A quick estimation method for residual capacity of ternary lithium battery based on dynamic virtual resistance and non-static calibration

Abstract

The application discloses a kind of based on dynamic virtual resistance and no static calibration ternary lithium battery remaining power quick estimation method, method includes: step A, based on fitting coefficient correlation OCV-SOC curve fast initialization is carried out;Step B, the battery model containing time-varying virtual resistance is established, adaptive forgetting factor recursive least square method is combined online identification parameter, and real-time estimation is carried out by fusing ampere-hour integral and extended Kalman filter;Step C, in dynamic load mode, trigger no static calibration mechanism.The application realizes high-precision, high robustness power estimation in dynamic operating mode without long-time static.

Description

Technical Field

[0001] This invention belongs to the field of energy storage technology, and in particular relates to a method for rapidly estimating the remaining capacity of a ternary lithium battery based on dynamic virtual resistance and without static calibration. Background Technology

[0002] Lithium-ion batteries, especially ternary lithium batteries, have become the core power source for electric vehicles and energy storage systems due to their high energy density and long cycle life. Accurate estimation of the battery's remaining charge by the battery management system is crucial for preventing overcharging / over-discharging, improving energy utilization efficiency, and ensuring system safety.

[0003] Currently, mainstream SOC estimation methods are mainly based on the open-circuit voltage method and the ampere-hour integration method. The open-circuit voltage method utilizes the relatively fixed mapping relationship between SOC (State of Charge) and OCV (Open Circuit Voltage), offering high accuracy, but requires the battery to be left undisturbed for a long time to reach an electrochemically stable state, making it unsuitable for online use. The ampere-hour integration method accumulates changes in charge by integrating current in real time, allowing for online operation, but its error accumulates over time and depends on a precise initial SOC value. To combine the advantages of both methods, the industry often adopts a strategy of "ampere-hour integration + periodic OCV calibration," but calibration still relies on unrealistic long-term undisturbed conditions.

[0004] Furthermore, battery models, such as equivalent circuit models, are widely used to describe battery dynamic characteristics and serve as the basis for advanced observers such as extended Kalman filters. However, traditional model parameters are usually considered fixed or varying within a limited range, making it difficult to accurately characterize the polarization and loss characteristics of batteries under complex dynamic operating conditions and different health states, leading to model mismatch and decreased observer accuracy.

[0005] The existing technology has the following three main limitations:

[0006] (1) Strong dependence on long-term static calibration: Traditional OCV-SOC relationship calibration requires the battery to complete a full charge-discharge cycle and be left to stand for several hours at each SOC point to achieve voltage stability, which is extremely time-consuming (several days to several weeks). Periodic calibration at the application end also requires the vehicle or equipment to be left to stand for a long time, which is not feasible in most practical application scenarios, making the calibration function of the OCV method "virtually useless" and the cumulative error of ampere-hour integration cannot be effectively eliminated.

[0007] (2) Insufficient model accuracy under dynamic operating conditions: The parameters of traditional equivalent circuit models (such as first-order or second-order RC models) are usually obtained offline based on specific operating conditions (such as standard pulse test). Under actual dynamic and variable operating conditions and high-rate charging and discharging, the internal polarization behavior of the battery is complex, and the offline fixed parameter model cannot accurately reflect the instantaneous state, resulting in an increase in the deviation between the predicted voltage and the actual voltage of the model-based observer (such as EKF), and the failure of SOC correction.

[0008] (3) Lack of effective online self-correction mechanism: Although existing fusion algorithms can run online, they are essentially "open-loop" or "weakly corrected". The accumulated error of the ampere-hour integral and the model error will continue to propagate. Although algorithms such as EKF can make a certain degree of optimal estimation, without the injection of external "real" reference values ​​(such as accurate OCV points), it is impossible to fundamentally "reset" the accumulated error, and there is still a risk of divergence in long-term operation.

[0009] Therefore, developing a fast SOC estimation method that can eliminate the dependence on static calibration and maintain high accuracy in dynamic operation under all working conditions has become a technical challenge that urgently needs to be solved in this field. Summary of the Invention

[0010] The purpose of this invention is to solve the problems mentioned in the background art and to propose a fast estimation method for the remaining capacity of ternary lithium batteries based on dynamic virtual resistance and no static calibration.

[0011] To achieve the objective of this invention, this invention provides a rapid estimation method for the remaining capacity of a ternary lithium battery based on dynamic virtual resistance and without static calibration. The method includes:

[0012] Step A, fast initialization of the OCV-SOC curve based on the correlation of the fitting coefficients; including:

[0013] Step A1: Establish a benchmark database corresponding to the correlation model of the fitting coefficients;

[0014] Step A2: Parameterization and Relation Learning;

[0015] Step A3, rapid calibration of the target battery;

[0016] Step A4: Generate the curve;

[0017] Step B: Establish a battery model with time-varying virtual resistance, identify parameters online using the adaptive forgetting factor recursive least squares method, and perform real-time estimation by integrating ampere-hour integral and extended Kalman filter.

[0018] Step C: In dynamic load mode, trigger the no-static calibration mechanism.

[0019] The significant advancement of this invention compared to existing technologies lies in:

[0020] 1. Completely eliminate reliance on long-term static storage and achieve rapid online application: By using the "fitting coefficient correlation" method, the battery calibration time is shortened from several days to several hours; by using the "dynamic load-triggered calibration" mechanism, accuracy correction can be completed while the battery is running, without the need for the battery to be static for a long time, which greatly improves the engineering practicality of the method and the user experience.

[0021] 2. Significantly improved dynamic estimation accuracy: The introduced "time-varying virtual resistance" is an adaptive variable that effectively captures the additional dynamic losses of the battery under complex conditions such as high rate, low temperature, and aging, fundamentally improving the accuracy of the battery model under all operating conditions and providing a reliable foundation for high-precision observers such as EKF.

[0022] 3. The system possesses closed-loop self-calibration capability and is extremely robust: This invention creatively transforms dynamic operating conditions into calibration opportunities, designing a complete "detection-estimation-injection" online calibration process. This mechanism can periodically and automatically correct the accumulated error of the ampere-hour integral and model bias, preventing the estimation results from diverging over time and ensuring the long-term stability and reliability of the system.

[0023] 4. The algorithm has a moderate computational load and is easy to implement in engineering: The algorithms used in the whole solution, such as RLS and EKF, are mature linear or quasi-linear algorithms. The computational complexity is within the processing capability range of mainstream battery management chips, which makes it easy to implement and deploy in vehicle BMS or embedded systems.

[0024] To more clearly illustrate the functional characteristics and structural parameters of the present invention, further explanation is provided below in conjunction with the accompanying drawings and specific embodiments. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method provided in the embodiments of this application. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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.

[0027] The purpose of this application is:

[0028] (1) The length of time required for OCV-SOC relationship calibration and subsequent application calibration is shortened or even avoided, so as to realize the rapid initialization and online calibration of the SOC estimation system.

[0029] (2) Construct a high-precision battery model that can adapt to dynamic working conditions, and use an innovative "virtual resistor" to characterize the dynamic loss not covered by the traditional model, thereby improving the fitting accuracy of the model under real complex conditions.

[0030] (3) Design an intelligent triggering mechanism that can automatically find the moment when “quasi-OCV” can be calculated during the dynamic operation of the battery, and use the information at that moment to strongly correct the core estimation algorithm, thereby constructing a closed-loop SOC estimation system with online self-correction capability.

[0031] Step A: Perform rapid initialization of the OCV-SOC curve based on the correlation of the fitting coefficients;

[0032] Theoretical basis: Ternary lithium batteries of the same model and batch exhibit high morphological similarity in their OCV-SOC curves. This can be analyzed using a piecewise exponential function. It can fit with high precision; for batteries of the same family, there is a stable linear mapping relationship between the fitting parameter vector P = [a, b, c, d, k];

[0033] Step A1, establish a benchmark database: select several sample batteries, conduct a complete low-rate charge and discharge test at standard temperature, and let them stand until the voltage stabilizes at each 5% SOC interval to obtain a high-precision benchmark OCV-SOC curve set;

[0034] Step A2, Parameterization and Relationship Learning: Perform piecewise exponential fitting on each baseline curve to obtain the corresponding parameter vector P. i ; Through machine learning methods such as multiple linear regression, learn the mapping matrix M from a small number of keypoint OCV values ​​to the complete parameter vector;

[0035] Step A3, Target Battery Rapid Calibration: For a new battery to be calibrated, it only needs to be left to stand for a short time (approximately 30 minutes at each of the set key SOC points, such as 20%, 50%, and 80%), and the measured voltage values ​​are used to construct a vector V. key =[V 20 V 50 V 80 ] T .

[0036] Step A4, Curve Generation: Calculate the fitting parameters of the target battery using the mapping matrix M: P target =M×V key ; P target Substituting the values ​​into the fitting function generates the complete initial OCV-SOC relationship curve for the battery. This method reduces the calibration time from several days to less than 2 hours.

[0037] Specific implementation plan:

[0038] Step A1: Establish a benchmark database corresponding to the correlation model of the fitting coefficients;

[0039] The goal of this step is to build a high-quality learning set for the "fit coefficient correlation" model.

[0040] Step 11, Sample cell selection:

[0041] N batteries were randomly selected from ternary lithium batteries of the same model and the same production process batch as a sample battery pack, where N ≥ 30 to ensure statistical significance.

[0042] All sample batteries underwent initial consistency treatment: one complete charge-discharge cycle was performed at a standard temperature (e.g., 25°C) at a rate of 0.5C; the batteries were discharged from 100% to 0% and then recharged to 100% to eliminate historical state effects.

[0043] Step 12, Standard Test Environment Setup:

[0044] The high-precision battery testing equipment was placed in a constant temperature environment, with the temperature controlled at 25±0.5℃.

[0045] Using a separate voltage acquisition module (such as a high-precision data acquisition card), the measurement accuracy is better than ±1mV;

[0046] All sample battery tests must be performed sequentially or in parallel under the same equipment and environment;

[0047] Step 13, execute the benchmark testing process:

[0048] For each sample battery, perform the following "resting point test method": a) Full charge resting: Charge the battery at a constant current of 0.3C to the cutoff voltage, then switch to constant voltage charging until the current drops to 0.05C, and then let it rest for 2 hours. Record the voltage at this time as the OCV corresponding to 100% SOC; b) Stepped discharge and resting: Discharge the battery at a small rate of 0.05C to the target SOC point (e.g., 95%). After the discharge is completed, immediately begin resting. The resting judgment criterion is: within 5 consecutive minutes, the rate of change of the terminal voltage is less than 0.1mV / min. Record the voltage after stabilization as the OCV of the corresponding SOC point; c) Repeat step b, with 5% intervals, and test the OCV value at SOCs of 95%, 90%, 85%, ..., 5%, 0%. It is recommended that the SOC interval be 5%, for a total of 21 points.

[0049] Data recording: For each battery i, a set of pairing data is finally obtained: {SOC j OCV ij}, j=1,2,...,21;

[0050] Step A2: Parameterization and Relation Learning

[0051] This is the core of the method, which aims to discover the mapping relationship between "a small number of key point OCVs" and "complete curve morphology parameters".

[0052] Step 21, Curve parameterization: Fit each baseline curve.

[0053] Choosing a fitting function: A five-parameter piecewise exponential function with clear physical meaning and high fit is adopted.

[0054] ;

[0055] a, b, c, d, and k are the parameters corresponding to the five-parameter piecewise exponential function. The fitting function is divided into two segments, including two exponential functions and one constant term. a and b are the constant and exponential terms of the first segment's fitting function, respectively; c and d are the constant and exponential terms of the second segment's fitting function; and k is the offset constant. Specifically, terms a and b describe the voltage variation characteristics in the high SOC region (>50%); terms c and d describe the voltage variation characteristics in the low SOC region (<50%); and k is the offset constant.

[0056] Perform nonlinear fitting: For the 21 sets of data of the i-th sample battery, perform fitting using the nonlinear least squares method (Levenberger-Marquardt algorithm) to obtain the optimal parameter vector:

[0057] ;

[0058] a, b, c, d, and k are five parameters of the fitting function describing the data of one battery sample. There are N battery samples in total. Let represent the fitting parameters of the i-th battery sample, where i takes values ​​from 1 to N;

[0059] Fit quality check: The coefficient of determination R² for all sample batteries must be > 0.999; any battery data with R² not meeting the standard should be removed to ensure database quality.

[0060] Step 22, establish the keypoint-parameter mapping relationship, that is, determine the learning mapping matrix M:

[0061] Key point OCV extraction: From the 21 OCV data points of each sample battery, the voltage values ​​corresponding to the 20%, 50%, and 80% SOC points are extracted to form the key point voltage vector.

[0062] ;

[0063] These three points were chosen because they are located in the low, medium, and high regions of the curve, where the linearity and rate of change differ significantly, thus best representing the characteristics of the curve.

[0064] Constructing the learning dataset: Pair the data from all N sample batteries to form the learning dataset.

[0065] ;

[0066] Learning the linear mapping matrix M: Assume there is a multivariate linear mapping relationship from the key point voltage to the complete parameters, i.e.:

[0067] ;

[0068] in, It is the augmented key point voltage vector, with a constant term of 1 added to absorb fitting bias;

[0069] M is a 5x4 matrix; stack N sets of data into a matrix form:

[0070] Parameter matrix: ;

[0071] Key point voltage matrix: ;

[0072] Calculate the optimal mapping matrix using the analytical solution of multiple linear regression:

[0073] ;

[0074] Cross-validation: "Leave one out" cross-validation is used; matrix M is trained with N-1 sets of data each time, and the remaining 1 set is used for validation; the final evaluation metric is the ratio of the trained M to the validated matrix. Predicted The average absolute voltage error (MAE) between the OCV curve generated after substituting into the fitting function and the actual 21-point OCV data of the verification battery is required to be < 5mV for all verification rounds. If the requirement is not met, the final mapping matrix M is recalculated using all N sets of data until the requirement is met.

[0075] Step A3, Target Battery Quick Calibration:

[0076] For any target battery, quickly obtain the OCV-SOC curve by following these steps;

[0077] Step 31, Condition Preparation: Ensure the battery is in a standard temperature environment (25±2°C). Connect the test equipment.

[0078] Step 32, obtain the OCV of three key points:

[0079] Charge to 80% SOC: If the initial state of the battery is unknown, first charge it to about 85% SOC using the normal method, and then discharge it to 80% SOC with a small current of 0.05C.

[0080] First short-term rest period: Disconnect the load and let the battery rest; the rest period is 30 minutes, or stop when the voltage change rate is below 0.2mV / min for 3 consecutive minutes (whichever comes first); record the stable voltage value V80;

[0081] Discharge to 50% and 20% SOC points: Discharge to 50% SOC and 20% SOC with a small current of 0.05C in sequence, and repeat the short resting process at each point, and record V50 and V20 respectively;

[0082] Step 33, Data Packaging: Key Point Voltage Vectors Constituting the Target Battery:

[0083] .

[0084] Step A4: Generate the curve;

[0085] Step 41, calculate the target battery curve parameters:

[0086] ;

[0087] This operation is a single matrix multiplication, which can be completed instantaneously on a microcontroller.

[0088] Step 42, Generate the complete OCV-SOC curve: The obtained curve... Substituting the five-parameter piecewise exponential function, we obtain a continuous and smooth OCV-SOC relationship curve from SOC 0% to 100%.

[0089] Step B: Establish a battery model with time-varying virtual resistance, identify parameters online using the adaptive forgetting factor recursive least squares method, and perform real-time estimation by integrating ampere-hour integral and extended Kalman filter.

[0090] Step B1, construct the battery model;

[0091] This invention adds a time-varying virtual resistor in parallel to the traditional second-order RC equivalent circuit model. The virtual resistor is not a real physical component, but rather a comprehensive parameter used to characterize dynamic losses not covered by the model (such as active lithium loss due to aging, additional polarization at high rates, etc.). The model is described as follows:

[0092] The terminal voltage equation is as follows:

[0093] ;

[0094] Polarization voltage differential equation: ;

[0095] ;

[0096] in, I is the terminal voltage, and I is the current (positive for discharge). Determined by the curve provided in step 1, For ohmic internal resistance, The parameters of the two RC elements in the traditional second-order RC equivalent circuit model of the battery are resistor 1, capacitor 1, resistor 2, and capacitor 2, respectively.

[0097] Step B2, online parameter identification (adaptive forgetting factor recursive least squares method); using an adaptive forgetting factor The RLS algorithm identifies model parameters online in real time: ; * , * ;

[0098] The forgetting factor update rule is as follows: ;

[0099] in, Let λ(k) be the change in current between time k and time k-1, and β be the adjustment coefficient. When the current changes drastically (dynamic operating condition), λ(k)→1, and the algorithm quickly forgets the old data and tracks the new dynamics. When the current is stable, λ(k) decreases to enhance the smoothness and stability of parameter estimation.

[0100] Step B3, Extended Kalman filter fusion estimation;

[0101] State equations: with SOC and two polarization voltages as state variables. ;

[0102] ;

[0103] ;

[0104] ;

[0105] in, For rated capacity, Coulomb efficiency;

[0106] Observation equation:

[0107] EKF Iteration Process: EKF uses the result of the ampere-hour integral as the predicted value, and the difference between the model-predicted voltage y(k) and the actual measured voltage z(k) as the observation information. It then uses the Kalman gain K to optimally correct the state vector (especially the SOC). The online identification module provides real-time... , , , These parameters ensure the accuracy of the observation equations.

[0108] Step C: In dynamic load mode, trigger the no-static calibration mechanism.

[0109] Step C1, Dynamic mode detection; Real-time monitoring of load current sequence Calibration is triggered when a dynamic pattern matching the "excitation-relaxation" characteristic is detected; a typical pattern is: a continuous... stable current (Such as constant current discharge) followed immediately; another continuous period tiny current The triggering condition is: and 0, and and Sufficiently long enough for the polarization voltages of both stages to reach a quasi-steady state;

[0110] Step C2, online OCV estimation and forced calibration;

[0111] Data is collected at the end of both phases of the detected dynamic pattern. and ; Estimate polarization voltage variation using currently identified model parameters ;

[0112] The core OCV estimation method is as follows:

[0113] ;

[0114] Using the estimated Query the initial OCV-SOC curve generated in step 1, and use reverse interpolation to obtain a high-confidence reference SOC value. ;

[0115] Will As absolute observations injected into the EKF system: at the corresponding time, the observation equations of the temporary EKF are changed to... The observed value is set as It assigns a very small observation noise covariance; EKF will strongly pull the SOC value in the state vector towards [the desired value] in this iteration. This allows for one-click clearing of the accumulated error of the ampere-hour integral; after calibration, the system immediately returns to normal voltage observation mode.

[0116] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for rapid estimation of remaining capacity of a ternary lithium battery based on dynamic virtual resistance and without static calibration, characterized in that, The method includes: Step A, fast initialization of the OCV-SOC curve based on the correlation of the fitting coefficients; including: Step A1: Establish a benchmark database corresponding to the correlation model of the fitting coefficients; Step A2: Parameterization and Relation Learning; Step A3, rapid calibration of the target battery; Step A4: Generate the curve; Step B: Establish a battery model with time-varying virtual resistance, identify parameters online using the adaptive forgetting factor recursive least squares method, and perform real-time estimation by integrating ampere-hour integral and extended Kalman filter. Step C: In dynamic load mode, trigger the no-static calibration mechanism.

2. The method according to claim 1, characterized in that, Step A1: Establish a benchmark database corresponding to the correlation model of the fitted coefficients; including: Step 11, Sample cell selection: N batteries were randomly selected from ternary lithium batteries of the same model and the same production process batch as a sample battery pack, where N ≥ 30; All sample batteries underwent initial consistency treatment: one complete charge-discharge cycle at 0.5C rate under standard temperature; discharged from 100% to 0%, and then recharged to 100% to eliminate historical state effects; Step 12, Standard Test Environment Setup: The high-precision battery testing equipment was placed in a constant temperature environment, with the temperature controlled at 25±0.5℃; Using an independent voltage acquisition module, the measurement accuracy is better than ±1mV; All sample battery tests must be performed sequentially or in parallel under the same equipment and environment; Step 13, execute the benchmark test process: For each sample battery, the following "resting point test method" was performed: a) Full charge resting: The battery was charged at a constant current of 0.3C to the cutoff voltage, then switched to constant voltage charging until the current dropped to 0.05C, and then rested for 2 hours. The voltage at this time was recorded as the OCV corresponding to 100% SOC; b) Stepped discharge and resting: The battery was discharged at a small rate of 0.05C to the target SOC point. After the discharge was completed, resting was started immediately. The resting judgment criterion was: the rate of change of the terminal voltage was less than 0.1mV / min within 5 consecutive minutes. The voltage after stabilization was recorded as the OCV of the corresponding SOC point; c) Repeat step b, with 5% as an interval, and test the OCV value at SOC of 95%, 90%, 85%, ..., 5%, 0% point by point. Data recording: For each battery i, a set of pairing data is finally obtained: {SOC j OCV ij }, j=1,2,...,21.

3. The method according to claim 2, characterized in that, Step A2: Parameterization and relation learning, including: Step 21, Curve parameterization: Fit each baseline curve. Choosing a fitting function: A five-parameter piecewise exponential function was adopted. ； a, b, c, d, and k are the parameters corresponding to the five-parameter piecewise exponential function. The fitting function is divided into two segments, including two exponential functions and one constant term. a and b are the constant and exponential terms of the first segment's fitting function, respectively; c and d are the constant and exponential terms of the second segment's fitting function; and k is the offset constant. Specifically, terms a and b describe the voltage change characteristics in the high SOC region (>50%); terms c and d describe the voltage change characteristics in the low SOC region (<50%); and k is the offset constant. Perform nonlinear fitting: For the 21 sets of data of the i-th sample battery, use nonlinear least squares method to fit the data and obtain the optimal parameter vector: ； a, b, c, d, and k are five parameters of the fitting function describing the data of one battery sample. There are N battery samples in total. Let represent the fitting parameters of the i-th battery sample, where i takes values ​​from 1 to N; Fit quality check: The coefficient of determination R² for all sample batteries must be > 0.999; any battery data with R² not meeting the standard should be removed to ensure database quality. Step 22, establish the key point-parameter mapping relationship, i.e., determine the learning mapping matrix M: Key point OCV extraction: From the 21 OCV data points of each sample battery, the voltage values ​​corresponding to the 20%, 50%, and 80% SOC points are extracted to form the key point voltage vector. ； Constructing the learning dataset: Pair the data from all N sample batteries to form the learning dataset. ； Learning the linear mapping matrix M: Assume there is a multivariate linear mapping relationship from the key point voltage to the complete parameters, i.e.: ； in, The key point voltage vector is augmented, and a constant term of 1 is added to absorb the fitting bias; M is a 5x4 matrix; stack N sets of data into a matrix form: Parameter matrix: ; Key point voltage matrix: ; Calculate the optimal mapping matrix using the analytical solution of multiple linear regression: ； Cross-validation: "Leave one out" cross-validation is used; matrix M is trained with N-1 sets of data each time, and the remaining 1 set is used for validation; the final evaluation metric is the ratio of the trained M to the validated matrix. Predicted The average absolute voltage error (MAE) between the OCV curve generated after substituting into the fitting function and the actual 21-point OCV data of the verification battery is required to be < 5mV for all verification rounds. If the requirement is not met, the final mapping matrix M is recalculated using all N sets of data until the requirement is met.

4. The method according to claim 3, characterized in that, Step A3, rapid calibration of the target battery, includes: For any target battery, quickly obtain the OCV-SOC curve by following these steps; Step 31, Condition Preparation: Ensure the battery is in a standard temperature environment (25±2°C). Connect the test equipment; Step 32, obtain the OCV of three key points: Charge to 80% SOC: If the initial state of the battery is unknown, first charge it to about 85% SOC using the normal method, and then discharge it to 80% SOC with a small current of 0.05C. First short-term rest period: Disconnect the load and let the battery rest; the rest period is 30 minutes, or stop when the voltage change rate is below 0.2mV / min for 3 consecutive minutes; record the stable voltage value V80; Discharge to 50% and 20% SOC points: Discharge to 50% SOC and 20% SOC with a small current of 0.05C in sequence, and repeat the short resting process at each point, and record V50 and V20 respectively; Step 33, Data Packaging: Key Point Voltage Vectors Constituting the Target Battery: 。 5. The method according to claim 4, characterized in that, Step A4: Generate the curve; include: Step 41, calculate the target battery curve parameters: ； Step 42, Generate the complete OCV-SOC curve: The obtained... Substituting the five-parameter piecewise exponential function, we obtain a continuous and smooth OCV-SOC relationship curve from SOC 0% to 100%.

6. The method according to claim 5, characterized in that, Step B: Establish a battery model with time-varying virtual resistance, identify parameters online using the adaptive forgetting factor recursive least squares method, and perform real-time estimation by integrating ampere-hour integral and extended Kalman filter. include: Step B1, construct the battery model; In the traditional second-order RC equivalent circuit model, a time-varying virtual resistor is connected in parallel. The virtual resistor is not a real physical component, but rather a comprehensive parameter used to characterize dynamic losses not covered by the model. The model is described as follows: The terminal voltage equation is as follows: ； Polarization voltage differential equation: ; ； in, I is the terminal voltage, and I is the current. Determined by the curve provided in step 1, For ohmic internal resistance, The parameters of the two RC elements in the traditional second-order RC equivalent circuit model of the battery are resistor 1, capacitor 1, resistor 2, and capacitor 2, respectively. Step B2, online parameter identification; using an adaptive forgetting factor. The RLS algorithm identifies model parameters online in real time: ; * , * ; The forgetting factor update rule is as follows: ; in, Let λ(k) be the change in current between time k and time k-1, and β be the adjustment coefficient. When the current changes drastically (dynamic operating condition), λ(k)→1, and the algorithm quickly forgets the old data and tracks the new dynamic. When the current is stable, λ(k) decreases, which enhances the smoothness and stability of parameter estimation. Step B3, Extended Kalman filter fusion estimation; State equations: with SOC and two polarization voltages as state variables. ; ； ； ； in, For rated capacity, Coulomb efficiency; Observation equation: EKF Iteration Process: EKF uses the result of the ampere-hour integration as the predicted value, and the difference between the model predicted voltage y(k) and the actual measured voltage z(k) as the observation information. The state vector is then optimally corrected through the Kalman gain K.

7. The method according to claim 6, characterized in that, Step C, in dynamic load mode, triggers a no-static calibration mechanism, including: Step C1, Dynamic mode detection; Real-time monitoring of load current sequence Calibration is triggered when a dynamic pattern matching the "excitation-relaxation" characteristic is detected; a typical pattern is: a continuous... stable current Then, immediately following; another continuous tiny current The triggering condition is: and 0, and and Sufficiently long enough for the polarization voltages of both stages to reach a quasi-steady state; Step C2, online OCV estimation and forced calibration; Data is collected at the end of both phases of the detected dynamic pattern. and ; Estimate polarization voltage variation using currently identified model parameters ; The core OCV estimation method is as follows: ； Using the estimated Query the initial OCV-SOC curve generated in step 1, and use reverse interpolation to obtain a high-confidence reference SOC value. ; Will As absolute observations injected into the EKF system: at the corresponding time, the observation equations of the temporary EKF are changed to... The observed value is set as It assigns a very small observation noise covariance; EKF will strongly pull the SOC value in the state vector towards [the desired value] in this iteration. This allows for one-click clearing of the accumulated error of the ampere-hour integral; after calibration, the system immediately returns to normal voltage observation mode.

Images