Battery equivalent circuit parameter identification method and storage medium

Abstract

The application provides a battery equivalent circuit parameter identification method and a storage medium. The method comprises the following steps: performing pulse discharge and standing test on the battery, recording corresponding data, obtaining key test parameters and initial voltage; setting a search range of resistance and capacitance parameters to be identified; generating a large number of sample points by using a system sampling method; performing mathematical transformation on the samples to optimize the calculation process; using multi-core parallel calculation to optimize all samples, and finding out a parameter combination that makes the predicted voltage closest to the measured value; converting the optimization result back to the original parameter form; and finally automatically or manually judging and screening the most reliable one from the multiple sets of optimization results as the identification result. Through wide-range system sampling, parameter space conversion, parallel efficient calculation and multi-solution screening mechanism, the global search ability, calculation speed and result reliability of the parameter identification are significantly improved, and the shortcomings of the traditional method, such as being easy to fall into local optimum, low efficiency and single result, are effectively overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of battery, in particular to a battery equivalent circuit parameter identification method and a storage medium. BACKGROUND

[0002] In the field of battery research and application represented by lithium-ion batteries and sodium-ion batteries, accurate identification of the equivalent circuit parameters is crucial for battery performance evaluation, state estimation and optimization of battery management system. Currently, the commonly used battery equivalent circuit parameter identification methods mainly include traditional optimization algorithms and intelligent optimization algorithms.

[0003] Traditional optimization algorithms, such as least squares method, often rely on initial parameter guess values and are sensitive to the selection of initial values. If the initial value deviates greatly from the true solution, it is easy to fall into a local optimal solution and it is difficult to accurately obtain the global optimal parameter combination. Although intelligent optimization algorithms, such as ant colony algorithm and genetic algorithm, improve the optimization ability to some extent, these algorithms usually use serial computing method and the calculation process depends on the previous steps, resulting in low calculation efficiency. Especially when dealing with complex models and a large number of parameters, it takes a long time and it is difficult to meet the demand for fast identification in practical application.

[0004] In addition, most of the existing identification methods can only provide a single set of parameter solutions. However, in actual battery systems, due to the complexity of battery characteristics and the influence of test conditions, a single solution may not accurately reflect the true state of the battery. For example, in the parameter identification process, a solution may have a high fitting degree and a low residual error, but due to the local optimal solution, the identification result is unreliable. When the automatically obtained parameter solution does not match the actual situation, there is a lack of effective adjustment mechanism and it is difficult to obtain more realistic and reliable results through manual intervention.

[0005] Therefore, there is an urgent need for a parameter identification method that has global search ability, efficient calculation and robust results. SUMMARY

[0006] To solve the above technical problems, the present application provides a battery equivalent circuit identification method which is scientific and reasonable, logical and clear, has strong global search ability, efficient calculation and high reliability, and provides a corresponding storage medium.

[0007] According to one aspect of the present application, a battery equivalent circuit parameter identification method is provided, the equivalent circuit includes a constant voltage source, a resistor R0, a first RC circuit and a second RC circuit connected in series, wherein the first RC circuit is composed of a resistor R1 and a capacitor C1 in parallel, and the second RC circuit is composed of a resistor R2 and a capacitor C2 in parallel, and the identification method comprises the following steps:

[0008] S1. Perform constant current pulse discharge and rest test on the battery after standing in sequence, record the time sequence, terminal voltage sequence and current sequence of the test process, obtain the discharge current I p and pulse length t p , take the standing end time as zero time, and take the terminal voltage at zero time as the constant voltage source voltage Voc;

[0009] S2. Determine the to-be-identified parameter vector θ = [θ1, θ2, θ3, θ4, θ5] = [R0, R1, C1, R2, C2] and its lower bound LB j = L x θ j (prior) , upper bound UB j = U x θ j (prior) , where L, U are preset coefficients and L < U, j is a serial number respectively taking values from 1 to 5, θ j (prior) is the original parameter estimate value;

[0010] S3. Generate an N x 5-dimensional uniformly distributed matrix X that satisfies the Latin hypercube design, where each column X{:,j} is uniformly distributed on the interval [0, 1] and the hierarchical interval is 1 / N;

[0011] Construct a parameter sample set in logarithmic space, each parameter sample θ j (i) satisfies the following formula:

[0012] log 10 (θ j (i) ) = log 10 (LB j ) + X i,j x [log 10 (UB j ) - log 10 (LB j )]

[0013] Above θ j (i) is the jth parameter of the ith sample, i takes values from 1 to N, and j takes values from 1 to 5;

[0014] S4. Transform each parameter sample θ j (i) space to φ j (i) : φ j (i) = ln(θ j (i) );

[0015] Construct the circuit model fb according to the following formula:

[0016] When 0 <t≤t p At that time, V p (t,θ j (i) )=Voc-I(t)θ1 (i) -I(t){θ2 (i) [1-exp(-t / (θ2 (i) θ3 (i) ))]+θ4 (i) [1-exp(-t / (θ4 (i) θ5 (i) ))]};

[0017] When t>t p At that time, V p (t,θ j (i) ) = Voc - V 1c exp[-(tt p ) / (θ2 (i) θ3 (i) )]-V 2c exp[-(tt p ) / (θ4 (i) θ5 (i) )],in:

[0018] V 1c =θ2 (i) I p [1-exp(-t p / (θ2 (i) θ3 (i) ))],V 2c =θ4 (i) I p [1-exp(-t p / (θ4 (i) θ5 (i) ))];

[0019] In the above, t represents time, I(t) is the measured current value corresponding to time t, and V p This is the predicted value of the terminal voltage;

[0020] In circuit model fb, when 0 <t≤t p When t > t, it corresponds to the pulse discharge stage; when t > t p At that time, corresponding to the shelving stage, where V 1c and V 2c These represent the voltage values ​​of the two RC circuits at the initial moment of the suspension.

[0021] Construct a circuit transformation model based on the fb model using the following formula:

[0022] V p(t,φ j (i) )=fb(t,exp(φ j (i) ),I(t),Voc,I p ,t p )

[0023] S5. Parallel computing is performed by calling multiple cores, each core being aimed at different sample φ j (i) Optimization is performed:

[0024]

[0025] Optimal solution set is generated

[0026] wherein the time sequence of the test process is [t1, t2, …, t k ,…,t K ], V m (t k ) is the terminal voltage test value corresponding to time t k , V m (t k ,φ j (i) ) is the terminal voltage prediction value obtained by the circuit transformation model;

[0027] S6. Optimal solution set is converted back to the physical space according to the following formula:

[0028] θ j (i) =exp(φ j (i) );

[0029] S7. The most suitable solution is found from each solution θ j (i) obtained in step S6 as the final identification result.

[0030] Optionally, the battery is a lithium ion battery or a sodium ion battery.

[0031] Optionally, the test process synchronously samples time, terminal voltage and current and the sampling time interval remains unchanged.

[0032] Optionally, the sampling time interval of the test process is between 0.01s and 1s.

[0033] Optionally, the preset coefficient L in step S2 is between 0.0001 and 0.1, and the preset coefficient U is between 10 and 10000.

[0034] Optionally, the number of cores used for parallel computation in the step S5 is greater than 4.

[0035] Optionally, the parallel computation process in the step S5 divides the sample set into a plurality of subsets, and each subset is allocated to a computing core. j (i) The computing core performs independent optimization computation according to the sample allocated to it, and the results are summarized into an optimal solution set.

[0036] Optionally, in the step S7, each solution θ j (i) is substituted into the circuit model fb to obtain a corresponding terminal voltage prediction value sequence, and compared with the terminal voltage measured value sequence to evaluate the pros and cons of each solution, so as to find the most suitable solution as the final identification result.

[0037] Optionally, the way to evaluate the pros and cons of each solution is to calculate the residual mean square root value RMSE or the goodness of fit R 2 of the terminal voltage prediction value sequence and the terminal voltage measured value sequence, and the solution corresponding to the minimum RMSE or the solution corresponding to the maximum R 2 is taken as the most suitable solution.

[0038] According to another aspect of the present application, a computer readable storage medium is provided, which stores a plurality of computer instructions adapted to be loaded by a processor to execute the battery equivalent circuit parameter identification method of any one of the above.

[0039] The specific principles and beneficial technical effects of the technical solutions of the present application are described below. In the above technical solutions, the detailed information about the equivalent circuit model structure, parameter identification, and calculation of the terminal voltage prediction value sequence from the current measured value sequence based on the equivalent circuit model can be referred to Ahmad Rahmoun's article "Modelling of Li-ion batteries using equivalent circuit diagrams".

[0040] The specific implementation steps and principles of the battery equivalent circuit parameter identification are as follows:

[0041] Step S1. Data acquisition and basic processing

[0042] Content: sequentially perform constant current pulse discharge and resting test on the battery after standing, synchronously acquire the time sequence terminal voltage sequence and current sequence in the test process, and take the voltage value at the end of standing as the constant voltage source voltage Voc.

[0043] Principle: Rest the battery before testing, which helps to stabilize the battery state, and then the voltage value of the battery at the end of the rest time, which is the time before pulse discharge, can be regarded as the voltage of the constant voltage source in the equivalent circuit. Constant current pulse discharge can produce a significant voltage response, and the subsequent rest can obtain the rebound curve of the terminal voltage, and the rest phase R0 does not work, which helps to provide reliable and easy-to-analyze raw data for subsequent parameter identification.

[0044] Step S2. Parameter range setting

[0045] Content: Determine the to-be-identified parameter vector θ = [θ1, θ2, θ3, θ4, θ5] = [R0, R1, C1, R2, C2] and its lower bound LB j = L x θ j (prior) , upper bound UB j = U x θ j (prior) , where L, U are preset coefficients and L < U, j is a serial number that can take values 1 to 5 respectively, θ j (prior) is the original parameter estimate value. For example, let j = 1, then θ1 (prior) is the original parameter estimate value of R0, the lower bound LB1 of R0 = Lθ1 (prior) , the upper bound UB1 of R0 = Uθ1 (prior) .

[0046] Principle: By setting a reasonable parameter range, a wide search space is provided for subsequent sampling on the basis of the original guess value, i.e. the original parameter estimate value, increasing the possibility of covering the true solution and reducing the dependence on user experience.

[0047] Step S3. Latin hypercube design sampling

[0048] Content: Generate an N-row 5-column uniformly distributed matrix X that meets the Latin hypercube design, each column is uniformly distributed in the [0, 1] interval and the hierarchical interval is 1 / N; construct a parameter sample set in the logarithmic space, each parameter sample θ j (i) satisfies:

[0049] log 10 (θ j (i) ) = log 10 (LB j ) + X i,j x [log 10 (UB j ) - log 10 (LB j )].

[0050] Principle: Latin hypercube design is a stratified sampling method, which can cover the whole parameter space more evenly than random sampling, ensuring that each parameter interval has sample points, thus maximizing the coverage of the parameter space and improving the probability of finding the true solution under limited sample size.

[0051] Step S4. Parameter space transformation and model construction

[0052] Content: Transform each parameter sample θ j (i) to natural logarithm space φ j (i) = ln(θ j (i) ), and construct the circuit transformation model V p (t, φ j (i) ) = fb(t, exp(φ j (i) ), I(t), Voc, I p , t p ) based on the original circuit model fb, where θ1 (i) to θ5 (i) in the original circuit model fb are replaced by exp(φ1 (i) ) to exp(φ5 (i) ), respectively, but the parameter samples input to the model, i.e., the initial guess values of the parameters to be identified, are changed from θ1 (i) to θ5 (i) in the original circuit model fb to φ1 (i) to φ5 (i) in the natural logarithm space.

[0053] Principle: Through parameter space transformation, the nonlinear parameter identification problem is converted into a form more suitable for optimization algorithm processing. The logarithm space transformation method converts the original constrained optimization problem into an unconstrained optimization problem through nonlinear parameter remapping, and its mathematical essence is:

[0054] (1) Parameter space transformation

[0055] Define new variable φ j (i) = ln(θ j (i) ), where θ j (i) is the original parameter, and the subsequent optimization process of parameter identification is performed in φ j (i) space, and the final result is converted back to the original space.

[0056] (2) Conversion of constraint conditions

[0057] If no parameter space transformation is performed, the original parameter θ j (i) Optimization needs to explicitly set a lower bound θ j (i) > 0, which will have an iterative oscillation effect on the optimization process, especially near the value of 0; after the parameter space transformation, the original constraint θ j (i) > 0 is naturally satisfied in the definition of φ j (i) = ln(θ j (i) ), and there is no need to explicitly set a lower bound.

[0058] (3) Optimization target equivalence

[0059] The final optimization target after parameter space transformation is mathematically equivalent to the optimization target without transformation.

[0060] The main advantages of this step are:

[0061] (1) Strictly guarantee the positivity of the parameters. In terms of physical meaning, it strictly ensures that the to-be-identified parameters in the battery model are greater than 0, always conforming to the physical reality; at the same time, it can eliminate the boundary constraint that the original parameters are greater than 0, avoiding the repeated trial of the optimizer at the boundary and reducing the convergence oscillation.

[0062] (2) Parameter dimension equalization. In terms of order of magnitude compression, the above logarithmic transformation can map the original parameter range to a narrower interval, alleviating the parameter order imbalance problem in gradient descent. For example, if the original parameter range of parameter R0 is [0.001, 1000], i.e. θ1 (i) ∈ [0.001, 1000], spanning 7 orders of magnitude; after transformation, φ1 (i) ∈ [-6.9, 6.9], spanning a very small range.

[0063] At the same time, this helps to improve the condition number, significantly reducing the condition number of the Hessian matrix, and improving the numerical stability.

[0064] (3) Optimization efficiency improvement. For gradient direction optimization, in the φ j (i) space, the sensitivity of parameter changes to the target function is more balanced, and the gradient direction is closer to the true optimal path. In terms of step size adaptation, the optimizer can more freely and better adjust the step size, avoiding the contradiction of "large parameter step size being too small and small parameter step size being too large" in the original space.

[0065] (4) Initial value sensitivity reduction. After spatial transformation, on the one hand, the global search ability is enhanced, that is, even if the initial guess deviates from the true value by several orders of magnitude, the optimizer can still converge effectively; on the other hand, the multimodal problem is alleviated, which can reduce the risk of falling into a local optimum, especially in complex non-convex loss surface.

[0066] (5) Numerical stability enhancement. After spatial transformation, the exponential operation has better smoothness, which can avoid the numerical underflow problem caused by the parameter approaching 0 in the original space; at the same time, the sparsity of the Jacobian matrix is improved, the parameter correlation is reduced, and the matrix inversion is more stable.

[0067] Step S5. Parallel optimization calculation

[0068] Content: In the present application, multiple cores (generally more than 4) are called for parallel calculation, and the parameter sample set is evenly distributed to each core; each core performs optimization calculation on the distributed sample to minimize the error between the predicted and measured values of the terminal voltage, and generates an optimal solution set.

[0069] Principle: Using parallel computing technology, large-scale parameter optimization problems are decomposed into multiple sub-problems for simultaneous processing. Each core independently performs optimization calculation, fully utilizes the computing power of GPU and other parallel computing devices, greatly improves the calculation efficiency, and shortens the identification time. This parallel processing method is different from serial algorithms such as ant colony algorithm, which avoids waiting for the previous step, and realizes efficient use of computing resources.

[0070] Step S6. Convert the optimal solution set back to the physical space

[0071] Convert the optimal solution set from the natural logarithm space back to the physical space, so that the parameters have actual physical meaning and are convenient for engineering application.

[0072] Step S7. Find the best solution as the final identification result

[0073] Content: From the converted solution set back to the physical space, the best solution is selected. The root mean square error (RMSE) or goodness of fit (R 2 ) of the terminal voltage prediction value sequence and the measured value sequence corresponding to each solution can be calculated, and the solution with the smallest RMSE or the largest R 2 is selected as the best solution. Further, after selecting the best solution, manual judgment can be performed to determine whether to adopt the best solution according to the actual physical meaning and manual experience; if the manual judgment does not adopt the best solution, it is deleted from the solution set and the next best solution is found and judged until the manual judgment of the best solution meets the requirements.

[0074] Principle: The merits and demerits of each solution are objectively measured by the error evaluation index, and the existence of the solution set enables manual intervention to remove invalid solutions and select the next optimal solution when the automatic screening result is not satisfactory, thereby increasing the reliability and flexibility of the result.

[0075] In summary, the beneficial effects of the technical solution include:

[0076] 1. Wide and uniform sampling coverage of the true solution: Based on the original guess value, the method scales the parameter range by a preset coefficient L and U, and uses Latin hypercube design for sampling. This sampling method can uniformly distribute sample points in the parameter space, avoiding the sample aggregation or omission problem that may occur in random sampling, thereby covering the parameter range as widely as possible, including the true solution. Compared with the traditional method which relies on a single initial value or a small range search, the probability of finding the global optimal solution is significantly improved, especially in the case of inaccurate initial estimation of battery parameters.

[0077] 2. Parallel computing improves efficiency: A multi-core parallel computing strategy is adopted to evenly distribute the parameter sample set to each computing core for simultaneous optimization calculation. This approach fully adapts to the architectural features of current parallel computing hardware such as GPU, significantly reducing the calculation time. Unlike the serial computing process of intelligent optimization algorithms such as ant colony algorithm, there is no sequence restriction on the execution of the steps in this method, and multiple computing tasks can be pushed forward simultaneously, significantly improving the computing efficiency. For example, in the case of large-scale parameter identification, parallel computing can shorten the calculation time by several times or even dozens of times, meeting the demand for fast parameter identification in practical applications.

[0078] 3. Solution set output combined with manual screening: The identification result of this method is not a single parameter combination, but a solution set containing multiple optimal solutions. This feature provides more selection space for parameter optimization. When the optimal solution obtained automatically is judged to be invalid due to changes in battery characteristics, test errors, etc., the solution can be removed through manual intervention, and the next optimal solution can be selected from the solution set to continue the judgment until a satisfactory final solution is obtained. Compared with the limitations of single solution in existing methods, this method improves the reliability and adaptability of the parameter identification result through the combination of solution set and manual screening, especially in complex working conditions or when the battery characteristics drift. BRIEF DESCRIPTION OF DRAWINGS

[0079] Figure 1 The equivalent circuit model schematic diagram in the embodiment of the application, Figure 1 The two terminals at the far right are used to connect the load, and the voltage difference between them is the battery terminal voltage.

[0080] Figure 2 The flowchart of the battery equivalent circuit parameter identification method in the embodiment of the application.

[0081] Figure 3 This is a graph showing the variation of voltage and current with time during the test process in an embodiment of the present invention.

[0082] Figure 4 This is a graph comparing the predicted terminal voltage value sequence and the measured value sequence in an embodiment of the present invention.

[0083] Figure 5 This is the voltage residual graph in an embodiment of the present invention. Detailed implementation manners

[0084] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0085] A method for identifying battery equivalent circuit parameters, as Figure 1 shown, the equivalent circuit includes a constant voltage source, a resistor R0, a first RC circuit, and a second RC circuit connected in series in sequence, where the first RC circuit is composed of a resistor R1 and a capacitor C1 connected in parallel, and the second RC circuit is composed of a resistor R2 and a capacitor C2 connected in parallel. As Figure 2 shown, the method includes the following steps:

[0086] S1. Perform a constant current pulse discharge and a rest test on the battery after standing still in sequence, record the time series, terminal voltage series, and current series during the test process, and obtain the discharge current I p and the pulse duration t p , take the end moment of standing still as the zero moment, and take the terminal voltage at the zero moment as the constant voltage source voltage Voc, where the variation curves of the terminal voltage and current with time can be understood by referring to Figure 3 ;

[0087] S2. Determine the parameter vector to be identified θ = [θ1, θ2, θ3, θ4, θ5] = [R0, R1, C1, R2, C2] and its lower bound LB j = L×θ j (prior) , upper bound UB j = U×θ j (prior) , where L and U are preset coefficients and L < U, j is a serial number that can take values from 1 to 5 respectively, and θ j (prior) is the original parameter estimated value;

[0088] S3. Generate an N×5-dimensional uniformly distributed matrix X that satisfies the Latin hypercube design, where each column X{:,j} is uniformly distributed in the interval [0, 1] and the stratification interval is 1 / N;

[0089] Construct a parameter sample set in the logarithmic space, and each parameter sample θ j (i)Satisfy the following formula:

[0090] log 10 (θ j (i) ) = log 10 (LB j )+X i,j ×[log 10 (UB j )-log 10 (LB j )]

[0091] The above θ j (i) Let θ2 be the j-th parameter of the i-th sample, where i ranges from 1 to N and j ranges from 1 to 5. For example, θ2 (1) Represents the parameter sample set {θ j (i) The first parameter of the second parameter sample in the array is the constructed value of parameter R0, which appears in the second row and first column.

[0092] S4. Sample each parameter θ j (i) Space transformation to φ j (i) φ j (i) =ln(θ) j (i) );

[0093] Construct the circuit model fb using the following formula:

[0094] When 0 <t≤t p At that time, V p (t,θ j (i) )=Voc-I(t)θ1 (i) -I(t){θ2 (i) [1-exp(-t / (θ2 (i) θ3 (i) ))]+θ4 (i) [1-exp(-t / (θ4 (i) θ5 (i) ))]};

[0095] When t>t p At that time, V p (t,θ j (i) ) = Voc - V 1c exp[-(tt p ) / (θ2 (i) θ3 (i) )]-V 2c exp[-(ttp ) / (θ4 (i) θ5 (i) )], wherein:

[0096] V 1c = θ2 (i) I p [1-exp(-t p / (θ2 (i) θ3 (i) ))], V 2c = θ4 (i) I p [1-exp(-t p / (θ4 (i) θ5 (i) ))] ;

[0097] wherein t is time, I(t) is the current test value corresponding to time t, V p is the terminal voltage prediction value;

[0098] In the circuit model fb, when 0 < t ≤ t p , it corresponds to the pulse discharge stage; when t > t p , it corresponds to the standby stage, wherein V 1c and V 2c represent the voltage values of the two RC circuits at the beginning of the standby stage.

[0099] The circuit transformation model is constructed based on the fb model according to the following formula:

[0100] V p (t, φ j (i) ) = fb(t, exp(φ j (i) ), I(t), Voc, I p , t p )

[0101] S5. Parallel computing is performed by calling multiple cores, and each core is for different sample φ j (i) Optimization is performed:

[0102]

[0103] An optimal solution set is generated

[0104] wherein the test process time sequence is [t1, t2, …, t k , …, t K ], V m (t k ) is the terminal voltage test value corresponding to time t k , and Vm (t k ,φ j (i) ) is an end voltage prediction value obtained by a circuit transformation model;

[0105] S6. Transform the optimal solution set back to the physical space according to the following formula:

[0106] θ j (i) = exp(φ j (i) );

[0107] S7. Find the optimal solution from each solution θ j (i) obtained in step S6 as the final recognition result.

[0108] Specifically, the recognition process can be realized by programming with Matlab software.

[0109] Further, the battery is a lithium ion battery or a sodium ion battery.

[0110] Further, the test process synchronously samples time, end voltage and current, and the sampling time interval remains unchanged.

[0111] Further, the sampling time interval of the test process is between 0.01s and 1s.

[0112] Further, the preset coefficient L in step S2 is between 0.0001 and 0.1, and the preset coefficient U is between 10 and 10000.

[0113] Further, the number of cores for parallel computing in step S5 is greater than 4.

[0114] Preferably, the parallel computing process is called by a multi-core CPU or GPU.

[0115] Preferably, the parameter recognition in step S5 is performed by an optimization method, which can be realized by using the “lsqcurvefit” function of Matlab.

[0116] Further, when using the “lsqcurvefit” function of Matlab for parameter optimization recognition, the “levenberg-marquardt” or “trust-region-reflective” algorithm can be used.

[0117] In one embodiment, parallel computing is performed in a 32-core CPU; in another embodiment, parallel computing is performed in a 1024-core GPU.

[0118] Further, the parallel computing process in the step S5, each sample φ j (i) The sample set is evenly distributed to each computing core, each computing core performs independent optimization calculation according to the sample obtained by distribution, and the results are summarized into an optimal solution set.

[0119] Further, in the step S7, each solution θ j (i) The corresponding terminal voltage prediction value sequence is obtained by substituting each solution into the circuit model fb, and compared with the terminal voltage measured value sequence to evaluate the advantages and disadvantages of each solution, so as to find the optimal solution as the final identification result.

[0120] Further, the way to evaluate the advantages and disadvantages of each solution is to calculate the residual mean square root value RMSE or goodness of fit R 2 The solution corresponding to the minimum RMSE or the solution corresponding to the maximum R 2 is taken as the optimal solution.

[0121] Specifically, the curves of the terminal voltage prediction value sequence and the terminal voltage measured value sequence, and the residual curve can be understood with reference to Figure 4 and Figure 5 .

[0122] The application also provides a computer readable storage medium, which stores a plurality of computer instructions, the computer instructions are suitable for being loaded by a processor to execute the above-mentioned battery equivalent circuit parameter identification method.

[0123] It should be pointed out that in some cases, due to the influence of various factors such as the optimization process falling into a local optimal solution or measured data noise, the optimal solution may not conform to the actual situation, for example, the parameter identification result is obviously inconsistent with the common sense. Then, after the computer program automatically selects the optimal solution corresponding to the minimum RMSE or the solution corresponding to the maximum R 2 , manual inspection and judgment are performed; if the optimal solution found by manual judgment does not conform to the actual situation, the optimal solution should be deleted from the solution set, and then the next optimal solution is automatically found and manual inspection and judgment are performed again until the optimal solution approved by manual is obtained.

[0124] Embodiment

[0125] Please refer to the attached Figure 1 to Figure 5 to understand this embodiment. In this embodiment, if not specially stated, the units of current, voltage, resistance and capacitance are A, V, Ω and F respectively.

[0126] A certain type of lithium ion battery is LFP18650-1.8Ah, the positive electrode material is lithium iron phosphate, the negative electrode material is graphite, the shape is cylindrical battery, and the rated capacity is 1.8Ah. The equivalent circuit model of the battery is a second-order RC model, as shown in Figure 1 .

[0127] As Figure 2 shown, parameter identification is carried out on the battery.

[0128] S1. Adjust the battery to 80% SOC state at room temperature, stand for 1h, then carry out 1C rate, 10s constant current pulse discharge and 40s standby test in turn, record the time sequence, terminal voltage sequence and current sequence of the test process with a sampling time interval of 0.1s, obtain the discharge current I p =1.8A and pulse duration t p =10s, take the end time of standby as the time t=0 moment, and take the terminal voltage at zero moment as the constant voltage source voltage Voc=3.3271V, wherein the terminal voltage and current curves with time are shown in Figure 3 .

[0129] S2. Determine the parameter vector to be identified θ=[θ1,θ2,θ3,θ4,θ5]=[R0,R1,C1,R2,C2], preset the coefficient and L=0.01, U=100. The original parameter estimate θ j (prior) =[0.0584,0.00892,217,0.0591,603]. Take R0 as an example, the original parameter estimate θ1(prior)=0.0584Ω, the lower bound LB1=L×θ1(prior)=0.01*0.0584=0.000584Ω, and the upper bound UB1=U×θ1 (prior) =100**0.0584=5.84Ω. The serial number j=1 corresponds to R0, the serial number j=1 corresponds to R1, and so on.

[0130] S3. Let N=1000, generate a 1000*5-dimensional uniformly distributed matrix X that satisfies the Latin hypercube design, wherein each column X{:,j} is uniformly distributed in the interval [0,1] and the hierarchical interval is 1 / 1000.

[0131] Take the first five rows as an example, i.e. the values of X{1,:} to X{5,:} are shown in Table 1, and the data in the table is accurate to five decimal places.

[0132] Table 1 Values of the first five rows of matrix X

[0133] Column 1 Column 2 Column 3 Column 4 Column 5 Row 1 0.13871 0.44203 0.57896 0.23226 0.36940 Row 2 0.95607 0.01642 0.07917 0.97873 0.90379 Row 3 0.85720 0.77503 0.37497 0.94876 0.52033 Row 4 0.48267 0.06296 0.04163 0.98283 0.57168 Row 5 0.39026 0.46181 0.39383 0.07639 0.79637

[0134] In the logarithmic space, a parameter sample set is constructed, and each parameter sample θ j(i) satisfies the following formula:

[0135] log 10 (θ j (i) ) = log 10 (LB j ) + X i,j × [log 10 (UB j ) - log 10 (LB j )]

[0136] where i takes values from 1 to N and j takes values from 1 to 5.

[0137] Taking X{1,1} as an example, i.e. X 1,1 , the corresponding parameter sample θ1 (1) satisfies:

[0138] log 10 (θ1 (1) ) = log 10 (LB1) + X 1,1 × [log 10 (UB1) - log 10 (LB1)]

[0139] = log 10 (0.000584) + 0.13871 × [log 10 (5.84) - log 10 (0.000584)]

[0140] = -2.67875

[0141] Therefore, θ1 1) = 10^(-2.67875) = 0.002095

[0142] Taking the first five lines as examples, i.e. taking i to be 1 to 5 respectively, the values of θ j (i) are shown in Table 2.

[0143] Table 2 Values of θ j (i) when i takes values from 1 to 5

[0144] j＝1 j＝2 j＝3 j＝4 j＝5 i＝1 0.002095 0.00523 449.0642 0.005019 181.0962 i＝2 3.896618 0.000104 4.499348 4.858672 24857.6 i＝3 1.567521 0.112329 68.60357 3.686468 727.1551 i＝4 0.049784 0.000159 3.18413 5.045754 1166.868 i＝5 0.021255 0.006275 81.61322 0.001194 9242.314

[0145] S4. Transform each parameter sample θ j (i) into φ j (i) : φ j (i) = ln(θj (i) ),

[0146] Taking the first five rows as an example, i takes values ​​from 1 to 5 respectively, to obtain φ. j (i) The values ​​are shown in Table 3.

[0147] When φ is taken as 1 to 5 in Table 3i j (i) value

[0148] j＝1 j＝2 j＝3 j＝4 j＝5 i＝1 -6.168 -5.2534 6.107166 -5.29451 5.199028 i＝2 1.360109 -9.17341 1.503932 1.580765 10.12092 i＝3 0.449496 -2.18632 4.228345 1.304669 6.58914 i＝4 -3.00006 -8.74476 1.158179 1.618547 7.062079 i＝5 -3.85116 -5.07119 4.401991 -6.73009 9.131548

[0149] Construct the circuit model fb using the following formula:

[0150] When 0 <t≤t p At that time, V p (t,θ j (i) )=Voc-I(t)θ1 (i) -I(t){θ2 (i) [1-exp(-t / (θ2 (i) θ3 (i) ))]+θ4 (i) [1-exp(-t / (θ4 (i) θ5 (i) ))]};When t>t p At that time, V p (t,θ j (i) ) = Voc - V 1c exp[-(tt p ) / (θ2 (i) θ3 (i) )]-V 2c exp[-(tt p ) / (θ4 (i) θ5 (i) )],in:

[0151] V 1c =θ2 (i) I p [1-exp(-t p / (θ2 (i) θ3 (i) ))],V 2c =θ4 (i) I p [1-exp(-t p / (θ4 (i) θ5 (i) ))];

[0152] In the above, t represents time, I(t) is the measured current value corresponding to time t, and V pThe end voltage prediction value.

[0153] In the circuit model fb, when 0 < t ≤ t p , it corresponds to the pulse discharge stage; when t > t p , it corresponds to the rest stage, wherein V 1c and V 2c represent the voltage values of the two RC circuits at the beginning of the rest stage, respectively.

[0154] The circuit transformation model is constructed based on the fb model according to the following formula:

[0155] V p (t,φ j (i) )=fb(t,exp(φ j (i) ),I(t),Voc,I p ,t p )

[0156] Content: transform each parameter sample θ j (i) to the natural logarithm space φ j (i) =ln(θ j (i) ), and construct the circuit transformation model V p (t,φ j (i) )=fb(t,exp(φ j (i) ),I(t),Voc,I p ,t p ) based on the original circuit model fb, wherein θ1 (i) to θ5 (i) in the original circuit model fb are replaced by exp(φ1 (i) ) to exp(φ5 (i) ), respectively, but the parameter sample, i.e., the initial guess value of the parameter to be identified, input to the model is changed from θ1 (i) to θ5 (i) in the original circuit model fb to φ1 (i) to φ5 (i) in the natural logarithm space.

[0157] Taking the first row of the matrix X as an example, X{1,:}=[0.13871,0.44203,0.57896,0.23226,0.36940], the corresponding parameter sample set θ j (i) The first row θ j (1)= [0.002095, 0.00523, 449.0642, 0.005019, 181.0962], i.e. the initial guess values of the parameter combination R0 to C2 to be optimized in the circuit model fb are 0.002095 Ω, 0.00523 Ω, 449.0642 F, 0.005019 Ω, 181.0962 F respectively; but the initial guess values sent into the circuit transformation model are φ j (1) = [-6.168, -5.2534, 6.107166, -5.29451, 5.199028], the initial guess values of the parameters to be identified in the transformation model are subjected to exp(φ j (i) ) operation, so the circuit model fb and the circuit transformation model are actually equivalent, only the parameters to be identified in the circuit transformation model are φ j (i) rather than the original θ j (i) .

[0158] S5. Parallel computing is performed by calling 10 cores, each core processing 100 groups of samples and φ j (i) Optimization is performed:

[0159]

[0160] The optimal solution set is generated

[0161] where the time sequence of the test process is [t1, t2, …, t k , …, t K ], V m (t k ) is the terminal voltage test value corresponding to time t k , V m (t k , φ j (i) is the terminal voltage prediction value obtained by the circuit transformation model.

[0162] Specifically, in this embodiment, the “lsqcurvefit” function provided by Matlab is applied to perform parameter optimization identification, and the “levenberg-marquardt” optimization algorithm is used.

[0163] S6. The optimal solution set is converted back to the physical space according to the following formula:

[0164] θ j (i) = exp(φ j(i) );

[0165] S7. Find the best solution from each solution θ j (i) in step S6 as the final recognition result.

[0166] The optimal solution set in the embodiment There are 1000 groups of solutions, and the largest fitting degree R 2 = 0.9997 in each group of solutions, and the corresponding solution is selected as the best solution and verified by artificial verification: R0=0.055995Ω, R1=0.008479Ω, C1=222.2023F, R2=0.062556Ω, C2=568.1999F. The measured value and the predicted value curve obtained according to the best solution of the terminal voltage are shown in Figure 4 ; the difference between the predicted value and the measured value of the terminal voltage, i.e. the residual curve, is shown in Figure 5 , Figure 5 In order to observe conveniently, the voltage residual uses the dimension of mV.

[0167] In the embodiment of the application, on the basis of the original guess value, the parameter range is scaled in a large range by the preset coefficients L and U, and Latin hypercube design is used for sampling. This sampling method can uniformly distribute sample points in the parameter space, avoiding the problems of sample aggregation or omission that may occur in random sampling, so as to cover the parameter range including the true solution as widely as possible. In the embodiment, a multi-core parallel computing strategy is adopted to evenly distribute the parameter sample set to each computing core for optimization calculation, so as to improve the efficiency. The recognition result in the embodiment is not a single parameter combination, but a solution set containing 1000 optimal solutions. This feature provides more selection space for parameter optimization. By combining the solution set with artificial screening, the reliability and adaptability of the parameter recognition result are improved.

Claims

1. A method for identifying the parameters of a battery equivalent circuit, wherein the equivalent circuit comprises a constant voltage source, a resistor R0, a first RC circuit, and a second RC circuit connected in series, wherein the first RC circuit is composed of a resistor R1 and a capacitor C1 connected in parallel, and the second RC circuit is composed of a resistor R2 and a capacitor C2 connected in parallel, characterized in that, Includes the following steps: S1. After the battery has been left to stand, perform constant current pulse discharge and rest tests sequentially, and record the time sequence, terminal voltage sequence, and current sequence of the test process to obtain the discharge current I. p and pulse duration t p The zero point is defined as the end of the static period, and the terminal voltage at the zero point is used as the constant voltage source voltage Voc. S2. Determine the parameter vector θ to be identified θ = [θ1, θ2, θ3, θ4, θ5] = [R0, R1, C1, R2, C2] and its lower bound LB j = L × θ j (prior) and its upper bound UB j = U × θ j (prior) , where L and U are preset coefficients and L < U, j is the serial number that can take values from 1 to 5 respectively, and θ j (prior) is the original parameter estimated value; S3. Generate an N×5 dimensional uniform distribution matrix X that satisfies the Latin hypercube design, where each column X{:,j} is uniformly distributed in the interval [0,1] and the layer interval is 1 / N; Construct a parameter sample set in logarithmic space, where each parameter sample θ j (i) Satisfy the following formula: log 10 (θ j (i) )=log 10 (LB j )+X i,j ×[log 10 (UB j )-log 10 (LB j )] The values ​​of i range from 1 to N, and the values ​​of j range from 1 to 5. S4. Sample each parameter θ j (i) Space transformation to φ j (i) φ j (i) =ln(θ) j (i) ); Construct the circuit model fb using the following formula: When 0 < t ≤ t p , V p (t, θ j (i) ) = Voc - I(t)θ1 (i) - I(t){θ2 (i) [1 - exp(-t / (θ2 (i) θ3 (i) ))] + θ4 (i) [1 - exp(-t / (θ4 (i) θ5 (i) ))]}; When t>t p At that time, V p (t,θ j (i) ) = Voc - V 1c exp[-(tt p ) / (θ2 (i) θ3 (i) )]-V 2c exp[-(tt p ) / (θ4 (i) θ5 (i) )],in: V 1c =θ2 (i) I p [1-exp(-t p / (θ2 (i) θ3 (i) ))],V 2c =θ4 (i) I p [1-exp(-t p / (θ4 (i) θ5 (i) ))]; In the above, t represents time, I(t) is the measured current value corresponding to time t, and V p This is the predicted value of the terminal voltage; The circuit transformation model is constructed based on the fb model using the following formula: V p (t,φ j (i) )=fb(t,exp(φ j (i) ),I(t),Voc,I p ,t p ) S5. Parallel computation is performed by calling multiple cores, with each core targeting different samples φ. j (i) Execution optimization: Generate the optimal solution set The time series of the test process is [t1, t2, ..., t]. k ,…,t K ], V m (t k (t) represents time t k The corresponding terminal voltage test value, V m (t k ,φ j (i) The terminal voltage prediction value is obtained from the circuit transformation model. S6. Obtain the optimal solution set Convert back to physical space using the following formula: i j (i) =exp(φ j (i) ); S7. Solutions θ obtained from step S6 j (i) The optimal solution is selected as the final identification result.

2. The battery equivalent circuit parameter identification method according to claim 1, characterized in that, The battery is a lithium-ion battery or a sodium-ion battery.

3. The battery equivalent circuit parameter identification method according to claim 1, characterized in that, The test process synchronously samples time, terminal voltage, and current, and the sampling time interval remains constant.

4. The battery equivalent circuit parameter identification method according to claim 1 or 3, characterized in that, The sampling time interval of the test process is between 0.01s and 1s.

5. The battery equivalent circuit parameter identification method according to claim 1, characterized in that, In step S2, the preset coefficient L is between 0.0001 and 0.1, and the preset coefficient U is between 10 and 10000.

6. The battery equivalent circuit parameter identification method according to claim 1, characterized in that, In step S5, the number of cores used for parallel computing is greater than 4.

7. The battery equivalent circuit parameter identification method according to claim 1, characterized in that, The parallel computing process in step S5 will process each sample φ j (i) The sample set is evenly distributed to each computing core. Each computing core performs independent optimization calculations based on the sample it is allocated to, and the results are aggregated into the optimal solution set.

8. The battery equivalent circuit parameter identification method according to claim 1, characterized in that, In step S7, each solution θ j (i) Substitute each value into the circuit model fb to obtain the corresponding predicted terminal voltage sequence, and compare it with the measured terminal voltage sequence to evaluate the quality of each solution, thereby finding the optimal solution as the final identification result.

9. The battery equivalent circuit parameter identification method according to claim 8, characterized in that, The evaluation of each solution involves calculating the root mean square residual (RMSE) or goodness-of-fit (R²) using both the predicted and measured terminal voltage sequences. 2 The optimal solution is the one with the smallest RMSE or the one with the largest R2.

10. A computer-readable storage medium storing a plurality of computer instructions, characterized in that, The computer instructions are adapted to be loaded by a processor to execute the battery equivalent circuit parameter identification method according to any one of claims 1-9.

Images