FIG. 4, the Kalman Filtering (KF) method has a good overall tracking effect on the fluctuation part, can well track the change of data volume, and has a good prediction effect;

as can be seen from fig. 5, the logistic regression (w + LR) is used to predict the main body trend part, the Kalman Filter (KF) is used to predict the fluctuation part, and the two parts are added together, i.e., the online prediction result of the selected feature quantity shows that the error between the predicted value and the true value of the feature quantity is very small, and the RMSE of the prediction result with the largest error is only 0.52%.

S4, combining the predicted main trend part with the secondary fluctuation part to obtain predicted characteristic data, and substituting the predicted characteristic data into a Radial Basis Function (RBF) neural network to realize the online prediction of the health state of the lithium battery; it should be noted that, in the following description,

the radial basis function neural network is a feedforward neural network with excellent performance and can approach any nonlinear function;

activation function F_activationExpressed as:

wherein | x_p-c_i|' represents the euclidean norm, c represents the center of the gaussian function, and σ represents the variance of the gaussian function;

according to the structure of the radial basis function neural network, the output of the network is obtained as follows:

wherein the content of the first and second substances,denotes the pth input sample, P1, 2,3, …, P denotes the total number of samples, c_iRepresenting the centres of nodes of the hidden layers of the network, w_nDenotes the connection weight from hidden layer to output layer, i ═ 1,2,3, …, h denotes the number of hidden layer nodes, y denotes the number of hidden layer nodes_jRepresenting the actual output of the jth output node of the network corresponding to the input sample;

assuming d is the desired output value of the sample, then the variance of the radial basis function is expressed as:

the characteristic quantity of online prediction is used as RBF network input, new characteristic and SOH data are observed each time, the characteristic quantity is used as historical data to update the network, and the RBF network is updated in real time to realize the real-time SOH prediction.

The RBF neural network is a forward neural network consisting of three layers of neuron nodes, the first layer is an input layer and consists of a sensing unit, the input layer is connected with the external input of the neural network, and an input vector is directly mapped to a hidden space of a hidden layer; the second layer is a hidden layer which transforms low-dimensional input data into a high-dimensional space, so that the problem that the linearity in the low-dimensional space is inseparable is linearly separable in the high-dimensional space; the third layer is the output layer, which responds to the inputs.

The invention utilizes a principal component analysis method to select proper charge-discharge characteristics to describe the aging process of the battery. Noise reduction processing is carried out on the original charging and discharging characteristic data by using CEEMDAN, the original charging and discharging characteristic data are divided into a main trend part and a secondary fluctuation part and are respectively predicted, and the proposed charging and discharging characteristic selection scheme based on PCA has good prediction accuracy; predicting the main trend data and the secondary fluctuation data of the processed charge and discharge characteristic quantity respectively by using a logistic regression and Kalman filtering method based on a sliding time window, wherein the prediction result of the charge and discharge characteristic quantity is good; the sum of the predicted main trend and the predicted secondary trend of the charge and discharge characteristic quantity is substituted into the RBF neural network, so that an SOH online prediction result can be obtained, and the SOH prediction precision in the lithium ion battery degradation process can be well described. In the method, as logistic regression (w + LR) and Kalman Filtering (KF) methods are adopted, historical data can be abandoned gradually, new data are adopted to update internal parameters of the RBF network in an iterative manner, the real-time prediction precision of SOH is improved by using the new data, and the prediction precision is better and better. The method can accurately predict the whole trend of the capacity attenuation, can quickly capture the capacity regeneration phenomenon and realizes the accurate prediction of the whole attenuation stage.

Example 2

Referring to fig. 6, this embodiment is a second embodiment of the present invention, which is different from the first embodiment in that a verification test of an online estimation method of a lithium battery health state based on feature analysis is provided, and to verify and explain technical effects adopted in the method, this embodiment adopts a conventional technical scheme and the method of the present invention to perform a comparison test, and compares test results by means of scientific demonstration to verify a true effect of the method.

The traditional technical scheme is as follows: the Kalman filtering method cannot realize quick updating on the capacity regeneration point, and the effect of capturing the capacity regeneration point is poor; when the support vector machine method maps data from low dimension to high dimension to do regression, the tracking performance of the capacity attenuation curve can not be ensured, and the local error is larger; the long-time and short-time neural network method is high in calculation complexity, and prediction accuracy is influenced by data cycle characteristics.

Compared with the traditional method, the method has higher prediction precision and capability of accurately capturing the capacity regeneration point.

In this embodiment, a traditional kalman filtering method, a support vector machine method, a long-and-short-term neural network method, and the method are respectively used to perform real-time measurement and comparison on the estimation accuracy of the health state of the lithium battery.

And (3) testing environment: all the batteries were subjected to a charge-discharge process at room temperature, which was as follows: charge in 1.5A Constant Current (CC) mode until the battery voltage reaches 4.2V, then charge in Constant Voltage (CV) mode until the charge current drops to 20 mA. The discharge was performed at a constant 2A current (CC) level until the voltage of batteries 5, 6, 7 dropped to 2.7V, 2.5V, 2.2V, respectively. By adopting the method, the estimation prediction of the method is realized by using MATLAB software programming, and the health state estimation data is obtained according to the experimental result. Three sets of data were tested for each method, and the error calculated by comparing the data with the actual capacity data input by the simulation is shown in table 3.

Table 3: the error comparison table is calculated by the traditional method and the method of the invention.

As shown in fig. 6, it can be seen that the online prediction accuracy of the method provided by the present invention is high, the robustness is strong, the health state estimation accuracy of the lithium battery is compared in real time according to the traditional kalman filtering method, the support vector machine method, the long-short time neural network method and the method, and the results are shown in table 3, and it is found that the MAE, MAPE, and RMSE error data of the method are smaller than the error data of the kalman filtering method, the support vector machine method, and the long-short time neural network method, and compared with the traditional method, the method has higher prediction accuracy and the capability of accurately capturing the capacity regeneration point.

It should be noted that the above-mentioned embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.