Enhanced battery parameter identification method based on sinusoidal pulse current

Abstract

The invention relates to the technical field of battery parameter identification, in particular to an enhanced battery parameter identification method based on sinusoidal pulse current, and the method comprises the steps: S101, building a test platform, and collecting data; s102, modeling a battery equivalent circuit; s103, performing population initialization; s104, generating a sinusoidal excitation current; s105, reasoning the response voltage; s106, constructing a response voltage envelope line; s107, optimizing a main target based on a quadratic sum error; s108, constructing an auxiliary optimization target based on sinusoidal features; s109, carrying out genetic evolution; and S110, judging a termination condition and outputting and extracting an optimal solution. According to the method, the dynamic process of different time constants in the battery is fully and differentially excited, on the premise of not increasing any measurement hardware or test cost, the information quality is improved from a data source, a brand new path is opened up for realizing parameter identification with higher precision and higher robustness, and the method has a wide application prospect. Convergence efficiency and identification precision of battery parameter identification can be significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery parameter identification, and more particularly, to an enhanced battery parameter identification method based on sinusoidal pulse current. BACKGROUND

[0002] As the core energy carrier of modern energy conversion, the safety, performance and life of the battery are the key to determine the success of the system. The battery management system as the brain to protect the reliable operation of the battery, all its advanced functions, such as the accurate estimation of the state of charge (SOC), the accurate assessment of the state of health, the dynamic prediction of the peak power and the thermal runaway warning, are highly dependent on a mathematical model that can accurately reproduce the internal and external characteristics of the battery. The fidelity of the model is directly determined by the accuracy of the internal electrical parameters. However, these key parameters are not static and unchangeable, but will change significantly and nonlinearly with temperature, aging degree and charge-discharge rate. Therefore, developing a technology that can identify these dynamic parameters online, quickly and accurately is the core technical prerequisite for improving the estimation and control accuracy of the battery management system, optimizing energy efficiency management, prolonging the service life and ultimately ensuring the safety throughout the life cycle, which has extremely important engineering significance and application value.

[0003] The current mainstream online battery parameter identification method is a time-domain method, which applies a preset dynamic pulse current excitation signal to the battery, collects the voltage response sequence of the battery, and then based on the selected equivalent circuit model, uses an optimization algorithm to continuously adjust the parameters in the model, so that the error between the voltage waveform of the model simulation output and the actual measured voltage waveform is minimized. When the error converges within the set threshold, the current model parameters are considered to be the true electrical parameters of the battery under this working condition. This method can accurately identify the battery parameters without using expensive measuring equipment (such as electrochemical testers, etc.), and is suitable for various complex equivalent circuit models.

[0004] However, the research focus in the current time-domain parameter identification field is highly concentrated on improving algorithms and models, and generally ignores the design and optimization of the input end, i.e. the excitation current signal. The excitation signal used in the application of online battery parameter identification is mostly periodic fixed duty cycle charge-discharge pulse current or charge-discharge pulse current without explicit rule characteristics. This single or even irregular excitation form can only excite limited battery dynamic response information, and cannot fully expose the characteristics of all electrical parameters at different time scales. When the responses of different parameters (such as fast and slow polarization processes) are mixed in the time domain, relying only on the decoding ability of the algorithm or model to separate these features seriously restricts the accuracy ceiling.

[0005] Therefore, how to get rid of the information limitation brought by the traditional fixed pulse excitation without increasing any additional measurement hardware cost, enhance the parameter distinguishability at the data source end by designing a composite current excitation which can fully stimulate the multi-time scale dynamic characteristics of the battery, is a key technical problem to be solved in the current battery management field, and has great academic research value and broad industrial application prospect. SUMMARY

[0006] The application aims to provide a sinusoidal pulse current-based enhanced battery parameter identification method to solve the above problems.

[0007] The purpose of the application is achieved by the following technical solutions:

[0008] A sinusoidal pulse current-based enhanced battery parameter identification method, comprising the following steps:

[0009] S101. Build a test platform and collect data; the test platform comprises a power electronic constant current source power supply, a battery to be tested and a measuring device for data acquisition; by applying a periodic and programmable current excitation signal to the battery, the system obtains and stores a set of time series data, i.e. port current i[t] and port voltage v[t], which will be the only data basis for all subsequent parameter identification and feature calculation of the method;

[0010] S102. Battery equivalent circuit modeling; a second-order RC equivalent circuit model is used to establish the discrete state space equation of the battery equivalent circuit model;

[0011] S103. Population initialization;

[0012] S104. Sinusoidal excitation current generation; a controlled current with a duty cycle modulated according to a sinusoidal law and regularly switched between charging and discharging two discrete states is generated by a digital signal processor controlling a power electronic constant current source power supply; a high-frequency current closed-loop control is combined with a low-frequency sinusoidal modulation strategy;

[0013] S105. Response voltage inference; based on the battery discrete state space equation in step S102, the corresponding voltage sequence is derived;

[0014] S106. Response voltage envelope construction; the fluctuating envelope line is extracted from the voltage response sequence;

[0015] S107. Optimal main target based on sum of square errors;

[0016] S108. Construction of auxiliary optimization target based on sinusoidal characteristics; Under sinusoidal excitation current, an auxiliary optimization target different from the conventional optimization target is constructed through the envelope of voltage response. The auxiliary optimization target includes envelope slope matching degree and envelope shape similarity.

[0017] S109. Genetic evolution;

[0018] S110. Termination condition judgment and optimal solution output extraction: Determine whether the fitness is converging. If it is converging, the algorithm terminates and the parameters are extracted. If it has not yet converged, return to step S104.

[0019] Furthermore, in step S102, the second-order RC equivalent circuit model consists of an ohmic resistance R that reflects the purely resistive portion inside the battery. o The equivalent capacitance C of a battery b It consists of two parallel RC branches connected in series, namely the R1C1 branch and the R2C2 branch; the key advantage of this model is that it can simultaneously characterize two core dynamic processes: the fast dynamic response related to electrochemical polarization and the slow dynamic response related to concentration polarization.

[0020] Based on fundamental circuit laws, the set of battery model parameters λ that needs to be identified includes {R} o , R1, C1, R2, C2,V1[0], V2[0], C b}; where V1(0) is the initial voltage value of the R1C1 branch and V2(0) is the initial voltage value of the R2C2 branch; considering the actual solution, assuming the sampling time interval of the high-precision measuring device is Δt, the discretized state-space equation of the battery equivalent circuit model at the t-th sampling time is established as follows:

[0021] (1)

[0022] (2)

[0023] (3)

[0024] (4)

[0025] Where: v1 is the terminal voltage of capacitor C1, v2 is the terminal voltage of capacitor C2, v is the battery terminal voltage, i is the battery terminal current, SOC is the remaining charge percentage, and Focv-soc is the battery open-circuit voltage characteristic curve, which can be obtained offline in advance through testing.

[0026] Furthermore, in step S103, population initialization aims to construct an initial candidate solution set for subsequent evolutionary algorithms. Each individual P in the initial candidate solution set... i Includes a complete set of model parameters {R o , R1, C1, R2, C2, V1(0),V2(0), C b For each parameter to be identified, uniform random sampling is performed within the logarithmic interval of its optimization interval. The initial population generated in this way can achieve uniform coverage across all orders of magnitude.

[0027] Furthermore, in step S104, the method for generating the sinusoidal excitation current is as follows:

[0028] Inside the digital signal processor, two key signals are generated digitally: a sinusoidal modulated wave v. m [t] and triangular carrier v c [t];

[0029] sinusoidal modulated wave v m [t] is a low-frequency, standardized digital sine wave sequence with frequency f. m This determines the fundamental frequency of the final generated sinusoidal square wave current waveform;

[0030] Triangular carrier v c [t] is a digital triangular wave sequence that is at a higher frequency than the modulated wave, generated by a counter in a digital signal processor;

[0031] During the operation of the digital signal processor, by comparing v m [t] and v c The value of [t] determines the reference value for the converter current; when v m [t] > v c At [t], the digital signal processor will execute the target instruction I of the current closed-loop controller. ref Set to a positive value to enable constant current charging of the battery; when v m [t] < v c At [t], the digital signal processor will execute the target instruction I of the current closed-loop controller. ref Set to a negative value to enable constant current discharge of the battery.

[0032] Furthermore, under the excitation of the sinusoidal pulse current, the battery's terminal voltage exhibits a periodic response with the same frequency as the fundamental frequency of the excitation current. In step S105, the specific method for inferring the response voltage is as follows:

[0033] Based on the discretized state-space equation of the battery obtained in step S102, for each individual P in the population iUsing a time-domain iterative approach, at each discrete time t, based on the known excitation current i[t] and other state variables (SOC[t], v1[t], v2[t]), the current battery port voltage v[t] and the state variables (SOC[t+1], v1[t+1], v2[t+1]) at the next time are derived. By performing the above iterative calculation on all sampling times, the voltage at time P is obtained. i Individual model parameters {R o , R1, C1, R2, C2, V1(0), V2(0), C b The complete response voltage sequence v r(i) .

[0034] Furthermore, in step S106, the specific method for constructing the response voltage envelope is as follows:

[0035] Define a set of transient events Q, which contains all discrete time points where the excitation current switches from a positive value to a negative value or from a negative value to a positive value; let q be the time points before and after the corresponding voltage transition for any jump event q in Q. pre and q aft Then, the center voltage value at each transient event is calculated:

[0036] (5)

[0037] By connecting all the calculated center voltage points, a continuous voltage centerline can be drawn. This centerline is the envelope of the voltage response, denoted as v. env For the measured battery port voltage v o The envelope of [t] is denoted as v. o_env [q], for using population individual P i The battery port voltage v is derived from the model parameters. r(i) The envelope of [t] is denoted as v. r_env(i) [q].

[0038] The purpose of this step is to extract the envelope of fluctuations from the voltage response sequence, which represents the slowly varying response caused by the fundamental component of the excitation, and to provide input for the subsequent optimization process of parameter identification.

[0039] Furthermore, in step S107, the specific method for optimizing the main objective based on the sum of squared errors is as follows:

[0040] When evaluating the fitting accuracy of time-domain signals, a commonly used metric is the sum of squares error between two waveforms. Therefore, the main optimization objective function in this step is defined as:

[0041] (6)

[0042] Where N is the total discrete time of data acquisition; E main (P i The magnitude of the value corresponds to the quality of the fit of the response voltage amplitude; the smaller the value, the better.

[0043] Furthermore, in step S108, the method for constructing the envelope slope matching degree is as follows:

[0044] The slope of the response voltage envelope curve can be used as an independent evaluation metric. Specifically, when constructing the optimization algorithm, in addition to considering the traditional amplitude error metric, an optimization objective based on slope matching degree is introduced, forming a multi-objective optimization framework. The slope parameter of the response voltage envelope curve is quantified by the following formula:

[0045] (7)

[0046] Where, Δt q It is the time difference between the q and q-1 jump events; substituting the measured voltage envelope v into formula (7) respectively o_env Voltage envelope v derived from individual populations r_env(i) The slope k of the measured voltage response envelope can be obtained. o [q] and individual P i The slope k of the voltage response envelope r(i) [q]; Similarly, using the sum of squares error method, the objective function reflecting the degree of matching of the envelope slope is constructed according to the following formula:

[0047] (8)

[0048] The lower the value of an individual for this objective, the closer the slope of the response voltage derived by that individual is to the actual situation.

[0049] Furthermore, shape similarity is an index used to evaluate the consistency of the changing trends of two time-domain signals, and its evaluation process does not rely on a measure of Euclidean distance. In step S108, the calculation of envelope shape similarity includes the following three steps:

[0050] Step 1: Calculate the response voltage waveform v according to formula (7). env The slope of the envelope, k[q];

[0051] Step 2: Determine the envelope of the response voltage signal v according to formula (9). env The shape pattern value S[q] is given by θ, where θ represents the decision threshold, function1(·) and function2(·) are calculated by formulas (10) and (11) respectively, and γ and δ are hyperparameters that determine the value of S[q].

[0052] (9)

[0053] (10)

[0054] (11)

[0055] S[q] can effectively characterize the changing features of a time signal: the sign of S[q] reflects the direction of change of the time series waveform at time point q, with positive values ​​indicating an increase, negative values ​​indicating a decrease, and zero values ​​indicating stability; its absolute value |S[q]| describes the dynamic characteristics of the rate of change of the time series waveform at time point q: when 1-γ < |S[q]| < 1, it indicates that the current trend is slowing down; when |S[q]| = 1, it indicates that the rate of change is stable; when 1 < |S[q]| < 1+γ, it indicates that the trend is accelerating; when 1+γ < |S[q]| < 1+1.5γ, it means that the trend has reversed; the greater the deviation of |S[q]| from 1, the greater the acceleration of change.

[0056] Step 3: Substitute the measured voltage envelope v into formula (9) respectively. o_env Voltage envelope v derived under population individual parameters r_env(i) The shape mode value S of the measured response voltage envelope can be obtained separately. o [q] and individual P i The characteristic mode value S of the response voltage envelope r(i) [q], and then, using the same sum of squares error method, construct the objective function reflecting the similarity of envelope shape as shown in formula (12):

[0057] (12)

[0058] E s (P i The smaller the value of P, the better the individual's performance. i The derived time series of response voltages is closer to the measured values.

[0059] Furthermore, in step S109, the method of genetic evolution is as follows:

[0060] Combining the primary objective and auxiliary optimization objective constructed in steps S107 and S108, an optimization objective is constructed as minimizing the three-dimensional objective [E]. main E k E s The problem involves a multi-objective optimization problem, where the parameters of the battery model to be identified are used as decision variables in the optimization problem.

[0061] Subsequently, a multi-objective evolutionary algorithm is invoked to solve the problem. This algorithm simulates the biological evolution process, maintaining a population of candidate solutions and iteratively updating the population through a series of evolutionary operators such as selection, crossover, and mutation, guiding it towards the Pareto optimal front. During the iteration process, the algorithm uses mechanisms such as non-dominated sorting and crowding calculation to retain individuals that perform well and are widely distributed across various objective dimensions, thus ensuring the diversity and convergence of the solution set. The optimization process is considered to have converged and terminated when a preset termination condition is met, i.e., the fitness difference of multiple consecutive iterations is less than a set threshold ε. The individual that has reached the optimality in the objective is then used as the final identified parameter.

[0062] In a specific, non-limiting embodiment, the industry-recognized Non-Dominated Sorting Genetic Algorithm II (NSGA-II) can be used as the algorithm for this step. Of course, any other algorithm capable of solving multi-objective optimization problems, such as SPEA2, MOEA / D, etc., is also applicable to this invention.

[0063] Furthermore, in step S110, the specific methods for determining the termination condition and extracting the optimal solution output are as follows:

[0064] In the main optimization objective E main If the fitness value of the best-performing individual stabilizes, the algorithm terminates immediately; otherwise, the program returns to the iterative loop to continue execution. Once the optimization algorithm reaches convergence, individuals are selected from the final population who meet the primary objective E. main The individual with the highest fitness is selected, and the parameters corresponding to this individual are used as the final estimated values ​​of the model parameters.

[0065] The specific criterion for determining that the fitness value tends to stabilize is: in consecutive R iterations, the fluctuation range of the fitness value is less than the preset minimum positive number ε.

[0066] Compared with the prior art, the beneficial effects of the present invention are:

[0067] Compared with existing technologies, this invention breaks through the fixed paradigm of focusing only on optimization algorithms and equivalent models in battery parameter identification research. It innovatively shifts the optimization perspective forward to the design of the excitation signal itself, starting from the source of information excitation. This solves the fundamental problem of aliasing of parameter response features and fuzzy identification results caused by the limited information content of traditional single-pulse excitation. This method fully and differentially excites the dynamic processes of different time constants within the battery. Without increasing any measurement hardware or testing costs, it improves information quality from the data source, opening up a new path for achieving higher accuracy and greater robustness in parameter identification. It can significantly improve the convergence efficiency and recognition accuracy of battery parameter identification. Attached Figure Description

[0068] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0069] Figure 1 This is a flowchart of the enhanced battery parameter identification method based on sinusoidal pulse current in this invention.

[0070] Figure 2 This is a schematic diagram of the equivalent model for battery measurement in this invention.

[0071] Figure 3 This is a schematic diagram of the voltage envelope in an embodiment of the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Example:

[0074] like Figure 1 As shown, the enhanced battery parameter identification method based on sinusoidal pulse current in this embodiment includes the following steps:

[0075] S101. Setting up the test platform and collecting data:

[0076] like Figure 2 As shown, the test platform includes a power electronic constant current source, a lithium battery under test, and a measurement device for data acquisition. By applying a periodic and programmable current excitation signal to the battery, the system acquires and stores a set of time-series datasets, namely the port current i[t] and the port voltage v[t]. This will serve as the sole data basis for all subsequent parameter identification and feature calculations in this method.

[0077] S102. Battery equivalent circuit modeling:

[0078] This embodiment uses a widely used second-order RC equivalent circuit model to describe the electrical characteristics of the battery, such as... Figure 2 As shown, it consists of the following parts connected in series: an ohmic resistor R of the purely resistive part inside the battery. o The equivalent capacitance C of a battery bIt also includes two parallel RC branches, R1C1 and R2C2. The key advantage of this model is that it can simultaneously characterize two core dynamic processes: the fast dynamic response related to electrochemical polarization and the slow dynamic response related to concentration polarization.

[0079] Based on fundamental circuit laws, the set of battery model parameters λ that needs to be identified includes {R} o , R1, C1, R2, C2,V1[0], V2[0], C b}. Where V1(0) is the initial voltage value of branch R1C1, and V2(0) is the initial voltage value of branch R2C2. Considering the actual solution, assuming the sampling time interval of the high-precision measuring device is Δt, the discretized state-space equation of the battery equivalent circuit model at the t-th sampling time is established as follows:

[0080] (1)

[0081] (2)

[0082] (3)

[0083] (4)

[0084] Where: v1 is the terminal voltage of capacitor C1, v2 is the terminal voltage of capacitor C2, v is the battery terminal voltage, i is the battery terminal current, SOC is the remaining charge percentage, and Focv-soc is the battery open-circuit voltage characteristic curve, which can be obtained offline in advance through testing.

[0085] S103. Population Initialization:

[0086] This step aims to construct an initial candidate solution set for subsequent evolutionary algorithms. Each individual P... i Includes a complete set of model parameters {R o , R1, C1, R2, C2, V1(0), V2(0), C b For each parameter to be identified, uniform random sampling is performed within the logarithmic interval of its optimization interval, thus generating an initial population that can achieve uniform coverage across all orders of magnitude.

[0087] S104. Generation of sinusoidal excitation current:

[0088] This step aims to control a power electronic constant current source using a digital signal processor (DSP) to generate a controlled current with a sinusoidally modulated duty cycle that switches regularly between two discrete states: charging and discharging. The core of this method lies in combining a high-frequency current closed-loop control with a low-frequency sinusoidal modulation strategy. The DSP is an example of such a device.

[0089] First, within the digital signal processor, two key signals are generated digitally: one is a sinusoidal modulated wave v. m [t] is a low-frequency, standardized digital sine wave sequence with frequency f. m The fundamental frequency of the final generated sinusoidal square wave current waveform is determined, and the sinusoidal modulation wave v m The frequency range of [t] can be selected based on the time constant of the battery's own RC model, and the period is usually on the order of seconds. In this embodiment, the period is 7 seconds; the second is a triangular carrier wave v. c [t] is a digital triangular wave sequence with a higher frequency than the modulated wave, generated by a counter in a digital signal processor. During the operation of the digital signal processor, it is compared with v... m [t] and v c The value of [t] determines the reference value for the converter current. When v m [t]>v c At [t], the digital signal processor will execute the target instruction I of the current closed-loop controller. ref Set to a positive value to enable constant current charging of the battery. When v m [t] <v c At [t], the digital signal processor will execute the target instruction I of the current closed-loop controller. ref Setting it to a negative value enables the battery to discharge at a constant current. The target command generation process is as follows: Figure 2 As shown.

[0090] S105. Response Voltage Inference:

[0091] like Figure 3 As shown, under the excitation of the sinusoidal pulsed current, the battery's terminal voltage exhibits a periodic response with the same frequency as the fundamental wave of the excitation current. This step is based on the discretized state-space equation of the battery obtained in S102. For each individual P in the population... i Using a time-domain iterative approach, at each discrete time t, based on the known excitation current i[t] and other states (SOC[t], v1[t], v2[t]), the battery port voltage v[t] at the current time and the state variables (SOC[t+1], v1[t+1], v2[t+1]) at the next time are derived. By performing the above iterative calculation on all sampling times, the voltage at P can be derived. i Individual model parameters {R o, R1, C1, R2, C2, V1(0), V2(0), C b The complete response voltage sequence v r(i) .

[0092] S106. Construction of the response voltage envelope:

[0093] The purpose of this step is to extract the envelope of fluctuations from the voltage response sequence, which represents the slowly varying response caused by the fundamental component of the excitation, and to provide input for the subsequent optimization process of parameter identification.

[0094] Define a set of transient events Q, which contains all discrete time points where the excitation current switches from a positive to a negative value or from a negative to a positive value. Let q be the time points before and after the corresponding voltage transition for any transient event q in Q. pre and q aft Then, the center voltage value at each transient event is calculated:

[0095] (5)

[0096] By connecting all the calculated center voltage points, a continuous voltage centerline can be drawn. This centerline is the envelope of the voltage response, denoted as v. env The result is as follows Figure 3 The curve is shown in the figure. For the measured battery port voltage v... o The envelope of [t] is denoted as v. o_env [q], for using population individual P i The battery port voltage v is derived from the model parameters. r(i) The envelope of [t] is denoted as v. r_env(i) [q].

[0097] S107. Optimization objective based on sum of squared errors:

[0098] When evaluating the fitting accuracy of time-domain signals, a commonly used metric is the sum of squares error between two waveforms. Therefore, the main optimization objective function in this study is defined as:

[0099] (6)

[0100] Where N is the total discrete time of data acquisition. E main (P i The magnitude of the value corresponds to the quality of the fit of the response voltage amplitude; the smaller the value, the better.

[0101] S108. Construction of auxiliary optimization target based on sinusoidal features:

[0102] Under sinusoidal excitation current, an auxiliary optimization objective different from the conventional optimization objective can be constructed through the envelope of the voltage response. This invention will construct two auxiliary optimization objectives: envelope slope matching degree and envelope shape similarity degree.

[0103] (1) Auxiliary optimization objective 1: envelope slope matching degree

[0104] The slope of the response voltage envelope curve can be used as an independent evaluation index. Specifically, when constructing the optimization algorithm, in addition to considering the traditional amplitude error index, an optimization objective based on slope matching degree is introduced to form a multi-objective optimization framework. The slope parameter of the envelope can be quantified by formula (7):

[0105] (7)

[0106] Where, Δt q It is the time difference between the q and q-1 transition events. Substitute the measured voltage envelope v into formula (7). o_env Voltage envelope v derived from individual populations r_env(i) The slope k of the measured voltage response envelope can be obtained. o [q] and individual P i The slope k of the voltage response envelope r(i) [q]. Similarly, using the sum of squares error method, an objective function reflecting the envelope slope matching degree is constructed according to formula (8). The lower the value of an individual on this objective, the closer the slope of the response voltage derived by that individual is to the actual situation.

[0107] (8)

[0108] (2) Envelope shape similarity: Shape similarity is an index used to evaluate the consistency of the changing trends of two time-domain signals. Its evaluation process does not depend on the measurement of Euclidean distance. The calculation of envelope shape similarity mainly includes the following three steps:

[0109] Step 1: Calculate the response voltage waveform v according to formula (7). env The slope of the envelope, k[q];

[0110] Step 2: Determine the envelope of the response voltage signal v according to formula (9). envThe shape pattern value S[q] is given by θ, where θ represents the decision threshold. function1(·) and function2(·) are calculated by formulas (10) and (11) respectively, and γ and δ are hyperparameters that determine the value of S[q]. The hyperparameters {θ, γ, δ} of shape similarity can be set freely. The optimal parameter selection can be determined based on the overall fitting accuracy of the algorithm. Usually, γ and δ are selected between 0 and 1.

[0111] (9)

[0112] (10)

[0113] (11)

[0114] S[q] effectively characterizes the changing features of a time signal: the sign of S[q] reflects the direction of change of the time series waveform at time point q, with positive values ​​indicating an increase, negative values ​​indicating a decrease, and zero representing stability. Its absolute value |S[q]| describes the dynamic characteristics of the rate of change of the time series waveform at time point q: when 1-γ<|S[q]|<1, it indicates that the current trend is slowing down; when |S[q]|=1, it represents a stable rate of change; when 1<|S[q]|<1+γ, it indicates that the trend is accelerating; when 1+γ<|S[q]|<1+1.5γ, it means that the trend has reversed (i.e., from increasing to decreasing or from decreasing to increasing). The greater the deviation of |S[q]| from 1, the greater the acceleration of change.

[0115] Step 3: Substitute the measured voltage envelope v into formula (9) respectively. o_env Voltage envelope v derived under population individual parameters r_env(i) The shape mode value S of the measured response voltage envelope can be obtained separately. o [q] and individual P i The characteristic mode value S of the response voltage envelope r(i) [q], and then, using the same sum of squares error method, construct the objective function reflecting the similarity of envelope shape as shown in formula (12). E s (P i The smaller the value of P, the better the individual's performance. i The derived time series of response voltages is closer to the measured values.

[0116] (12)

[0117] S109. Genetic Evolution:

[0118] Combining the main objective and auxiliary optimization objectives constructed in S107 and S108, an optimization objective is constructed as minimizing the three-dimensional objective [E]. main Ek E s This is a multi-objective optimization problem. The parameters of the battery model to be identified are used as decision variables in the optimization problem.

[0119] Subsequently, a multi-objective evolutionary algorithm is invoked to solve the problem. This algorithm simulates biological evolution, maintaining a population of candidate solutions and iteratively updating the population through a series of evolutionary operators such as selection, crossover, and mutation, guiding it towards the Pareto optimal front. During the iteration process, the algorithm uses mechanisms such as non-dominated sorting and crowding calculation to retain individuals that perform well and are widely distributed across various objective dimensions, thus ensuring the diversity and convergence of the solution set. The optimization process is considered to have converged and terminated when a preset termination condition is met—that is, the fitness difference across multiple consecutive iterations is less than a set threshold ε—and the individual that has reached the optimal state in the objective is taken as the final identified parameter.

[0120] In a specific, non-limiting embodiment, the industry-recognized Non-Dominated Sorting Genetic Algorithm II (NSGA-II) can be used as the algorithm for this step. Of course, any other algorithm capable of solving multi-objective optimization problems, such as SPEA2, MOEA / D, etc., is also applicable to this invention.

[0121] S110. Termination condition determination and optimal solution output extraction:

[0122] This step is responsible for monitoring the running status of the evolutionary algorithm and automatically stopping the calculation when preset conditions are met. Once the main optimization objective E is achieved... main If the fitness value of the best-performing individual stabilizes, the algorithm terminates immediately; otherwise, the program returns to the iterative loop to continue execution. Once the optimization algorithm reaches convergence, individuals are selected from the final population who meet the primary objective E. main The individual with the highest fitness is selected, and the parameters corresponding to this individual are used as the final estimated values ​​of the model parameters.

[0123] The specific criterion for determining whether the fitness value changes tend to be stable is that the fluctuation range of the fitness value is less than a preset minimum positive number ε throughout R consecutive iterations. In this embodiment, R=5, and the value of ε is 1e. -4。

[0124] Tables 1 and 2 are comparison tables of the results of the true values, the traditional algorithm, and the method used in this embodiment. According to the experimental results, the algorithm in this case improves the fitting error accuracy by 2 times and the stability by 4 times without adding any additional measurement hardware. Furthermore, the identification accuracy of each parameter in the equivalent model is also significantly improved compared with the traditional scheme.

[0125] Table 1

[0126]

[0127] Table 2

[0128]

[0129] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for enhanced battery parameter identification based on sinusoidalized pulse current, characterized by, Comprising the following steps: S101. Building a test platform and collecting data; the test platform includes a power electronic constant current source power supply, a battery to be tested, and a measuring device for data collection; by applying a periodic and programmable current excitation signal to the battery, the system obtains and stores a set of time series data, i.e. port current i[t] and port voltage v[t], which will be the only data basis for all subsequent parameter identification and feature calculation of this method; S102. Battery equivalent circuit modeling; a second-order RC equivalent circuit model is adopted to establish the discrete state space equation of the battery equivalent circuit model; S103. Population initialization; S104. Sine excitation current generation; a digital signal processor controls the power electronic constant current source power supply to generate a controlled current with a duty cycle modulated according to a sine law and regularly switched between two discrete states of charging and discharging; a high-frequency current closed-loop control is combined with a low-frequency sine modulation strategy; S105. Response voltage inference; based on the battery discrete state space equation in step S102, the corresponding voltage sequence is derived; S106. Response voltage envelope construction; the fluctuating envelope is extracted from the voltage response sequence; S107. Main optimization target based on sum of square errors; S108. Auxiliary optimization target construction based on sine characteristics; Under the sine excitation current, the envelope of the voltage response is constructed to form an auxiliary optimization target different from the conventional optimization target, which includes envelope slope matching degree and envelope shape similarity; S109. Genetic evolution; S110. Termination condition judgment and optimal solution output extraction; judge whether the fitness tends to converge, if it tends to converge, the algorithm is terminated, and the parameters are extracted; if it has not converged, return to step S104.

2. The enhanced battery parameter identification method based on sinusoidal pulse current according to claim 1, wherein, In step S102, the second-order RC equivalent circuit model consists of an ohmic resistance R that reflects the purely resistive portion inside the battery. o The equivalent capacitance C of a battery b It consists of two parallel RC branches connected in series, namely R1C1 and R2C2. According to the basic law of circuit, the set of battery model parameters λ to be identified includes {R o , R1, C1, R2, C2, V1[0], V2[0], C b}; Wherein, V1(0) is the initial voltage of R1C1 branch, V2(0) is the initial voltage of R2C2 branch; considering the actual solution, assuming that the sampling time interval of high-precision measuring device is Δt, the discrete state space equation of battery equivalent circuit model at the tth sampling time is as follows: （1） （2） （3） （4） Wherein: v1 is the terminal voltage of capacitor C1, v2 is the terminal voltage of capacitor C2, v is the battery terminal voltage, i is the battery terminal current, SOC is the percentage of remaining power, Focv-soc is the open circuit voltage characteristic curve of the battery.

3. The enhanced battery parameter identification method based on sinusoidal pulse current according to claim 2, wherein, In step S103, each individual P i contains a complete set of model parameters {R o , R1, C1, R2, C2, V1(0), V2(0), C b}; for each parameter to be identified, uniform random sampling is performed in the logarithmic interval within the optimization interval thereof.

4. The enhanced battery parameter identification method based on sinusoidal pulse current according to claim 3, wherein, In step S104, the method for generating sine excitation current is: Inside the digital signal processor, two key signals are generated digitally: a sinusoidal modulation wave v m [t] and a triangular carrier wave v c [t]; sinusoidal modulation wave v m [t], is a low frequency, normalized digital sinusoidal wave sequence with frequency f m determines the fundamental frequency of the resulting sinusoidal square wave current waveform; Triangular carrier v c [t], is a sequence of relatively high frequency digital triangular waves, generated by a counter in the digital signal processor; by comparing v m [t] and v c [t] to determine the reference value of the inverter current; when v m [t] > v c [t], the digital signal processor sets the target instruction I ref of the current closed-loop controller to a positive value to make the battery constant-current charge; when v m [t] < v c [t], the digital signal processor sets the target instruction I ref of the current closed-loop controller to a negative value to make the battery constant-current discharge.

5. The enhanced battery parameter identification method based on sinusoidal pulsed current according to claim 4, wherein, In step S105, the specific method for response voltage inference is: Based on the battery discretized state space equation obtained in step S102, for each individual P in the population i , in a time domain iteration manner, at each discrete time t, according to the known excitation current i[t] and other state quantities (SOC[t], v1[t], v2[t]), the battery port voltage v[t] at the current time and the state quantities (SOC[t+1], v1[t+1], v2[t+1]) at the next time are derived; through the above iteration calculation on all sampling times, the complete response voltage sequence v i is derived under the model parameters {R o , R1, C1, R2, C2, V1(0), V2(0), C b} of the P r(i) individuals.

6. The enhanced battery parameter identification method based on sinusoidal pulse current according to claim 5, wherein, In step S106, the specific method for response voltage envelope construction is: A set of transient events Q is defined, which contains all the discrete time points at which the excitation current occurs a switch from positive to negative or from negative to positive; let any one of the jump events q in Q be the time point before and after the corresponding voltage jump of q pre and q aft ; then, the central voltage value at each transient event is calculated: (5) By connecting all the calculated center voltage value points, a continuous voltage center line is drawn, which is the envelope of the voltage response, denoted as v env For the measured battery port voltage v o [t] of the envelope, denoted as v o_env [q] for the battery port voltage v i [t] derived using the model parameters of the population individuals P r(i) [t] of the envelope, denoted as v r_env(i) [q].

7. The enhanced battery parameter identification method based on sinusoidal pulse current according to claim 6, wherein, In step S107, the specific method for main optimization target based on sum of square errors is: The main optimization target function is defined as: (6) Wherein, N is the total amount of discrete time of data acquisition; E main (P i ) The size of the value corresponds to the fitting of the response voltage amplitude. The smaller the better.

8. The enhanced battery parameter identification method based on sinusoidal pulse current according to claim 7, wherein, In step S108, the construction method of envelope slope matching degree is: The curve slope parameter of response voltage envelope is quantified by the following formula: (7) Where, Δt q It is the time difference between the q and q-1 jump events; substituting the measured voltage envelope v into formula (7) respectively o_env Voltage envelope v derived from individual populations r_env(i) The slope k of the measured voltage response envelope is obtained. o [q] and individual P i The slope k of the voltage response envelope r(i) [q]; Similarly, using the sum of squares error method, the objective function reflecting the degree of matching of the envelope slope is constructed according to the following formula: (8) The lower the value of a certain individual on this target, the closer the response voltage slope derived by the individual to the true situation.

9. The enhanced battery parameter identification method based on sinusoidal pulsed current according to claim 8, wherein, In step S108, the calculation of envelope shape similarity includes the following three steps: Step one: Calculate the envelope slope k[q] of the response voltage waveform v env according to formula (7) Step two: determine the shape pattern value S[q] of the response voltage signal envelope v env according to formula (9), wherein θ represents the determination threshold, function1(·) and function2(·) are calculated by formula (10) and (11) respectively, and γ and δ are hyperparameters for determining the value of S[q]; (9) (10) (11) S[q] can effectively depict the variation characteristics of the time signal: the sign of S[q] reflects the direction of the time series waveform variation at the q time point, the positive value is rising, the negative value is falling, and the zero value represents keeping stable; its absolute value |S[q]| describes the dynamic characteristics of the time series waveform variation rate at the q time point: when 1-γ < |S[q]| < 1, it indicates that the current trend is slowing down; when |S[q]| = 1, it represents the stable change rate; when 1 < |S[q]| < 1+γ, it shows that the trend is accelerating, and when 1+γ < |S[q]| < 1+1.5γ, it means that the trend has reversed; the greater the deviation of |S[q]| from 1, the greater the change acceleration; Step three: the measured voltage envelope v o_env and the voltage envelope v r_env(i) deduced under the population individual parameters are respectively brought into formula (9) to obtain the shape mode value S o [q] of the measured response voltage envelope and the shape mode value S i [q] of the response voltage envelope of the individual P r(i) , respectively, and then the objective function reflecting the envelope shape similarity as shown in formula (12) is constructed by using the square error method again: (12) E s (P i The smaller the value of P, the better the individual's performance. i The derived time series of response voltages is closer to the measured values.

10. The enhanced battery parameter identification method based on sinusoidal pulsed current according to claim 9, wherein, In step S109, the genetic evolution method is: Combining the main target and the auxiliary optimization target constructed in steps S107 and S108, a multi-objective optimization problem is constructed, in which the optimization target is to minimize the three-dimensional target [E main , E k , E s ]. The battery model parameters to be identified are taken as the decision variables of the optimization problem; Then, a multi-objective evolutionary algorithm is called to solve the problem; The algorithm simulates the biological evolution process, maintains a population of candidate solutions, and iteratively updates the population through a series of evolutionary operators such as selection, crossover, and mutation, guiding it to approach the Pareto optimal frontier.

Images