A method for estimating the health status of lithium batteries

Abstract

The present invention provides a lithium battery health state estimation method, comprising the following steps: A, establishing a second-order RC equivalent circuit model, and transforming the second-order RC equivalent circuit model equation into a mathematical form that can be recognized by a computer; B, converting the mathematical form into a least squares format to calculate various parameters to be determined; C, preprocessing the surface temperature data in the Oxford battery aging data set; D, obtaining a characteristic curve DTV according to the preprocessed surface temperature data T and the battery terminal voltage V in the Oxford battery aging data set. m , and obtain the characteristic curve DTV m The valley value P1, the peak value P2, and the voltage difference ΔP between the valley value P1 and the peak value P2; E, using a regression support vector machine to complete the mapping from the model parameters of the second-order RC equivalent circuit model identified in step B and the valley value P1, the peak value P2, and the voltage difference ΔP obtained in step D to the battery health state, and establish a lithium battery health state model. The present invention does not require a large number of experiments, saves calculation costs, and effectively improves estimation accuracy.

Description

Technical Field

[0001] The invention relates to a method for estimating the health state of a lithium battery. Background Art

[0002] The battery management system is one of the most important components of new energy vehicles. As the number of charge and discharge cycles increases, lithium batteries continue to age, and their operating efficiency and system safety also decrease. The state of health (SOH) of lithium batteries is an important indicator of battery aging and an important indicator of the operating life of new energy vehicles. SOH cannot be measured directly, and is often estimated using easily accessible external parameters such as voltage and temperature.

[0003] The commonly used SOH definition is: ,in is the maximum capacity in the current charging cycle, The initial nominal capacity. The initial SOH of a new battery is 100%. When the capacity decays to a threshold of 80% of the initial capacity, the battery is considered to be failed. Under laboratory conditions, the battery can be obtained by charging the battery with constant current and constant voltage and then discharging it with constant current. , the battery SOH is calculated, but it is difficult to estimate the battery health status online in the actual battery management system (BMS).

[0004] At present, battery SOH estimation can be roughly divided into two methods: model-based and data-driven. The model-based SOH estimation method starts from the mechanism of battery aging, identifies model parameters of external parameters such as physical and electrochemical models by measuring lithium battery signals, and then estimates the current battery aging degree based on the model. This method has good real-time performance and strong adaptability, but it is highly dependent on the accuracy of the model and has a high computational cost.

[0005] The data-driven method requires a large amount of experimental data as prior knowledge to establish and train a direct mapping relationship model between current, voltage, temperature and other data and battery SOH. This method does not require an in-depth understanding of the internal electrochemical mechanism of lithium batteries and is easier to implement for nonlinear problems, but the training data used needs to fully reflect the battery characteristics, otherwise it is easy to cause overfitting of the model. Therefore, the purely data-driven lithium battery state estimation method has a high algorithm complexity and a high degree of dependence on the rationality of the training data selection. Summary of the invention

[0006] The present invention proposes a lithium battery health state estimation method, which combines an equivalent model and a support vector machine and uses the Oxford battery aging dataset to identify model parameters. This method does not require a large number of experiments, saves computing costs, and effectively improves estimation accuracy.

[0007] The present invention is achieved through the following technical solutions:

[0008] A method for estimating the health status of a lithium battery comprises the following steps:

[0009] A. Establish a second-order RC equivalent circuit model and transform the second-order RC equivalent circuit model equation into a mathematical form that can be recognized by a computer. The mathematical form involves the terminal voltage, open circuit voltage, current and parameters to be determined of the battery. The terminal voltage, current and open circuit voltage data are obtained through the Oxford battery aging data set. The parameters to be determined involve the parameters of the second-order RC equivalent circuit model.

[0010] B. Convert the mathematical form described in step A into a least squares format, thereby calculating the parameters to be determined in step A, and then completing the identification of the model parameters of the second-order RC equivalent circuit model;

[0011] C. Use the robust local weighted regression algorithm to preprocess the surface temperature data in the Oxford battery aging dataset to remove outliers in the data;

[0012] D. The surface temperature data pre-processed in step C The battery terminal voltage V in the Oxford battery aging dataset is extracted by the following formula to obtain the characteristic curve DTVm: , and obtain the valley value P1, peak value P2, and voltage difference between valley value P1 and peak value P2 of characteristic curve DTVm , where L is the sampling interval;

[0013] E. Use regression support vector machine to complete the model parameters of the second-order RC equivalent circuit model identified in step B and the valley value P1, peak value P2, voltage difference obtained in step D Mapping to battery health status and establishing a lithium battery health status model.

[0014] Furthermore, in step A, the second-order RC equivalent circuit model equation is:

[0015] ,in, U 1. U 2 is the initial polarization voltage of the lithium battery, C 1. C 2 is the polarization capacitance of the lithium battery, R 1. R 2 is the polarization internal resistance of the lithium battery, U Indicates the terminal voltage of the lithium battery, U OC Indicates the open circuit voltage of the lithium battery. R 0 means ohmic internal resistance, I Indicates current.

[0016] Furthermore, the mathematical form that can be recognized by a computer is: ,in, k 1. k 2. k 3. k 4. k 5 is the parameter to be sought.

[0017] Furthermore, in the mathematical form, the lithium battery terminal voltage U , Current I Directly use the terminal voltage and current data in the Oxford battery aging dataset and use the following formula Get open circuit voltage data, where V 0 is the initial charging value selected based on the terminal voltage in the Oxford battery aging data set. N Increasing number of samples for constant current charging of the Oxford battery aging dataset, R is any number in the voltage range selected according to the charging voltage change.

[0018] Furthermore, the Oxford battery aging data set was obtained by performing aging tests on eight lithium batteries with a nominal capacity of 740 mAh. The aging test process was: a 1C charge and discharge cycle was performed using a constant current-constant voltage method, and the MPG-205 battery tester was used to measure and record the lithium battery current, voltage and surface temperature every 100 aging cycles.

[0019] Furthermore, in the robust local weighted regression algorithm of step C, for the surface temperature data t The time is calculated according to the following formula t Temperature fitting value at time : ,in, T i is the sampling time of the surface temperature data t i The corresponding temperature value, N Indicates t is the width of the sampling interval selected as the center, K Represents the kernel function, which represents the relationship between position and weight. After calculating the temperature fitting value, the first residual is calculated according to the following formula , and calculate the improved weights based on the first residual , in which S is the median value of the first residual absolute value sequence, , and finally the weights will be improved With kernel function K The weights in are multiplied as the final weights, and the new .

[0020] Furthermore, the sampling interval in step D is L The value range is [15,25].

[0021] Furthermore, the step E specifically includes:

[0022] E1, model parameters and valley value of the second-order RC equivalent circuit model P 1. Peak P 2. Voltage difference , create a training set: ,in, Indicates i The input feature vector of samples, Indicates the health status value of the lithium battery corresponding to the second-order RC equivalent circuit model, n Indicates the number of training set samples;

[0023] E2. Using radial basis kernel function , map the training set to a high-dimensional space and obtain the sample set ;

[0024] E3. Use the following first objective function: ,in, For the i The predicted value of samples, is the corresponding true value, and the regression function is obtained by minimizing the first objective function The weight vector in and bias .

[0025] Furthermore, the step B specifically includes:

[0026] B1. Convert the mathematical form described in step A into the least squares format: ,in, , then k The estimated output of subsampling is ;

[0027] B2. Define the second residual , and set the second objective function : ,in, is the weighting coefficient, ;

[0028] B3, use the following formula to obtain Smallest : , and the obtained Feedback to the least squares format to calculate the parameters to be determined in step A.

[0029] Furthermore, the method further comprises the following steps:

[0030] F. Use grid search to jointly optimize the hyperparameters of the support vector machine.

[0031] The present invention has the following beneficial effects:

[0032] 1. The present invention first establishes a second-order RC equivalent circuit model to characterize the battery aging behavior, and then transforms the model equation into a mathematical form that can be recognized by a computer. The terminal voltage, current and open circuit voltage data are obtained through the Oxford battery aging data set, and the parameters to be determined in the mathematical form are obtained by combining the least squares method, thereby completing the identification of the parameters of the second-order RC equivalent circuit model. The battery surface temperature data and the battery terminal voltage in the Oxford battery aging data set are then used to obtain the characteristic curve. Finally, the regression support vector machine is used to complete the mapping of the model parameters, the valley value, the peak value and the voltage difference between the peak and valley values ​​of the characteristic curve to the health status of the lithium battery, and a lithium battery set health status model is established. Combining the equivalent model and the support vector machine, it can The algorithm complexity is effectively reduced and the estimation accuracy is improved. The second-order RC equivalent circuit model can simulate the dynamic and static characteristics of the battery more accurately and is more applicable in practical engineering. By transforming the model equation into a mathematical form and then using the Oxford battery aging data set combined with the least squares method to identify the model parameters, there is no need to conduct a large number of experiments to identify the model parameters as in the prior art, thus saving computing and time costs. Due to limitations in sensor accuracy or measurement conditions, there are many outliers in the battery surface temperature data in the Oxford battery aging data set, resulting in large fluctuations in the data and being very unstable. The robust local weighted regression algorithm can be used to remove outliers in the data, thereby further improving the final estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention is further described in detail below with reference to the accompanying drawings.

[0034] Figure 1 It is a flow chart of the present invention.

[0035] Figure 2 This is the second-order RC equivalent circuit model diagram. DETAILED DESCRIPTION

[0036] like Figure 1 As shown, the lithium battery health status estimation method based on the equivalent circuit model includes the following steps:

[0037] Establish as Figure 2 The second-order RC equivalent circuit model shown in the figure shows that the second-order RC equivalent circuit model equation is: ,in, U 1. U2 is the initial polarization voltage of the lithium battery, C 1. C 2 is the polarization capacitance of the lithium battery, R 1. R 2 is the polarization internal resistance of the lithium battery, U Indicates the terminal voltage of the lithium battery, U OC Indicates the open circuit voltage of the lithium battery. R 0 means ohmic internal resistance, I Indicates current;

[0038] Compared with the electrochemical model, the second-order RC equivalent circuit model is easier to achieve the accuracy of model parameter identification, and the meanings of various physical quantity parameters are clear, which is easy to implement and verify the model;

[0039] Compared with the Thevenin and PNGV models, the second-order RC equivalent circuit model uses a resistor and two RC loops in series, which can more accurately simulate the dynamic and static characteristics of the battery and take into account both the accuracy of the model and the complexity of the subsequent algorithm. Therefore, the second-order RC equivalent circuit model is more applicable in practical engineering.

[0040] The second-order RC equivalent circuit model equation is transformed into a mathematical form that can be recognized by the computer. The specific transformation process is:

[0041] According to the above second-order RC equivalent circuit model equation, the following relationship can be obtained:

[0042] ,

[0043] make: ,

[0044] The above formula can be rearranged as:

[0045] ,

[0046] make: , T is the sampling period;

[0047] The above formula can be organized into the following mathematical form that can be recognized by computers:

[0048] ,in, k 1. k 2. k 3. k 4. k 5 is the parameter to be sought:

[0049] , , , , ;

[0050] This mathematical form involves the terminal voltage of the lithium battery U , open circuit voltage U OC , Current I and the parameters to be requested k 1. k 2. k 3. k 4. k 5;

[0051] Terminal voltage U Data and current I The data directly adopts the terminal voltage and current data in the Oxford battery aging data set. For the open circuit voltage data, the formula Get, where V 0 is the initial charging value selected based on the terminal voltage in the Oxford battery aging data set. N Increasing number of samples for constant current charging of the Oxford battery aging dataset, R is any number in the voltage interval selected according to the charging voltage change;

[0052] The Oxford battery aging data set was obtained by aging tests on eight lithium batteries with a nominal capacity of 740 mAh. The aging test process was: a 1C charge and discharge cycle was performed using a constant current-constant voltage method, and the MPG-205 battery tester was used to measure and record the lithium battery current, voltage, and surface temperature every 100 aging cycles;

[0053] Select the above formula to obtain the open circuit voltage U OC The reasons are as follows: it takes a lot of time to obtain pulse experiments in the laboratory, and the relevant conditions are usually not available in engineering; the open circuit voltage data obtained in the laboratory is different from the actual working conditions; the electromotive force increases slowly during the constant current charging process until a sudden change occurs at the end of charging, which can generally be considered to be equal to the open circuit voltage of the battery;

[0054] The mathematical form described in step A is converted into a least squares format, so as to calculate the parameters to be determined in step A, and then the identification of the model parameters of the second-order RC equivalent circuit model is completed; specifically:

[0055] B1. Convert the mathematical form described in step A into the least squares format: ,in, , then k The estimated output of subsampling is ; The mean is 0 and the variance is A white noise sequence that obeys Gaussian distribution;

[0056] B2. Define the second residual , and set the second objective function : ,in, is the weighting coefficient, ;

[0057] B3, use the following formula to obtain Smallest : , and the obtained Feedback to the least squares format to calculate the parameters to be determined in step A. After the parameters to be determined are obtained, two more back-calculations can be performed to obtain These five model parameters; among them, the obtained Feedback to the basic principle of the least squares method to calculate the specific process of the parameters to be determined in step A is an existing technology; the use of the least squares method can enhance the robustness of model parameter identification;

[0058] The surface temperature data in the Oxford battery aging dataset is preprocessed using a robust locally weighted regression algorithm to remove outliers in the data.

[0059] Due to the limitations of sensor accuracy or measurement conditions, there are many outliers in the Oxford battery aging data set. This phenomenon becomes increasingly obvious with the increase in the number of cycles, which will have a significant impact on the final regression results. Therefore, in order to effectively extract features, the surface temperature data is preprocessed. For voltage and current data, since their fluctuations are not as drastic as those of surface temperature data, simple filtering such as interval averaging can be used to preprocess them to meet the requirements.

[0060] The specific processing process is as follows: First, for the surface temperature data t The time is calculated according to the following formula t Temperature fitting value at time : ,in, T i is the sampling time of the surface temperature data t i The corresponding temperature value, N Indicates t is the width of the sampling interval selected as the center, N The specific value of can be tested and modified continuously according to the test results to finally obtain the best effect. N value, Kis a kernel function, which represents the relationship between position and weight. This kernel function is one of the existing kernel functions. After calculating the temperature fitting value, the first residual is calculated according to the following formula , and calculate the improved weights based on the first residual , in which S is the median value of the first residual absolute value sequence, , and finally the weights will be improved With kernel function K The weights in are multiplied as the final weights, and the new ;

[0061] The surface temperature data preprocessed in step C and the battery terminal voltage in the Oxford battery aging dataset V , feature extraction is performed through the following formula to obtain the characteristic curve DTV m : , and obtain the characteristic curve DTV m The valley value P 1. Peak P 2. And valley value P 1 and peak P The voltage difference between 2 ,in, L is the sampling interval;

[0062] The reason for using the above formula for feature extraction is that both temperature and voltage have been proven to reflect the battery aging state by experiments, and the number of battery features is controlled by division; due to the accuracy limitation of the sensor, directly using continuous measurement data will be affected by relatively large noise. L The finite difference approximation is used on the sampling intervals;

[0063] Because too large a sampling interval may cause the loss of significant features, and too small a sampling interval may cause the measurement noise to be inconspicuous, in this embodiment, the sampling interval L The value of is 20;

[0064] For the above valley P 1. Peak P 2. And voltage difference , as the number of lithium battery cycles increases, the valley value P 1 gradually increases and moves to the left, the peak P 2 gradually decreases and keeps moving to the left with a relatively small amplitude. The voltage difference It gradually increases;

[0065] The model parameters of the second-order RC equivalent circuit model identified in step B and the valley value obtained in step D are completed using a regression support vector machine.P 1. Peak P 2. Voltage difference Mapping to the battery health status and establishing a lithium battery health status model includes the following steps:

[0066] E1. Model parameters of the second-order RC equivalent circuit model and valley value P 1. Peak P 2. Voltage difference , create a training set: ,in, Indicates i The input feature vector of samples is , Indicates the health status value of the lithium battery corresponding to the second-order RC equivalent circuit model, n represents the number of samples in the training set, refer to dimensional real number space;

[0067] E2. Using radial basis kernel function , the radial basis function is an existing technology, which maps the training set to a high-dimensional space to obtain a sample set ,in, is a real number, Also for the prior art;

[0068] The following first objective function is used: ,in, For the i The predicted value of samples, is the corresponding true value, that is, the health status value of the lithium battery corresponding to the second-order RC equivalent circuit model. The regression function is obtained by minimizing the first objective function. The weight vector in and bias ;

[0069] set up is the error between the predicted value and the true value. If the error is less than That is, it is defined as 0, and the mathematical language description is:

[0070]

[0071] Introducing slack variables and To describe the i The degree to which a sample does not satisfy the constraints, the first objective function is transformed into:

[0072]

[0073]

[0074] At this time, the obtained These are the model parameters of the second-order RC equivalent circuit model and the valley value obtained in step D. P 1. Peak P 2. Voltage difference Mapping to battery health status;

[0075] Since the model hyperparameters determine its generalization performance, after preliminarily determining the value range of the hyperparameters, grid search is used to jointly optimize the hyperparameters of the support vector machine, that is, the network is divided according to a certain common ratio, and the parameter pairs are determined through grid search. This can enable the support vector machine to have higher accuracy and stronger robustness. The grid search process is a prior art.

[0076] The above description is only a preferred embodiment of the present invention, and therefore cannot be used to limit the scope of implementation of the present invention. That is, equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the specification should still fall within the scope covered by the patent of the present invention.

Claims

1. A method for estimating the health status of a lithium battery, characterized in that: The steps include: A. Establish a second-order RC equivalent circuit model and transform the second-order RC equivalent circuit model equation into a mathematical form that can be recognized by a computer. The mathematical form involves the terminal voltage, open circuit voltage, current and parameters to be determined of the battery. The terminal voltage, current and open circuit voltage data are obtained through the Oxford battery aging data set. The parameters to be determined involve the parameters of the second-order RC equivalent circuit model. B. Convert the mathematical form described in step A into a least squares format, thereby calculating the parameters to be determined in step A, and then completing the identification of the model parameters of the second-order RC equivalent circuit model; C. Use the robust local weighted regression algorithm to preprocess the surface temperature data in the Oxford battery aging dataset to remove outliers in the data; D. The surface temperature data pre-processed in step C and the battery terminal voltage in the Oxford battery aging dataset V , feature extraction is performed through the following formula to obtain the characteristic curve DTV m : , and obtain the characteristic curve DTV m The valley value P 1. Peak P 2. And valley value P 1 and peak P The voltage difference between 2 ,in, L is the sampling interval; E. Use regression support vector machine to complete the model parameters of the second-order RC equivalent circuit model identified in step B and the valley value obtained in step D P 1. Peak P 2. Voltage difference Mapping to battery health status and establishing a lithium battery health status model; In step A, the second-order RC equivalent circuit model equation is: , where U1 and U2 are the initial polarization voltages of the lithium battery, C1 and C2 are the polarization capacitances of the lithium battery, R1 and R2 are the polarization internal resistances of the lithium battery, U represents the terminal voltage of the lithium battery, UOC represents the open circuit voltage of the lithium battery, R0 represents the ohmic internal resistance, and I represents the current; The mathematical form that can be recognized by the computer is: , where k1, k2, k3, k4, and k5 are the parameters to be determined, and are respectively expressed as: , , , , , , , .

2. A lithium battery health status estimation method according to claim 1, characterized in that: In the mathematical form, the lithium battery terminal voltage U , Current I Directly use the terminal voltage and current data in the Oxford battery aging dataset and use the following formula Get open circuit voltage data, where V 0 is the initial charging value selected based on the terminal voltage in the Oxford battery aging data set. N Increasing number of samples for constant current charging of the Oxford battery aging dataset, R is any number in the voltage range selected according to the charging voltage change.

3. A lithium battery health status estimation method according to claim 1 or 2, characterized in that: The Oxford battery aging data set was obtained by performing aging tests on eight lithium batteries with a nominal capacity of 740 mAh. The aging test process was as follows: a 1C charge and discharge cycle was performed using a constant current-constant voltage method, and the MPG-205 battery tester was used to measure and record the lithium battery current, voltage and surface temperature every 100 aging cycles.

4. A lithium battery health status estimation method according to claim 1 or 2, characterized in that: In the robust local weighted regression algorithm of step C, for the surface temperature data t The time is calculated according to the following formula t Temperature fitting value at time : ,in, T i is the sampling time of the surface temperature data t i The corresponding temperature value, N Indicates t is the width of the sampling interval selected as the center, K Represents the kernel function, which represents the relationship between position and weight. After calculating the temperature fitting value, the first residual is calculated according to the following formula , and calculate the improved weights based on the first residual , in which, S is the median value of the first residual absolute value sequence, , and finally the weights will be improved With kernel function K The weights in are multiplied as the final weights, and the new .

5. A lithium battery health status estimation method according to claim 1 or 2, characterized in that: The sampling interval in step D L The value range is [15,25].

6. A lithium battery health status estimation method according to claim 1 or 2, characterized in that: The following steps are also included: F. Use grid search to jointly optimize the hyperparameters of the support vector machine.

Images