Lithium ion battery health state estimation method based on SVD and improved GMDH

Abstract

The invention provides a lithium ion battery health state estimation method based on SVD and improved GMDH, and belongs to the technical field of lithium ion batteries. The technical problem that feature extraction is easily influenced by human experience and the technical problem that a grouped data processing method is hyper-parameter configured are solved. According to the technical scheme, the method comprises the following steps of 1) extracting health index feature data of a lithium ion battery degradation data set by using a singular value decomposition method; 2) searching optimal hyper-parameter configuration of the grouped data processing method by using a particle swarm optimization algorithm; and 3) estimating the health state of the lithium ion battery by using an improved grouped data processing method. The beneficial effects of the invention are that the estimation method provided by the invention can effectively avoid the influence of human experience in the feature extraction process, and the improved grouped data processing method is high in estimation precision and good in robustness.

Description

Technical Field

[0001] The invention relates to the technical field of lithium ion batteries, and in particular to a lithium ion battery health state estimation method based on SVD and improved GMDH. Background Art

[0002] As an important energy storage device, lithium-ion batteries are widely used in many fields such as electric vehicles and portable electronic products due to their advantages such as long life, high power and wide operating temperature. Lithium-ion batteries will gradually age during use, which is a nonlinear process. Accurate state of health (SOH) estimation is of great significance to the life and safe use of lithium-ion batteries. Lithium-ion battery SOH estimation is one of the core tasks of battery management systems.

[0003] Data-driven methods are a hot topic in the current research of SOH estimation methods. Data-driven methods mainly extract feature data that can characterize the battery SOH from the historical data of lithium-ion batteries, and then use data-driven algorithms to estimate SOH based on the extracted feature data. For example, in the article "State of health estimation of lithium-ion batteries based on multi-health features extraction and improved long short-term memory neural network", Peng et al. proposed a lithium-ion battery health state estimation method based on multi-health feature extraction and improved long short-term memory neural network, and introduced a quantum particle swarm optimization algorithm to solve the problem of neural network hyperparameter selection. In the article "Robust state of health estimation of lithium-ion batteries using convolutional neural network and random forest", Yang et al. used convolutional neural network and random forest algorithm to realize the estimation of SOH of partial discharge of batteries under different initial and final charge states. However, the above data-driven algorithms often have the defects of long running time and manual adjustment of network structure. It is an important issue to estimate the SOH of lithium-ion batteries using data-driven algorithms with simple calculation, adaptive adjustment of network structure, higher accuracy and better robustness.

[0004] How to solve the above technical problems is the subject faced by the present invention. Summary of the invention

[0005] The purpose of the present invention is to provide a lithium-ion battery health state estimation method based on SVD and improved GMDH. The estimation method adopts the SVD method for data feature extraction, effectively reduces the complexity of feature extraction, and avoids the influence of human experience. The estimation method introduces a particle swarm optimization PSO algorithm to search for the optimal hyperparameters of GMDH on the basis of extracting feature data using SVD, and finally obtains the improved GMDH. The improved GMDH is used to estimate the SOH of lithium-ion batteries, which has the advantages of simple calculation, high accuracy, good robustness, etc.

[0006] The present invention is achieved by the following measures: A method for estimating the health state of a lithium-ion battery based on SVD and improved GMDH, comprising the following steps:

[0007] Step 1) extracting health indicator feature data of lithium-ion battery degradation data set using singular value decomposition method;

[0008] Step 2) using a particle swarm optimization algorithm to search for an optimal hyperparameter configuration of the grouped data processing method, and setting it as a hyperparameter of the grouped data processing method;

[0009] Step 3) Estimate the health status of the lithium-ion battery using the improved grouped data processing method.

[0010] As a further optimization scheme of the lithium ion battery health state estimation method based on SVD and improved GMDH provided by the present invention, the specific steps of step 1) are as follows:

[0011] Step 1-1) Define the health status of the lithium-ion battery as follows:

[0012]

[0013] Where SOH is the health status of the lithium-ion battery, C r is the rated capacity of the lithium-ion battery, C a It is the actual capacity of the lithium-ion battery after aging.

[0014] Step 1-2) Obtain degradation data such as voltage, charge and discharge time, and temperature of the lithium-ion battery during multiple charge and discharge cycles.

[0015] Step 1-3) In practical applications, the discharge state of lithium-ion batteries is usually uncontrollable. For example, during the actual driving process of electric vehicles, the discharge requirements of the batteries are different due to the different road conditions and required output power at each moment. Compared with the discharge state of lithium-ion batteries, the charging state is easier to control. Based on the practical application characteristics, the present invention extracts the charging data of lithium-ion batteries.

[0016] Step 1-4) Use SVD to extract the health indicators of each parameter. Taking the charging voltage data as an example, the voltage data of each charging cycle is extracted in turn and regarded as a single-column matrix, Z j is the charging voltage data of the jth aging characteristic test, where j = 1, 2, ..., n, and n is the total number of aging characteristic tests. j Perform SVD, since Z j It is a single-column matrix. According to the principle of SVD, for each aging test voltage data Z j A singular value υ can be obtained to represent its characteristic information i Finally, the singular values ​​obtained from all aging tests are arranged in order to form the voltage health index HI V Similarly, we can get the temperature health index HI T . Health indicator HI for charging time t Calculated from the charging start and end time in the aging data.

[0017] As a further optimization scheme of the lithium ion battery health state estimation method based on SVD and improved GMDH provided by the present invention, the specific steps of step 2) are as follows:

[0018] Since in the grouped data processing method, key parameters such as the maximum number of neurons in the network layer, the maximum number of network layers, the selection pressure coefficient, and the learning rate will affect the estimation accuracy of the algorithm, the present invention uses a particle swarm optimization (PSO) algorithm to search for the optimal values ​​of the above key parameters to improve the hyperparameter configuration of GMDH, thereby improving the estimation accuracy of GMDH.

[0019] Step 2-1) divides the health indicator data such as charging voltage, charging time and temperature extracted in step 1-4) into a training set and a test set.

[0020] Step 2-2) Initialize the PSO algorithm and set the appropriate particle search space range and algorithm stop conditions.

[0021] Step 2-3) Combine the PSO algorithm and GMDH to build a PSO-GMDH estimation model. GMDH is a polynomial neural network that uses a recursive hierarchical structure to gradually build higher-order feature representations through adaptive polynomial functions, thereby realizing the nonlinear modeling capability of the model. The GMDH polynomial neural network will autonomously select two variables from all variables in the current layer as the input of a candidate model in the next layer. The SOH of lithium-ion batteries can be estimated by a series of partial quadratic polynomials containing two variables, as shown below:

[0022]

[0023] in, is the estimated health status of the lithium-ion battery obtained by GMDH, HIs is the feature input vector, f is the health status estimation function of the lithium-ion battery, and f (out) For connection The function of the last hidden layer, f (i) is the function connecting the hidden layer i and i-1, f (1) is the function connecting the first hidden layer and the input layer, is the variable selected in the i-th hidden layer; f (out) For example, the specific form of the partial quadratic polynomial is as follows:

[0024]

[0025] in and are the coefficients of the partial quadratic polynomial, which can be obtained by minimizing the mean square error (MSE) using the least squares algorithm.

[0026] Step 2-4) Use the PSO-GMDH estimation model to estimate the SOH of the lithium-ion battery for the training set data, and use the mean square error of the estimation result as the fitness value of the PSO.

[0027] Step 2-5) When the algorithm iteration ends, the key parameters corresponding to the minimum fitness value are used as the optimal hyperparameters of GMDH.

[0028] As a further optimization scheme of the lithium ion battery health state estimation method based on SVD and improved GMDH provided by the present invention, the specific steps of step 3) are as follows:

[0029] Step 3-1) configures the hyperparameters of GMDH to the optimal hyperparameters obtained in step 2-4).

[0030] Step 3-2) Use the improved GMDH to estimate the SOH of the lithium-ion battery based on the test set data.

[0031] Step 3-3) Analyze the obtained estimation results.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] (1) The present invention uses a singular value decomposition method when extracting features from degradation data of lithium-ion batteries. The size of the singular value is closely related to the important information hidden in the original degradation data. Compared with methods such as extracting data at a specific time point specified by humans or extracting geometric characteristic values ​​of degradation data curves, this feature extraction method requires less work, effectively reduces the complexity of feature extraction, and also effectively avoids the influence of human experience. The extracted features can well characterize the health status of lithium-ion batteries.

[0034] (2) The present invention adopts an improved grouped data processing method when estimating the SOH of lithium-ion batteries. The network hyperparameters of GMDH are configured as the optimal hyperparameters searched by the PSO algorithm, which effectively solves the hyperparameter selection problem of GMDH. Compared with GMDH that has not been optimized by the PSO algorithm, the improved GMDH has significantly improved accuracy when estimating the SOH of lithium-ion batteries. This estimation method can not only achieve high-precision estimation of the subsequent SOH value based on the historical data of a single battery, but also achieve high-precision estimation of the SOH value of the same type of batteries based on the historical data of multiple batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0036] Figure 1 The overall framework diagram of the lithium-ion battery SOH estimation method based on SVD and improved GMDH provided by the present invention.

[0037] Figure 2 The GMDH structure diagram with 5 hidden layers provided by the present invention.

[0038] Figure 3 This is a fitness function convergence curve diagram of the PSO in Example 1 provided by the present invention.

[0039] Figure 4 This is a lithium-ion battery SOH estimation curve diagram based on single battery historical data provided by the present invention.

[0040] Figure 5 This is a lithium-ion battery SOH estimation error curve diagram based on single battery historical data provided by the present invention.

[0041] Figure 6 This is a fitness function convergence curve diagram of PSO in Example 2 provided by the present invention.

[0042] Figure 7 This is a lithium-ion battery SOH estimation curve diagram based on multi-battery historical data provided by the present invention.

[0043] Figure 8 This is a lithium-ion battery SOH estimation error curve diagram based on multi-battery historical data provided by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0045] Example 1

[0046] This embodiment is based on the long-term aging test data of Cell 1 in the Oxford University battery degradation dataset, and uses the proposed method to estimate the subsequent SOH based on the historical data of Cell 1 single battery. Cell 1 is a 740mAh lithium-ion soft-pack battery, which is charged in a 40°C constant temperature box using a constant current and constant voltage charging method. The constant current used in the constant current charging stage is 740mA, and the charging cut-off voltage is 4.2V. The discharge cut-off voltage in the discharge stage is 2.7V. After every 100 charge and discharge cycles, the battery is subjected to an aging characteristic test. The overall framework diagram of the method proposed in this embodiment is shown in the figure. Figure 1 shown.

[0047] See also Figures 1 to 5 This embodiment provides a lithium-ion battery health state estimation method based on SVD and improved GMDH, comprising the following steps:

[0048] Step 1) extracting health indicator feature data of lithium-ion battery degradation data set using singular value decomposition method;

[0049] Step 2) using a particle swarm optimization algorithm to search for an optimal hyperparameter configuration of the grouped data processing method, and setting it as a hyperparameter of the grouped data processing method;

[0050] Step 3) Estimate the health status of the lithium-ion battery using the improved grouped data processing method.

[0051] Specifically, step 1) includes the following steps:

[0052] Step 1-1) Define the health status of the lithium-ion battery as follows:

[0053]

[0054] Where SOH is the health status of the lithium-ion battery, C r is the rated capacity of the lithium-ion battery, C a It is the actual capacity of the lithium-ion battery after aging.

[0055] Step 1-2) Obtain the long-term aging test data of Cell 1 in the Oxford University battery degradation data set, specifically including the voltage, charge and discharge time and temperature degradation data of multiple charge and discharge cycles of Cell 1.

[0056] Step 1-3) Based on the actual application characteristics, this embodiment extracts the charging data of Cell 1.

[0057] Step 1-4) Use SVD to extract the health indicators of each parameter. Taking the charging voltage data of Cell 1 as an example, extract the voltage data of each charging cycle in turn and regard it as a single-column matrix, Z j (j=1,2,…,n) is the charging voltage data of the jth aging characteristic test, and n is the total number of aging characteristic tests. j Perform SVD, since Z j It is a single-column matrix. According to the principle of SVD, for each aging test voltage data Z j A singular value υ can be obtained to represent its characteristic information i Finally, the singular values ​​obtained from all aging tests are arranged in order to form the voltage health index HI V Similarly, we can get the temperature health index HI T . Health indicator HI for charging time t Calculated from the charging start and end time in the aging data.

[0058] Specifically, step 2) includes the following steps:

[0059] Since in the grouped data processing method, key parameters such as the maximum number of neurons in the network layer, the maximum number of network layers, the selection pressure coefficient, and the learning rate will affect the estimation accuracy of the algorithm. The present invention uses a particle swarm optimization (PSO) algorithm to search for the optimal values ​​of the above four key parameters to improve the hyperparameter configuration of GMDH, thereby improving the estimation accuracy of GMDH.

[0060] Step 2-1) divides the first 70% of the health indicator data such as charging voltage, charging time and temperature extracted in step 1-4) into a training set, and the remaining 30% into a test set.

[0061] Step 2-2) Initialize the PSO algorithm, set the particle swarm population space dimension to 4, the scale N to 50, and the maximum number of iterations D to 50, that is, when the number of iterations reaches 50, the algorithm stops running. The range of the particle search space is: α∈[10,30], β∈[3,10], γ∈[0.1,0.9], δ∈[0.1,0.9], where α is the maximum number of neurons in the network layer, β is the maximum number of network layers, γ is the selection pressure coefficient, and δ is the learning rate. The GMDH hyperparameter configuration without PSO optimization is α=20, β=5, γ∈0.5, δ=0.7.

[0062] Step 2-3) Combine the PSO algorithm and GMDH to build a PSO-GMDH estimation model. GMDH is a polynomial neural network that uses a recursive hierarchical structure to gradually build higher-order feature representations through adaptive polynomial functions, thereby realizing the nonlinear modeling capability of the model. The GMDH polynomial neural network will autonomously select two variables from all the variables in the current layer as the input of a candidate model in the next layer. The SOH of the lithium-ion battery can be estimated by a series of partial quadratic polynomials containing two variables. When the number of hidden layers of the GMDH polynomial neural network is 5, the structure diagram of GMDH is as follows: Figure 2 As shown, the partial quadratic polynomial is as follows:

[0063]

[0064] in, is the estimated health status of the lithium-ion battery obtained by GMDH, HIs is the feature input vector, f is the health status estimation function of the lithium-ion battery, and f (out) For connection The function of the last hidden layer, f (i) is the function connecting the hidden layer i and i-1, f (1) is the function connecting the first hidden layer and the input layer, is the variable selected in the i-th hidden layer. (out) For example, the specific form of the partial quadratic polynomial is as follows:

[0065]

[0066] in, are the coefficients of the partial quadratic polynomial, which are obtained by minimizing the mean square error (MSE) using the least squares algorithm.

[0067] Step 2-4) Use the PSO-GMDH estimation model to estimate the SOH of the lithium-ion battery for the training set data, and use the mean square error of the estimation result as the fitness value of the PSO. The fitness function convergence curve of the PSO in this embodiment is as follows: Figure 3As shown, the minimum fitness value is 3.7359×10 -7 , the number of iterations when the algorithm fitness value converges is 12.

[0068] Step 2-5) When the algorithm iteration ends, the four key parameters corresponding to the minimum fitness value, namely α=22, β=4, γ∈0.719953, δ=0.815986, are used as the optimal hyperparameters of GMDH.

[0069] Specifically, step 3) includes the following steps:

[0070] Step 3-1) configures the four hyperparameters of GMDH to the optimal hyperparameters obtained in step 2-4).

[0071] Step 3-2) Use the improved GMDH to estimate the SOH of the lithium-ion battery based on the test set data.

[0072] Step 3-3) Get the SOH estimation curve of Cell 1 as follows Figure 4 As shown in Figure 2, the SOH estimation error curve of Cell 1 is as follows: Figure 5 See Figure 4 and Figure 5 It can be found that compared with the GMDH without PSO optimization, the improved GMDH has higher accuracy in estimating the subsequent SOH based on the historical data of the single battery, and the estimated curve is closer to the real SOH curve. The estimation error of the improved GMDH is less than 0.25%, and the error is still low in the final SOH estimation. The estimation error of the GMDH without PSO optimization has a trend of gradually increasing. The estimation performance of the improved GMDH has obvious advantages.

[0073] This embodiment proves that the lithium-ion battery SOH estimation method based on SVD and improved GMDH has good estimation performance and high estimation accuracy when estimating the subsequent SOH of a single battery using its historical data.

[0074] Example 2

[0075] This embodiment is based on the long-term aging test data of Cell 5, Cell 6, Cell 7 and Cell 8 in the Oxford University battery degradation data set, and uses the proposed method to estimate the SOH of Cell 8 based on the historical data of multiple batteries of Cell 5, Cell 6 and Cell 7. The above 4 batteries are all 740mAh lithium-ion soft-pack batteries, and the other test conditions and operating conditions are the same as those in Example 1.

[0076] See also Figures 6 to 8 This embodiment provides a lithium-ion battery health state estimation method based on SVD and improved GMDH, comprising the following steps:

[0077] Step 1) extracting health indicator feature data of lithium-ion battery degradation data set using singular value decomposition method;

[0078] Step 2) using a particle swarm optimization algorithm to search for an optimal hyperparameter configuration of the grouped data processing method, and setting it as a hyperparameter of the grouped data processing method;

[0079] Step 3) Estimate the health status of the lithium-ion battery using the improved grouped data processing method.

[0080] Specifically, step 1) includes the following steps:

[0081] Step 1-1) is the same as that in Example 1 above.

[0082] Step 1-2) Obtain the long-term aging test data of Cell 5, Cell 6, Cell 7 and Cell 8 in the Oxford University battery degradation data set, specifically including the voltage, charge and discharge time and temperature degradation data of multiple charge and discharge cycles of the above four batteries.

[0083] Step 1-3) Based on the actual application characteristics, this embodiment extracts the charging data of the above four batteries.

[0084] Step 1-4) SVD is used to extract the health index of each parameter of each battery. The principle is the same as that of the above embodiment 1.

[0085] Specifically, step 2) includes the following steps:

[0086] Step 2-1) divides the health indicator data of Cell 5, Cell 6, and Cell 7 extracted in step 1-4) into a training set, and divides the health indicator data of Cell 8 into a test set.

[0087] Step 2-2) to step 2-4) are the same as those in the above-mentioned embodiment 1.

[0088] The fitness function convergence curve of PSO in this embodiment is as follows: Figure 6 As shown. The minimum fitness value is 5.3790×10 -5 , the number of iterations when the algorithm fitness value converges is 6.

[0089] Step 2-5) When the algorithm iteration ends, the four key parameters corresponding to the minimum fitness value, namely α=17, β=5, γ∈0.402582, δ=0.708038, are used as the optimal hyperparameters of GMDH.

[0090] Specifically, step 3) includes the following steps:

[0091] Step 3-1) to step 3-2) are the same as those in the above-mentioned embodiment 1.

[0092] Step 3-3) Get the SOH estimation curve of Cell 8 as follows Figure 7 As shown in Figure 2, the SOH estimation error curve of Cell 8 is as follows: Figure 8 See Figure 7 and Figure 8 It can be found that compared with the GMDH without PSO optimization, the improved GMDH has higher accuracy in estimating the SOH of Cell 8 based on the historical data of multiple batteries of Cell 5, Cell 6, and Cell 7, and the estimated curve is closer to the real SOH curve. The estimation error of the improved GMDH is mostly less than 0.5%, and the error is still low in the final SOH estimation.

[0093] This embodiment proves that the lithium-ion battery SOH estimation method based on SVD and improved GMDH has good estimation performance and high estimation accuracy when estimating the SOH of batteries of the same model using multi-battery historical data.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A lithium-ion battery health state estimation method based on SVD and improved GMDH, characterized in that: The following steps are involved: Step 1) extracting health indicator feature data of lithium-ion battery degradation data set using singular value decomposition method; Step 2) using a particle swarm optimization algorithm to search for an optimal hyperparameter configuration of the grouped data processing method, and setting it as a hyperparameter of the grouped data processing method; Step 3) Estimate the health status of the lithium-ion battery using the improved grouped data processing method.

2. The lithium-ion battery health state estimation method based on SVD and improved GMDH according to claim 1, characterized in that: The step 1) comprises the following steps: Step 1-1) defining the health state SOH of a lithium-ion battery as the ratio of the actual capacity of the battery to the rated capacity of the battery, and obtaining the voltage, charge and discharge time and temperature degradation data of the lithium-ion battery after multiple charge and discharge cycles; The health status of lithium-ion batteries is defined as follows: Among them, SOH is the health status of lithium-ion batteries, C r is the rated capacity of the lithium-ion battery, C a is the actual capacity of the lithium-ion battery after aging; Step 1-2) extracting various parameter data of the charging stage; Step 1-3) Use SVD to extract the health index of each parameter, extract the voltage data of each charging cycle in turn and treat it as a single-column matrix, Z j is the charging voltage data of the jth aging characteristic test, where j = 1, 2, ..., n, n is the total number of aging characteristic tests, respectively for Z j Perform SVD, since Z j It is a single-column matrix. According to the principle of SVD, for each aging test voltage data Z j Get a singular value υ representing its characteristic information i ; Step 1-4) The singular values ​​obtained from all aging tests are arranged in order to form the voltage health index HI V , get the temperature health index HI T , the health indicator HI of charging time t Calculated from the charging start and end time in the aging data.

3. The lithium-ion battery health state estimation method based on SVD and improved GMDH according to claim 1, characterized in that: The step 2) comprises the following steps: Step 2-1) Divide the extracted health indicator data such as charging voltage, charging time and temperature into a training set and a test set; Step 2-2) Initialize the PSO algorithm and set the appropriate particle search space range and algorithm stop condition; Step 2-3) Combine the PSO algorithm and GMDH to build a PSO-GMDH estimation model, use the estimation model to estimate the SOH of the lithium-ion battery for the training set data, and use the mean square error of the estimation result as the fitness value of the PSO; The GMDH polynomial neural network autonomously selects two variables from all the variables in the current layer as the input of a candidate model in the next layer. The SOH of the lithium-ion battery is estimated by a series of partial quadratic polynomials containing two variables, as shown below: in, is the estimated health status of the lithium-ion battery obtained by GMDH, HIs is the feature input vector, f is the health status estimation function of the lithium-ion battery, and f (out) For connection The function of the last hidden layer, f (i) is the function connecting the hidden layer i and i-1, f (1) is the function connecting the first hidden layer and the input layer, is the variable selected in the i-th hidden layer, f (out) The partial quadratic polynomial of is of the following form: in, and are the coefficients of the partial quadratic polynomial, obtained by minimizing the mean square error MSE using the least squares algorithm; Step 2-4) When the algorithm iteration ends, the key parameter corresponding to the minimum fitness value is used as the optimal hyperparameter of GMDH.

4. The lithium-ion battery health state estimation method based on SVD and improved GMDH according to claim 1, characterized in that: The step 3) comprises the following steps: Step 3-1) configure the hyperparameters of GMDH to the optimal hyperparameters obtained in step 2); Step 3-2) Using the improved GMDH to estimate the SOH of the lithium-ion battery for the test set data; Step 3-3) Analyze the obtained estimation results.

Images