Lithium-ion battery health status estimation method based on IPSO-GPR algorithm

Abstract

The present invention discloses a lithium-ion battery state of health (SOH) estimation method based on the IPSO-GPR algorithm, comprising the following steps: 1. extracting and optimizing lithium-ion battery health indicators; 2. constructing a mapping relationship between battery health characteristics and health status using Gaussian process regression; 3. optimizing the GPR kernel parameters using an improved particle swarm optimization algorithm with a variation factor that can adaptively adjust weights based on population diversity; and 4. establishing a GPR-based state of health (SOH) estimation model to perform SOH estimation. The present invention is based on feature extraction based on discrete wavelet packet transform and Box-Cox transform, utilizes GPR to construct a mapping relationship between features and SOH, utilizes an improved particle swarm optimization algorithm to optimize the GPR kernel parameters, and establishes a GPR-based SOH estimation model, enabling accurate and rapid SOH estimation of lithium-ion batteries.

Description

Technical Field

[0001] The present invention relates to the field of lithium-ion battery state estimation, and in particular to a lithium-ion battery health state estimation method based on an IPSO-GPR algorithm. Background Art

[0002] To address the dual pressures of a growing population on both the environment and energy fronts, society is rapidly developing various clean energy sources to achieve carbon peak and carbon neutrality goals. Lithium-ion batteries, with their ability to strike a good balance between performance parameters such as cost, lifespan, and power, play a vital role in grid energy storage systems and new energy vehicles. However, due to complex internal mechanisms, the performance of lithium-ion batteries gradually declines with age, increasing the maintenance and replacement costs of energy storage equipment and even creating serious safety hazards. Therefore, timely and accurate health status prediction is crucial. Real-time estimation of the health quality of lithium-ion batteries can enable timely planning and improved usage of facilities that rely on lithium-ion batteries as their primary energy supply or storage.

[0003] Existing SOH estimation methods for lithium-ion batteries focus on analyzing the relationship between the health status characteristics and the actual health status of lithium-ion batteries. They do not consider that the health status characteristics of lithium-ion batteries are restricted by various factors during the extraction process, resulting in insufficient estimation accuracy. Existing SOH estimation is mainly achieved through direct measurement, model-driven methods, and data-driven methods. Direct measurement methods are mainly based on experiments, which require professional personnel and facilities, have relatively poor accuracy, and cannot be applied online. Model-driven methods can reflect the physical and chemical properties inside the battery, but problems such as model establishment and parameter identification difficulties also limit the development and application of such methods. Data-driven methods do not need to consider the specific reaction mechanism and can be implemented in battery management systems, but are subject to computing power limitations in actual applications. Data-driven methods are rarely used in actual engineering. Summary of the Invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the present invention proposes a lithium-ion battery health state estimation method based on the IPSO-GPR algorithm, in order to achieve accurate and rapid online SOH estimation under complex and changeable discharge conditions, thereby improving the health state prediction accuracy when the accuracy of lithium-ion battery health state feature extraction is poor, and improving the applicability of the lithium-ion battery health state estimation method.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] The present invention is characterized in that a method for estimating the health status of a lithium-ion battery based on the IPSO-GPR algorithm is performed according to the following steps:

[0007] Step 1: Use the historical voltage and current data of lithium-ion batteries as a training set to extract lithium-ion battery health indicators:

[0008] Step 1.1, construct health indicators of lithium-ion batteries;

[0009] Step 1.1.1. Set the upper and lower voltage limits of the lithium-ion battery during constant current charging and the upper and lower current limits of the lithium-ion battery during constant voltage charging;

[0010] Step 1.1.2, the constant current charging process of the lithium-ion battery is divided into several voltage intervals; for any a-th voltage interval in the constant current charging process of the lithium-ion battery, the voltage variation coefficient of the equal charging voltage difference is extracted from the historical voltage data in the a-th voltage interval using formula (1): and serves as the first health indicator;

[0011] The constant voltage charging process of the lithium-ion battery is divided into several current intervals. For the cth current interval in the constant voltage charging process of the lithium-ion battery, the current variation coefficient HI2 of equal charging current difference is extracted from the historical current data in the cth current interval using formula (2): c and serves as a second health indicator;

[0012]

[0013]

[0014] In formula (1) and formula (2), are respectively the standard deviation of the historical voltage data in the a-th voltage interval and the standard deviation of the historical current data in the c-th current interval; are the average values ​​of historical voltage data in the a-th voltage interval and the average values ​​of historical current data in the c-th current interval respectively; and are the number of samples of historical voltage data in the a-th voltage interval and the number of samples of historical current data in the c-th current interval respectively; is the i-th historical voltage data in the a-th voltage interval; is the jth historical current data in the cth current interval;

[0015] Step 1.1.3, using formula (3) to extract the charging energy of equal charging voltage difference from the historical voltage data in the a-th voltage interval and as a third health indicator;

[0016] The charging energy HI4 of equal charging current difference is extracted from the current data in the cth current interval using formula (4):c and as the fourth health indicator;

[0017]

[0018]

[0019] In formulas (3) and (4), is the current value of the ath voltage interval in the constant current charging process; is the voltage value changing with time t in the ath voltage interval during the constant current charging process; and Indicates the start time and end time of the ath voltage interval; V CV is the voltage value in the cth current interval of the constant voltage charging process; I CV (t) is the current value changing with time t in the cth current interval of the constant voltage charging process; and Indicates the start time and end time of the cth current interval;

[0020] Step 1.2: Use the ergodic method combined with the Pearson correlation coefficient to find the optimal voltage range and the optimal current range;

[0021] Step 1.2.1: Calculate the first health index of the ath voltage interval using formula (5): Its battery health status Pearson correlation coefficient between Thus, the Pearson correlation coefficient between the first health index and the battery health state SOH of all voltage intervals is obtained, and the voltage interval corresponding to the maximum first health index is selected as the suboptimal voltage interval;

[0022] Calculate the third health index of the bth voltage interval using formula (6): Its battery health status Pearson correlation coefficient between Thus, the Pearson correlation coefficient between the third health index and the battery health status SOH of all voltage intervals is obtained, and the voltage interval corresponding to the maximum third health index is selected as another suboptimal voltage interval;

[0023]

[0024]

[0025] In formulas (5) and (6), E() is the mean function, E 2 () is the square of the mean function;

[0026] Step 1.2.2: Calculate the second health index of the cth current interval using formula (7): Its battery health status Pearson correlation coefficient between Thus, the Pearson correlation coefficient between the second health index and the battery health state SOH of all current intervals is obtained, and the voltage interval corresponding to the maximum second health index is selected as another suboptimal current interval;

[0027] Calculate the fourth health index of the dth current interval using formula (8): Its battery health status Pearson correlation coefficient between Thus, the Pearson correlation coefficient between the fourth health index and the battery health state SOH of all current intervals is obtained, and the current interval corresponding to the maximum fourth health index is selected as another suboptimal current interval;

[0028]

[0029]

[0030] Step 1.2.3: Select the voltage interval corresponding to the larger absolute value from the two suboptimal voltage intervals as the optimal voltage interval; select the current interval corresponding to the larger absolute value from the suboptimal current voltage intervals as the optimal current interval;

[0031] Step 2: Use Box-Cox transform and discrete wavelet packet transform to optimize the health indicators of lithium-ion batteries:

[0032] Step 2.1, using a measuring device to obtain online current data of the lithium-ion battery, and performing real-time processing based on the online current data to obtain the second health indicator and the fourth health indicator of the full cycle;

[0033] Step 2.2: Normalize the health index of the optimal current range, the battery health status SOH, and the health index of the entire cycle to obtain the second health index of the normalized optimal current range. and the fourth health indicator Corresponding battery health status And the normalized second health indicator of the whole cycle and the fourth health indicator

[0034] Step 2.3: Use equations (9) to (12) to and as well as and Perform Box-Cox transformation to obtain the second health index of the optimal current interval after transformation The second health indicator of the whole cycle after transformation The fourth health indicator of the optimal current range after transformation And the fourth health indicator of the whole cycle after transformation

[0035]

[0036]

[0037]

[0038]

[0039] In formula (9-12), c is a constant that ensures that the transformed data is positive; λ1 and λ2 are two transformation parameters;

[0040] Step 2.4: Perform discrete wavelet packet decomposition on the lithium-ion battery health index to obtain the reconstructed overall health index THI DWPT Wavelet packet tree;

[0041] Step 3: SOH estimation based on the improved GPR model;

[0042] Step 3.1. Define health indicators for lithium-ion battery training in, Represents the reconstructed health indicator THI DWPT The wavelet packet tree of the second health indicator and the wavelet packet tree of the fourth health indicator in the wavelet packet tree, and define their corresponding health states

[0043] Step 3.2: Take THIG as the input of GPR model and the health status corresponding to THIG As the output of the GPR model, the SE kernel function k of the GPR model is constructed using formula (13) SE :

[0044]

[0045] In formula (13), THIG m ,THIG n are the mth and nth health indicators in the training health indicator THIG, and m=1,2,3,4, n=1,2,3,4, m≠n, θ is the scale parameter that controls the decay speed of the kernel function;

[0046] Step 3.3: Use the improved particle swarm optimization algorithm to optimize the SE kernel parameter k of the GPR model. SE Optimize to get the optimal scale parameter θ * , and used to construct the optimal GPR model for SOH estimation;

[0047] Step 3.4: Use the measuring device to obtain the online voltage and current data of the lithium-ion battery and process them to obtain the online health index. The online health index is input into the optimal GPR model to obtain the optimal SOH estimation result.

[0048] The lithium-ion battery health status estimation method based on the IPSO-GPR algorithm of the present invention is also characterized in that step 2.4 includes:

[0049] Step 2.4.1. Define the overall health indicators of lithium-ion batteries The overall health index THI is decomposed by 3-layer discrete wavelet packet to obtain the wavelet packet tree of the overall health index THI; wherein the overall health index set after the 3rd layer decomposition is recorded as THI3 = {THI 3,0 , THI 3,1 ,…,THI 3,i ,…,THI 3,7}; Among them, THI 3,i represents the i-th overall health indicator after the third-level decomposition;

[0050] Step 2.4.2: Define the overall health status of the lithium-ion battery The SOH is decomposed by 3-layer discrete wavelet packet to obtain the wavelet packet tree of the overall health state SOH; wherein the overall health state set after the 3rd layer decomposition is recorded as SOH3 = {SOH 3,0 , SOH 3,1 ,…,SOH 3,i ,…,SOH 3,7}, where SOH 3,i represents the i-th overall health status after the third-level decomposition;

[0051] Step 2.4.3, calculate the Pearson correlation coefficient P3 between THI3 and SOH3 respectively;

[0052] Step 2.4.4: Based on P3, the wavelet packet coefficient corresponding to the terminal node THI3 of the wavelet packet tree of the overall health index THI is modified to obtain the terminal node THI3′ of the modified wavelet packet tree;

[0053] Step 2.4.5: Perform discrete wavelet packet reconstruction on the terminal node THI3′ of the modified wavelet packet tree to obtain the reconstructed overall health index THI DWPT Wavelet packet tree.

[0054] The step 3.3 includes:

[0055] Step 3.3.1. Set the parameter θ of the GPR model as the position vector of the particle, and set the population size of the particle swarm to P and the maximum number of iterations to t. max ; Define the current number of iterations as t and initialize t = 1;

[0056] Step 3.3.2: Randomly initialize the velocity of the pth particle in the tth generation particle swarm and location

[0057] Step 3.3.3: Calculate the diversity d of the particle swarm of generation t using formula (14): t :

[0058]

[0059] In formula (14), δ is the longest diagonal length of the search space, represents the d-th dimension position of the p-th particle in the t-th generation particle swarm; is the average value of all particles in the t-th generation particle swarm at the d-th dimension position; P is the population size; D is the dimension of the particle;

[0060] Step 3.3.4: Use formula (15) to calculate the relationship weight w of the t-th generation particle swarm t ;

[0061]

[0062] In formula (15), d low and d high are the low and high thresholds of population diversity respectively; w min and w max are the low threshold and high threshold of the relationship weight between populations respectively; w t-1 is the relationship weight of the t-1 generation particle swarm; when t=1, let w t-1 =δ, δ represents a fixed value;

[0063] Step 3.3.5: Use the root mean square error as the fitness function of IPSO to calculate the fitness of the pth particle in the tth generation particle swarm. Thus, the local optimal solution of the p-th particle in the t-th generation is obtained Select the optimal solution corresponding to the maximum fitness value from the local optimal solutions of all particles in the tth generation as the global optimal position gbest of the tth generation t ;

[0064] Step 3.3.6: Use formula (16) to get the position θ of the pth particle in the t+1th generation particle swarm.p t+1 and speed

[0065]

[0066] In formula (20), c1 and c2 are two learning factors, r1 and r2 are two random functions with values ​​ranging from [0,1];

[0067] Step 3.3.7: After assigning t+1 to t, determine whether t>t max Is it true? If so, output the tth max Global optimal position of the generation population And as the optimal hyperparameter θ * , and used to construct the optimal GPR model for SOH estimation; otherwise, go to step 3.3.2 to continue optimization.

[0068] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the lithium-ion battery health status estimation method, and the processor is configured to execute the program stored in the memory.

[0069] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the lithium-ion battery health status estimation method when the computer program is executed by a processor.

[0070] Compared with the prior art, the beneficial effects of the present invention are embodied in:

[0071] 1. This invention addresses the difficulty in measuring capacity parameters that reflect the battery's state of health (SOH). By analyzing lithium-ion battery charging data, it extracts four highly correlated health indicators for SOH estimation. This method boasts high efficiency and robust applicability, making it applicable to most charging scenarios.

[0072] 2. To address the noise problem in the feature extraction process, the present invention proposes a new feature processing method based on discrete wavelet packet transform and Box-Cox transform, which eliminates noise and improves the correlation between features and capacity.

[0073] 3. This paper addresses the hyperparameter selection problem of the GPR model by introducing an IPSO algorithm that adaptively adjusts weights based on population diversity, resulting in a high global search capability. Using the IPSO algorithm to optimize the GPR kernel parameters effectively avoids the algorithm from falling into local extremes and achieves a global optimal solution.

[0074] 4. To address the problems of inaccurate health indicator extraction and poor model training when estimating the SOH of lithium-ion batteries using machine learning methods, the present invention adopts an improved particle swarm optimization algorithm to optimize the hyperparameters of the GPR model and constructs an integrated SOH estimation framework based on partial charging curve feature extraction, feature processing and IPSO-GPR, which can achieve accurate and fast SOH estimation of lithium-ion batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is the overall flow chart of the health status estimation method of lithium-ion batteries based on improved particle swarm optimization and Gaussian process regression;

[0076] Figure 2 Flowchart for lithium-ion battery health characterization processing;

[0077] Figure 3 Flowchart of the IPSO-GPR algorithm. DETAILED DESCRIPTION

[0078] In this embodiment, a lithium-ion battery state of health (SOH) estimation method based on an improved particle swarm optimization (IPSO) and Gaussian process regression (GPR) is constructed. This method addresses the problem of decreased accuracy caused by noise in the feature extraction process. By fully studying the law of battery charge and discharge data degradation with SOH under different cycles, four health indicators (HI) with high extraction efficiency and strong applicability that characterize battery SOH degradation are extracted from part of the charging process data. The features are further processed using Box-Cox Transformation (BCT) and Discrete Wavelet Packet Transform (DWPT) to eliminate noise and enhance the degree of linear correlation. On this basis, GPR is used to construct the mapping relationship between features and SOH, and the improved particle swarm optimization algorithm (IPSO) is used to optimize the kernel parameters of GPR. The four health indicators extracted and optimized are used as model inputs, and SOH is used as model output. The IPSO-GPR estimation model is constructed to estimate SOH. Specifically, a lithium-ion battery health state estimation method based on the IPSO-GPR algorithm is proposed. Figure 1 As shown, the steps are as follows:

[0079] Step 1: Use the historical voltage and current data of lithium-ion batteries as a training set to extract lithium-ion battery health indicators:

[0080] Step 1.1. Construct health indicators for lithium-ion batteries; when extracting features for SOH estimation, the correlation between features and SOH determines the estimation accuracy to a certain extent. Reasonable and highly correlated health indicators are conducive to improving the accuracy of battery degradation modeling. Due to the complexity and variability of actual working conditions, battery management systems are always affected by external noise, transient faults and other similar problems on sensors in actual applications, which may cause some online monitoring data to show fluctuating values. Since battery data sets are inevitably affected by noise, the extraction of certain features will also carry noise, which in turn affects the accuracy of the model. The feature extraction method proposed in the present invention has high extraction efficiency and strong applicability, can effectively improve the correlation between features and capacity, and ensure the stability of the model;

[0081] Step 1.1.1. Set the upper and lower voltage limits of the lithium-ion battery during constant current charging and the upper and lower current limits of the lithium-ion battery during constant voltage charging;

[0082] Step 1.1.2: Divide the constant current charging process of the lithium-ion battery into several voltage intervals. Due to battery life concerns and battery protection, users often do not perform full discharge and charging, which means that it is difficult to obtain data under extreme SOC during the charging process. In addition, it is worth noting that the larger the interval, the lower the calculation efficiency; the smaller the interval, the more sensitive the features extracted based on the interval data will be to interference factors such as current fluctuations, ambient temperature, and test errors. For any a-th voltage interval in the constant current charging process of the lithium-ion battery, use formula (1) to extract the voltage variation coefficient of equal charging voltage difference from the historical voltage data in the a-th voltage interval. and serves as the first health indicator;

[0083] The constant voltage charging process of the lithium-ion battery is divided into several current intervals. For the cth current interval in the constant voltage charging process of the lithium-ion battery, the current variation coefficient of equal charging current difference is extracted from the historical current data in the cth current interval using formula (2): and serves as a second health indicator;

[0084]

[0085]

[0086] In formula (1) and formula (2), are respectively the standard deviation of the historical voltage data in the a-th voltage interval and the standard deviation of the historical current data in the c-th current interval; are the average values ​​of historical voltage data in the a-th voltage interval and the average values ​​of historical current data in the c-th current interval respectively; and are the number of samples of historical voltage data in the ath voltage interval and the number of samples of historical current data in the cth current interval respectively; V i a is the i-th historical voltage data in the a-th voltage interval; is the jth historical current data in the cth current interval;

[0087] Step 1.1.3, using formula (3) to extract the charging energy of equal charging voltage difference from the historical voltage data in the a-th voltage interval and as a third health indicator;

[0088] The charging energy of equal charging current difference is extracted from the current data in the cth current interval using formula (4). and as the fourth health indicator;

[0089]

[0090]

[0091] In formulas (3) and (4), is the current value of the ath voltage interval in the constant current charging process; is the voltage value changing with time t in the ath voltage interval during the constant current charging process; and Indicates the start time and end time of the ath voltage interval; V CV is the voltage value in the cth current interval of the constant voltage charging process; I CV (t) is the current value changing with time t in the cth current interval of the constant voltage charging process; and Indicates the start time and end time of the cth current interval;

[0092] Step 1.2: Use the ergodic method combined with the Pearson correlation coefficient to find the optimal voltage range and the optimal current range;

[0093] Step 1.2.1: Calculate the first health index of the ath voltage interval using formula (5): Its battery health status SOH1 a Pearson correlation coefficient P1 between a; Thus, the Pearson correlation coefficient between the first health index and the battery health state SOH of all voltage intervals is obtained, and the voltage interval corresponding to the maximum first health index is selected as the suboptimal voltage interval;

[0094] Calculate the third health index of the bth voltage interval using formula (6): Its battery health status Pearson correlation coefficient between Thus, the Pearson correlation coefficient between the third health index and the battery health status SOH of all voltage intervals is obtained, and the voltage interval corresponding to the maximum third health index is selected as another suboptimal voltage interval;

[0095]

[0096]

[0097] In formulas (5) and (6), E() is the mean function, E 2 () is the square of the mean function;

[0098] Step 1.2.2: Calculate the second health index of the cth current interval using formula (7): Its battery health status Pearson correlation coefficient between Thus, the Pearson correlation coefficient between the second health index and the battery health state SOH of all current intervals is obtained, and the voltage interval corresponding to the maximum second health index is selected as another suboptimal current interval;

[0099] Calculate the fourth health index of the dth current interval using formula (8): Its battery health status Pearson correlation coefficient between Thus, the Pearson correlation coefficient between the fourth health index and the battery health state SOH of all current intervals is obtained, and the current interval corresponding to the maximum fourth health index is selected as another suboptimal current interval;

[0100]

[0101]

[0102] Step 1.2.3: Select the voltage interval corresponding to the larger absolute value from the two suboptimal voltage intervals as the optimal voltage interval; select the current interval corresponding to the larger absolute value from the suboptimal current voltage intervals as the optimal current interval;

[0103] Step 2: In actual applications, sensors are inevitably affected by external noise and transient faults, which can cause the initially extracted health indicator data to have weak linear correlation with the target quantity and some unrealistic fluctuations. Therefore, eliminating noise and enhancing the degree of linear correlation are crucial for obtaining reliable results. The health indicators of lithium-ion batteries are optimized using Box-Cox transform and discrete wavelet packet transform. The process of lithium-ion battery health feature processing is shown in Figure 2 ;

[0104] Step 2.1, using a measuring device to obtain online current data of the lithium-ion battery, and performing real-time processing based on the online current data to obtain the second health indicator and the fourth health indicator of the full cycle;

[0105] Step 2.2: Normalize the health index of the optimal current range, the battery health status SOH, and the health index of the entire cycle to obtain the second health index of the normalized optimal current range. and the fourth health indicator Corresponding battery health status And the normalized second health indicator of the whole cycle and the fourth health indicator

[0106] Step 2.3: Use equations (9) to (12) to and as well as and Perform Box-Cox transformation to obtain the second health index of the optimal current interval after transformation The second health indicator of the whole cycle after transformation The fourth health indicator of the transformed optimal current range And the fourth health indicator of the whole cycle after transformation

[0107]

[0108]

[0109]

[0110]

[0111] In formula (9-12), c is a constant that ensures that the transformed data is positive; λ1 and λ2 are two transformation parameters;

[0112] Step 2.4: Perform discrete wavelet packet decomposition on the lithium-ion battery health indicators. DWPT is based on discrete wavelet transform (DWPT). While DWPT can only perform coarse-grained analysis of the target signal, DWPT can simultaneously decompose both high-frequency and low-frequency components, achieving fine-grained signal analysis. It has excellent time-frequency characteristics and multi-scale and multi-resolution properties. DWPT is used to further process the features. Wavelet packet transform is mainly divided into two processes: decomposition and reconstruction.

[0113] Step 2.4.1. Define the overall health indicators of lithium-ion batteries The overall health index THI is decomposed by 3-layer discrete wavelet packet to obtain the wavelet packet tree of the overall health index THI; wherein the overall health index set after the 3rd layer decomposition is recorded as THI3 = {THI 3,0 , THI 3,1 ,…,THI 3,i ,…,THI 3,7}; Among them, THI 3,i represents the i-th overall health indicator after the third-level decomposition;

[0114] Step 2.4.2: Define the overall health status of the lithium-ion battery The SOH is decomposed by 3-layer discrete wavelet packet to obtain the wavelet packet tree of the overall health state SOH; wherein the overall health state set after the 3rd layer decomposition is recorded as SOH3 = {SOH 3,0 , SOH 3,1 ,…,SOH 3,i ,…,SOH 3,7}, where SOH 3,i represents the i-th overall health status after the third-level decomposition;

[0115] Step 2.4.3, calculate the Pearson correlation coefficient P3 between THI3 and SOH3 respectively;

[0116] Step 2.4.4: Based on P3, modify the wavelet packet coefficients corresponding to the terminal node THI3 of the wavelet packet tree for the overall health indicator THI, thereby obtaining the modified terminal node THI3′ of the wavelet packet tree. Set the wavelet packet coefficients corresponding to the frequency bands with correlation coefficients lower than 0.2 to 0, and use the remaining coefficients as multipliers for the wavelet packet coefficients corresponding to the frequency bands for modification. Correlation coefficients lower than 0.2 indicate extremely weak or no correlation between the two. This part of the data may be caused by random noise and has high volatility and heteroscedasticity, so this part of the information is eliminated.

[0117] Step 2.4.5: Perform discrete wavelet packet reconstruction on the terminal node THI3′ of the modified wavelet packet tree to obtain the reconstructed overall health index THI DWPTWavelet packet tree:

[0118] Step 3: SOH estimation based on the improved GPR model; GPR is a machine learning method based on kernel function, which can describe the relevant uncertainty by providing confidence intervals of predicted values; the flowchart of the IPSO-GPR algorithm is as follows Figure 3 As shown in the figure; when using GPR to construct the mapping relationship between features and SOH, the setting of hyperparameters has a great influence on the performance of the machine learning algorithm. Improper parameter selection will cause the model to fall into local optimality and produce large estimation errors. Therefore, to address the hyperparameter selection problem of the GPR model, an improved particle swarm optimization algorithm that can adaptively adjust weights according to population diversity is introduced, which has a high global search capability. The IPSO algorithm is used to optimize the kernel parameters of the GPR, and the optimal particle position is used as the final parameter of the GPR model, which can effectively improve the accuracy of the algorithm.

[0119] Step 3.1. Define health indicators for lithium-ion battery training in, Represents the reconstructed health indicator THI DWPT The wavelet packet tree of the second health indicator and the wavelet packet tree of the fourth health indicator in the wavelet packet tree, and define their corresponding health states

[0120] Step 3.2: Take THIG as the input of GPR model and the health status corresponding to THIG As the output of the GPR model, the SE kernel function k of the GPR model is constructed using formula (13) SE :

[0121]

[0122] In formula (13), THIG m ,THIG n are the mth and nth health indicators in the training health indicator THIG, and m=1,2,3,4, n=1,2,3,4, m≠n, θ is the scale parameter that controls the decay speed of the kernel function;

[0123] Step 3.3: Use the improved particle swarm optimization algorithm to optimize the SE kernel parameter k of the GPR model. SE Optimize to get the optimal scale parameter θ * ;

[0124] Step 3.3.1. Set the parameter θ of the GPR model as the position vector of the particle, and set the population size of the particle swarm to P and the maximum number of iterations to t. max; Define the current number of iterations as t and initialize t = 1;

[0125] Step 3.3.2: Randomly initialize the velocity of the pth particle in the tth generation particle swarm and location

[0126] Step 3.3.3: Calculate the diversity d of the particle swarm of generation t using formula (14): t :

[0127]

[0128] In formula (14), δ is the longest diagonal length of the search space, represents the d-th dimension position of the p-th particle in the t-th generation particle swarm; is the average value of all particles in the t-th generation particle swarm at the d-th dimension position; P is the population size; D is the dimension of the particle;

[0129] Step 3.3.4: Use formula (15) to calculate the relationship weight w of the t-th generation particle swarm t During the random initialization phase of the particle swarm algorithm, the population diversity is high and the particles are relatively dispersed. A larger w should be assigned to enable the particles to quickly approach the optimal search area. As the iteration proceeds, the population diversity gradually decreases. At this time, w can be reduced to enable the particles to better perform local search.

[0130]

[0131] In formula (15), d low and d high are the low and high thresholds of population diversity respectively; w min and w max are the low threshold and high threshold of the relationship weight between populations respectively; w t-1 is the relationship weight of the t-1 generation particle swarm; when t=1, let w t-1 =δ, δ represents a constant; δ is usually set to a smaller value to reduce the relationship weight of the subgroup so that the particles can perform local search better.

[0132] Step 3.3.5: Use the root mean square error as the fitness function of IPSO to calculate the fitness of the pth particle in the tth generation particle swarm. Thus, the local optimal solution pbest of the p-th particle in the t-th generation is obtained t p , repeatedly calculate to obtain the local optimal solution of all particles in the tth generation, and select the optimal solution corresponding to the maximum fitness value as the global optimal position gbest of the tth generation t ;

[0133] Step 3.3.6: Use formula (16) to get the position θ of the pth particle in the t+1th generation particle swarm. p t+1 and speed

[0134]

[0135] In formula (20), c1 and c2 are two learning factors, r1 and r2 are two random functions with values ​​ranging from [0,1];

[0136] Step 3.3.7: After assigning t+1 to t, determine whether t>t max Is it true? If so, output the tth max Global optimal position of the generation population And as the optimal hyperparameter θ * , and used to construct the optimal GPR model for SOH estimation; otherwise, go to step 3.3.2 to continue optimization;

[0137] Step 3.4: Use the measuring device to obtain the online voltage and current data of the lithium-ion battery and process them to obtain the online health index. The online health index is input into the optimal GPR model to obtain the optimal SOH estimation result.

[0138] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0139] In this embodiment, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are executed.

Claims

1. A lithium-ion battery health status estimation method based on the IPSO-GPR algorithm, characterized in that: The steps are as follows: Step 1: Use the historical voltage and current data of lithium-ion batteries as a training set to extract lithium-ion battery health indicators: Step 1.1, construct health indicators of lithium-ion batteries; Step 1.1.

1. Set the upper and lower voltage limits of the lithium-ion battery during constant current charging and the upper and lower current limits of the lithium-ion battery during constant voltage charging; Step 1.1.2, the constant current charging process of the lithium-ion battery is divided into several voltage intervals; for any a-th voltage interval in the constant current charging process of the lithium-ion battery, the voltage variation coefficient of the equal charging voltage difference is extracted from the historical voltage data in the a-th voltage interval using formula (1): and serves as the first health indicator; The constant voltage charging process of the lithium-ion battery is divided into several current intervals. For the cth current interval in the constant voltage charging process of the lithium-ion battery, the current variation coefficient of equal charging current difference is extracted from the historical current data in the cth current interval using formula (2): and serves as a second health indicator; (1) (2) In formula (1) and formula (2), 、 are respectively the standard deviation of the historical voltage data in the a-th voltage interval and the standard deviation of the historical current data in the c-th current interval; 、 are the average values ​​of historical voltage data in the a-th voltage interval and the average values ​​of historical current data in the c-th current interval respectively; and are the number of samples of historical voltage data in the a-th voltage interval and the number of samples of historical current data in the c-th current interval respectively; is the i-th historical voltage data in the a-th voltage interval; is the jth historical current data in the cth current interval; Step 1.1.3, using formula (3) to extract the charging energy of equal charging voltage difference from the historical voltage data in the a-th voltage interval and as a third health indicator; The charging energy of equal charging current difference is extracted from the current data in the cth current interval using formula (4). and as the fourth health indicator; (3) (4) In formulas (3) and (4), is the current value of the ath voltage interval in the constant current charging process; is the voltage value changing with time t in the ath voltage interval during the constant current charging process; and Indicates the start time and end time of the a-th voltage interval; is the voltage value in the cth current interval during the constant voltage charging process; is the current value changing with time t in the cth current interval of the constant voltage charging process; and Indicates the start time and end time of the cth current interval; Step 1.2: Use the ergodic method combined with the Pearson correlation coefficient to find the optimal voltage range and the optimal current range; Step 1.2.1: Calculate the first health index of the ath voltage interval using formula (5): Its battery health status Pearson correlation coefficient between ; Thus, the Pearson correlation coefficient between the first health index and the battery health state SOH of all voltage intervals is obtained, and the voltage interval corresponding to the maximum first health index is selected as the suboptimal voltage interval; Calculate the third health index of the bth voltage interval using formula (6): Its battery health status Pearson correlation coefficient between ; Thus, the Pearson correlation coefficient between the third health index and the battery health status SOH of all voltage intervals is obtained, and the voltage interval corresponding to the maximum third health index is selected as another suboptimal voltage interval; (5) (6) In formulas (5) and (6), E( ) is the mean function, E 2 ( ) is the square of the mean function; Step 1.2.2: Calculate the second health index of the cth current interval using formula (7): Its battery health status Pearson correlation coefficient between , thereby obtaining the Pearson correlation coefficient between the second health index and its battery health status SOH in all current intervals, and selecting the voltage interval corresponding to the maximum second health index as another suboptimal current interval; Calculate the fourth health index of the dth current interval using formula (8): Its battery health status Pearson correlation coefficient between ; Thus, the Pearson correlation coefficient between the fourth health index and the battery health state SOH of all current intervals is obtained, and the current interval corresponding to the maximum fourth health index is selected as another suboptimal current interval; (7) (8) Step 1.2.3: Select the voltage interval corresponding to the larger absolute value from the two suboptimal voltage intervals as the optimal voltage interval; select the current interval corresponding to the larger absolute value from the suboptimal current voltage intervals as the optimal current interval; Step 2: Use Box-Cox transform and discrete wavelet packet transform to optimize the health indicators of lithium-ion batteries: Step 2.1, using a measuring device to obtain online current data of the lithium-ion battery, and performing real-time processing based on the online current data to obtain the second health indicator and the fourth health indicator of the full cycle; Step 2.2: Normalize the health index of the optimal current range, the battery health status SOH, and the health index of the entire cycle to obtain the second health index of the normalized optimal current range. and the fourth health indicator , corresponding battery health status 、 And the normalized second health indicator of the whole cycle and the fourth health indicator ; Step 2.3: Use equations (9) to (12) to and as well as and Perform Box-Cox transformation to obtain the second health index of the optimal current interval after transformation , the second health indicator of the whole cycle after transformation , the fourth health indicator of the optimal current range after transformation And the fourth health indicator of the whole cycle after transformation : (9) (10) (11) (12) In formula (9-12), c is a constant that ensures that the transformed data is positive; 、 are 2 transformation parameters; Step 2.4: Perform discrete wavelet packet decomposition on the lithium-ion battery health index to obtain the reconstructed overall health index. Wavelet packet tree; Step 3: SOH estimation based on the improved GPR model; Step 3.1, define the health index of lithium-ion battery training THIG = { 、 、 、 },in, 、 Represents the health indicator after reconstruction The wavelet packet tree of the second health indicator and the wavelet packet tree of the fourth health indicator in the wavelet packet tree, and define their corresponding health states 、 ; Step 3.2: Take THIG as the input of GPR model and transform the health status SOH corresponding to THIG into { 、 、 、 } is used as the output of the GPR model, and the SE kernel function of the GPR model is constructed using formula (13) : (13) In formula (13), 、 is the mth and nth health index in the training health index THIG, and m=1,2,3,4, n=1,2,3,4, m≠n, It is the scale parameter that controls the decay speed of the kernel function; Step 3.3: Use the improved particle swarm optimization algorithm to optimize the SE kernel parameters of the GPR model. Optimize to get the optimal scale parameter , and used to construct the optimal GPR model for SOH estimation; Step 3.4: Use the measuring device to obtain the online voltage and current data of the lithium-ion battery and process them to obtain the online health index. The online health index is input into the optimal GPR model to obtain the optimal SOH estimation result.

2. The lithium-ion battery health status estimation method based on the IPSO-GPR algorithm according to claim 1, characterized in that: The step 2.4 includes: Step 2.4.1, define the overall health index of lithium-ion batteries THI = { 、 、 、 }, and perform 3-layer discrete wavelet packet decomposition on the overall health index THI to obtain the wavelet packet tree of the overall health index THI; among them, the overall health index set after the 3rd layer decomposition is recorded as ={ , ,…, ,…, };in, represents the i-th overall health indicator after the third-level decomposition; Step 2.4.2: Define the overall health status of the lithium-ion battery ={ , }, and Perform 3-layer discrete wavelet packet decomposition to obtain the overall health status The wavelet packet tree of ; among them, the overall health status set after the third level decomposition is recorded as ={ , ,…, ,…, },in, represents the i-th overall health status after the third-level decomposition; Step 2.4.3, calculate separately and Pearson correlation coefficient between ; Step 2.4.4, based on The terminal nodes of the wavelet packet tree for the overall health index THI The corresponding wavelet packet coefficients are corrected to obtain the terminal nodes of the corrected wavelet packet tree ; Step 2.4.5: The terminal nodes of the modified wavelet packet tree Perform discrete wavelet packet reconstruction to obtain the overall health index after reconstruction Wavelet packet tree.

3. The lithium-ion battery health status estimation method based on the IPSO-GPR algorithm according to claim 2, characterized in that: The step 3.3 includes: Step 3.3.

1. Set the scale parameter of the GPR model As the position vector of the particle, and set the population size of the particle swarm to P and the maximum number of iterations to t max ; Define the current number of iterations as t and initialize t=1; Step 3.3.2: Randomly initialize the velocity of the pth particle in the tth generation particle swarm and location ; Step 3.3.3: Calculate the diversity of the particle swarm of the tth generation using formula (14) : (14) In formula (14), is the longest diagonal length of the search space, represents the d-th dimension position of the p-th particle in the t-th generation particle swarm; is the average value of all particles in the t-th generation particle swarm at the d-th dimension position; P is the population size; D is the dimension of the particle; Step 3.3.4: Use formula (15) to calculate the relationship weight w of the t-th generation particle swarm t ; (15) In formula (15), d low and d high are the low and high thresholds of population diversity respectively; w min and w max are the low and high thresholds of the relationship weights between populations, respectively; is the relationship weight of the t-1 generation particle swarm; when t=1, let , Indicates a fixed value; Step 3.3.5: Use the root mean square error as the fitness function of IPSO to calculate the fitness of the pth particle in the tth generation particle swarm. , thus obtaining the local optimal solution of the p-th particle in the t-th generation , select the optimal solution corresponding to the maximum fitness value from the local optimal solutions of all particles in the tth generation as the global optimal position of the tth generation ; Step 3.3.6: Use formula (16) to get the position of the pth particle in the t+1th generation particle swarm and speed : (16) In formula (20), 、 are two learning factors, 、 are two random functions with values ​​ranging from [0,1]; Step 3.3.7: After assigning t+1 to t, determine whether t>t max Is it true? If so, output the tth max Global optimal position of the generation population and as the optimal scale parameter , and used to construct the optimal GPR model for SOH estimation; otherwise, go to step 3.3.2 to continue optimization.

4. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the lithium-ion battery health status estimation method according to any one of claims 1 to 3, and the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the lithium-ion battery health status estimation method according to any one of claims 1 to 3 are executed.

Images