Method for optimizing dynamic balancing window of power battery pack

Abstract

The application discloses a kind of power battery pack dynamic equalization window optimization method, comprising: collecting battery pack real-time operation data, determine the continuous aging characteristic parameter containing capacity attenuation rate;Utilize state estimator to calculate single charge state, and total estimation error is decomposed into four independent multi-source uncertainty components of measurement noise, model mismatch, aging parameter time-varying and temperature inhomogeneity;Based on the source heterogeneity of uncertainty component, calculate the differentiated safety margin, combine the continuous aging characteristic parameter, and construct the dynamic equalization window parameter of adaptation single attenuation state by continuous mapping function;Based on the window parameter, generate equalization control instruction.The application realizes the accurate quantification of safety boundary and the smooth self-adaptation of equalization strategy by multi-source uncertainty decomposition and continuous aging factor shaping, solves the problem of undifferentiated error source and strategy switching jitter, and improves the equalization safety and efficiency of whole life cycle.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of battery management, and particularly relates to a dynamic equalization window optimization method for a power battery pack. BACKGROUND

[0002] The consistency of a power battery pack can determine the available capacity and safety of an energy storage system, and active or passive equalization technology can be used to solve the inconsistency of single cells. In equalization control, dynamically adjusting the window parameters of equalization start and stop can balance the equalization efficiency, reduce energy loss, and prolong the battery life, which belongs to the research direction of battery management system (BMS) algorithms.

[0003] Existing equalization strategies, on the one hand, rely on model-based state estimation, such as extended Kalman filter (EKF), to obtain the state of charge (SOC), and set the equalization window according to fixed empirical thresholds or discrete classification based on the state of health (SOH), such as distinguishing new batteries from old batteries. On the other hand, the error covariance of state estimation is introduced to dynamically adjust the boundary, but all sources of error are usually treated as a single statistical uncertainty for unified processing, and a fixed safety factor is used to determine the safe operating range.

[0004] However, the adaptability of the prior art under complex working conditions still has the risk of misjudgment caused by the homogenization processing of uncertain sources and the problem of control discontinuity caused by discrete aging strategies. Therefore, further research and innovation are needed to solve the above problems existing in the prior art. SUMMARY

[0005] The application provides a dynamic equalization window optimization method for a power battery pack.

[0006] Technical scheme: According to one aspect of the application, a dynamic equalization window optimization method for a power battery pack includes:

[0007] Collecting real-time operation data of the power battery pack, and determining aging characteristic parameters of each battery cell according to the real-time operation data;

[0008] Estimating the state of charge estimation value of each battery cell based on the real-time operation data and the aging characteristic parameters, and calculating a multi-source uncertainty component of the state of charge estimation value;

[0009] Calculating the differentiated safety margin of each battery cell according to the multi-source uncertainty component, and constructing a dynamic equalization window parameter adapted to the attenuation state of the single cell in combination with the aging characteristic parameters; the dynamic equalization window parameter at least includes an equalization trigger threshold and an equalization termination threshold;

[0010] Comparing the state of charge estimation value with the dynamic equalization window parameter to generate an equalization control instruction, and driving the equalization circuit to act.

[0011] Beneficial effects: The present application realizes accurate quantification of safety boundaries and smooth self-adaptation of balanced strategies by multi-source uncertainty decomposition and continuous aging factor shaping, solves the problems of undifferentiated error sources and policy switching jitter, and improves the balanced safety and efficiency throughout the life cycle. Related technical effects will be described in detail below in conjunction with specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 A flowchart of a power battery pack dynamic balancing window optimization method provided for an embodiment of the present application.

[0013] Figure 2 A flowchart of calculating multi-source uncertainty components provided for an embodiment of the present application.

[0014] Figure 3 A flowchart of calculating differential safety margins provided for an embodiment of the present application.

[0015] Figure 4 A flowchart of spatial correlation correction provided for an embodiment of the present application.

[0016] Figure 5 A flowchart of establishing a corresponding relationship between a comprehensive attenuation factor and a target state of charge interval using a continuous mapping function provided for an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0018] It should be noted that the terms first, second, etc. in the specification of the present application and in the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms include and have and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to the clearly listed steps or units, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] To solve the above problems, the applicant has conducted in-depth search and analysis, and found that:

[0020] Specifically, the existing state estimator only outputs the total error variance, and cannot distinguish between random measurement noise (low risk) and systematic model mismatch or parameter drift (high risk), resulting in a safety boundary that is too conservative in some operating conditions and too aggressive in the later stages of aging. In addition, the discrete classification method based on health / medium / severe makes the battery prone to strategy jump due to parameter perturbation near the category critical point, causing chattering of the equalization control; and the existing method often ignores the influence of aging rate, and cannot protect the rapidly deteriorating single body in advance.

[0021] To solve these problems, in combination with Figures 1 to 5 The present application is specifically illustrated by the following embodiments.

[0022] An example provides an exemplary scheme of a power battery pack dynamic equalization window optimization method. This embodiment constructs a closed-loop control system, covering the whole process from bottom-layer data acquisition to equalization instruction generation, introduces multi-source uncertainty decomposition and continuous aging characteristics, solves the problem of conservative or aggressive equalization strategy caused by the use of fixed safety factor, and realizes the dynamic adaptation of safety boundary and battery state.

[0023] Specifically, the embodiment can perform the following steps:

[0024] Step 101, collecting real-time running data of the power battery pack, and determining the aging characteristic parameters of each battery monomer according to the real-time running data.

[0025] In this embodiment, the real-time running data specifically refers to the physical quantity sequence collected from the analog front end of the battery management system (BMS), which includes but is not limited to the end voltage sequence, working current sequence, monomer surface temperature sequence and environmental temperature of each battery monomer. In addition, the system also needs to load the pre-configured battery pack configuration data, which includes the monomer rated capacity, rated internal resistance and topology structure information of the battery module. In order to make the data available, the sliding window median filtering algorithm can be used to remove abnormal outliers in the original signal during the collection process, and the Kalman filtering preprocessing method is used to smooth the measurement noise.

[0026] On this basis, the process of determining the aging characteristic parameters specifically involves modeling the battery capacity attenuation trajectory. Optionally, this embodiment can use a piecewise exponential decay model to describe the capacity degradation law of the battery monomer. Specifically, the historical charge and discharge cycle data is used to fit the capacity attenuation model in the following form:

[0027] Q(N)=Q _nom -a×N-b×N 1 / 2 ;

[0028] wherein N represents the equivalent cycle number, a is a linear decay coefficient representing the active lithium loss rate during normal use; b is an early decay coefficient representing the rapid capacity decline caused by the growth of the solid electrolyte interface film (SEI) in the early stage of battery life. That is, Q(N) represents the actual capacity of the battery after N equivalent cycles, Q _nom The coefficients a and b are obtained by online identification or table lookup, thereby constituring the aging characteristic parameter, which is used to quantify the health state difference of different monomers in the subsequent steps.

[0029] In step 102, the state of charge estimation value of each battery monomer is estimated based on the real-time operation data and the aging characteristic parameter, and a multi-source uncertainty component of the state of charge estimation value is calculated.

[0030] In this embodiment, considering the strong nonlinear characteristics of lithium-ion batteries, an unscented Kalman filter (UKF) algorithm can be selected as a state observer. Specifically, a state space model containing the state of charge (SOC), ohmic internal resistance and polarization voltage is established, the terminal voltage sequence and working current sequence in the real-time operation data are taken as observation inputs, and the aging characteristic parameter is mapped as a time-varying capacity parameter in the model, and the state of charge estimation value of each battery monomer is recursively updated.

[0031] Further, this step not only outputs the mean value of the state estimation, but also outputs a multi-source uncertainty component describing the estimation confidence. In this embodiment, the total estimation error variance is decomposed into multiple components with independent physical meanings by signal decomposition technology, which correspond to sensor noise, model parameter deviation, parameter drift caused by aging and thermal field inhomogeneity, respectively. Through this decomposition, the system can identify the main source of current estimation uncertainty, and provide a basis for subsequent targeted safety margin adjustment.

[0032] In step 103, the differential safety margin of each battery monomer is calculated according to the multi-source uncertainty component, and a dynamic equalization window parameter adapted to the attenuation state of the monomer is constructed in combination with the aging characteristic parameter. The dynamic equalization window parameter at least includes an equalization trigger threshold and an equalization termination threshold.

[0033] In this embodiment, in the calculation of the differential safety margin, the statistical uncertainty is converted into a physical safety margin. According to the preset confidence level, such as 99% confidence, the system derives the upper and lower limits of the state of charge allowed to run at the current time for each monomer by using the comprehensive standard deviation calculated by the multi-source uncertainty component.

[0034] Meanwhile, the process of constructing the dynamic balancing window parameters introduces an adaptive adjustment to the battery degradation status. For the single cells with good health status, the system generates a wider balancing target interval, allowing them to undertake more energy throughput tasks; for the single cells with serious degradation characteristics, the system automatically shrinks their balancing target interval and offsets the center thereof to the neutral state of charge (e.g., 50% SOC), avoiding long-time operation in the high-voltage or low-voltage interval. On this basis, the generated dynamic balancing window parameters include the trigger threshold for starting balancing and the termination threshold for stopping balancing for each single cell. In other words, the dynamic balancing window parameters include the balancing trigger threshold and the balancing termination threshold.

[0035] Step 104, comparing the state of charge estimation value with the dynamic balancing window parameters to generate a balancing control instruction to drive the balancing circuit to act.

[0036] In this embodiment, an event-triggering mechanism can be optionally used to reduce the computational load and reduce the switching loss of the balancing circuit. Further, the system calculates the deviation of the state of charge estimation value of each single cell from the center of the target balancing interval in real time. Only when the absolute value of the deviation exceeds the balancing trigger threshold, a valid balancing request is generated.

[0037] In other words, only when the absolute value of the deviation exceeds the balancing trigger threshold, the generated balancing request is valid.

[0038] Next, the system generates a balancing control instruction according to the balancing topology structure, such as passive dissipation resistance or active transfer DC-DC converter. The balancing control instruction includes the balancing channel number to be started, the balancing current size setting value, and the expected balancing duration. By driving the balancing circuit to act, the redistribution of the energy inside the battery pack is realized until the deviation falls within the balancing termination threshold.

[0039] Another example describes an optional implementation of differential safety margin calculation based on multi-source uncertainty decomposition. That is, by uncertainty decomposition and space correction, the differential safety margin with working condition adaptability and spatial consistency is calculated.

[0040] In this embodiment, it specifically includes:

[0041] Step 201, the multi-source uncertainty components at least include the following independently calculated components: a measurement noise component representing the random error of sensor measurement; a model mismatch component representing the structural deviation of the battery model; an aging parameter time-varying component representing the influence of the change of the battery capacity parameter on the state estimation; and a temperature non-uniformity component representing the influence of the temperature gradient between single cells.

[0042] In this embodiment, the total uncertainty variance σ 2 _totalOrthogonal decomposition into measurement noise uncertainty variance σ 2 _meas , model mismatch uncertainty variance σ 2 _model , aging parameter time variation uncertainty variance σ 2 _aging , and temperature non-uniformity uncertainty variance σ 2 _thermal This decomposition enables the system to distinguish error risks of different natures. For example, measurement noise is usually a high-frequency random signal with zero mean, which can be suppressed by filtering and has a low risk weight; while model mismatch often manifests as a systematic bias and has a high risk weight.

[0043] In step 202, the multi-source uncertainty components are calculated, including: constructing a voltage residual sequence according to the difference between the single-cell voltage in real-time operation data and the model-predicted voltage; performing power spectrum analysis on the high-frequency component of the voltage residual sequence to calculate the measurement noise component; extracting the low-frequency trend component of the voltage residual sequence and calculating the model mismatch component based on the offset of the low-frequency trend component; calculating the aging parameter time variation component according to the estimated variance of the aging characteristic parameter and the sensitivity coefficient of the state of charge to capacity change; and calculating the deviation between the single-cell temperature in real-time operation data and the module average temperature, and calculating the temperature non-uniformity component based on the deviation.

[0044] In other words, calculating the multi-source uncertainty components includes:

[0045] constructing a voltage residual sequence including high-frequency components and low-frequency trend components according to the difference between the single-cell voltage in real-time operation data and the model-predicted voltage;

[0046] performing power spectrum analysis on the high-frequency component to calculate the measurement noise component;

[0047] extracting the low-frequency trend component and calculating the model mismatch component based on the offset of the low-frequency trend component;

[0048] calculating the aging parameter time variation component according to the estimated variance of the aging characteristic parameter and the sensitivity coefficient of the state of charge to capacity change;

[0049] calculating the deviation between the single-cell temperature in real-time operation data and the module average temperature, and calculating the temperature non-uniformity component based on the deviation.

[0050] Specifically, the voltage residual r _V (k) = V _meas (k) - V _pred (k) is calculated.

[0051] wherein r _V (k) represents the battery single-cell voltage residual at the kth sampling time; V _meas(k) is the measured value of the single cell voltage at this moment by the acquisition device; V _pred (k) is the predicted value of the single cell voltage at this moment calculated by the voltage estimation model.

[0052] For the measurement noise component, the high frequency component in r _V (k) is extracted by using fast Fourier transform (FFT) or wavelet transform, and the power spectral density integral is calculated to obtain σ 2 _meas . For the model mismatch component, the low frequency trend item r _V (k) in r _trend (k) is extracted by using a moving average filter, and the error propagation formula σ 2 _model = (r _trend (k) / G _ocv ) 2 is mapped back to the SOC domain, where G _ocv is the slope of the open circuit voltage curve.

[0053] On the one hand, for the time-varying component of the aging parameter, the uncertainty σ 2 _Q of the capacity estimation (which can be obtained from the fitting residual of the aging model) and the sensitivity of the SOC to the capacity under the current working condition are calculated, and the formula can be expressed as:

[0054] σ 2 _aging = (ЯSOC / ЯQ) 2 × σ 2 _Q ;

[0055] Where Я corresponds to the partial derivative, or the partial derivative of the state of charge SOC to the capacity Q, which can be obtained by taking the derivative of the SOC calculation formula.

[0056] For example, at the end of charging and discharging, the sensitivity coefficient is large, and this component increases. For the temperature inhomogeneity component, the difference between the single cell temperature T _i and the module average temperature T _avg is calculated, and the empirical formula σ 2 _thermal = k _T × (T _i -T _avg ) 2 is used to estimate, where k _T is the temperature influence coefficient.

[0057] In step 203, the differentiated safety margin is calculated, including: identifying the current working condition category of the power battery pack, determining the working condition modulation weight corresponding to the working condition category; using the working condition modulation weight to weight and sum the measurement noise component, the model mismatch component, the aging parameter time-varying component and the temperature non-uniformity component to obtain a weighted equivalent uncertainty; calculating the proportion of the aging parameter time-varying component in the weighted equivalent uncertainty, modifying the pre-configured reference safety factor based on the proportion to obtain a working condition adaptive safety factor; and determining the differentiated safety margin based on the working condition adaptive safety factor and the weighted equivalent uncertainty.

[0058] In the embodiment, the system predefines a weight vector w c =[w _meas , w _model , w _aging , w _thermal ] for different working condition categories, where w _meas , w _model , w _aging , w _thermal correspond to the measurement noise uncertainty weight coefficient, the model mismatch uncertainty weight coefficient, the aging parameter time-varying uncertainty weight coefficient and the temperature non-uniformity uncertainty weight coefficient respectively.

[0059] For example, in the fast charging working condition, the model polarization error and thermal effect are large due to large current, and the weight vector can be set as [1.0, 1.5, 1.2, 1.8]; while in the standing working condition, the weight vector can be set as [0.8, 1.0, 1.0, 0.5]. Based on this, the weighted equivalent uncertainty σ _weighted =(∑w _j ×σ 2 _j ) 1 / 2 .

[0060] Further, in order to preventively control the potential risk brought by aging, the aging component proportion Ratio _aging =(σ 2 _aging ) / (σ 2 _weighted ) is calculated. The formula k _safe =k _base ×(1+λ×Ratio _aging ) is used, where k _base corresponds to the modified reference safety factor. For example, if k _base =3, corresponding to 99.7% confidence, λ=0.5, when the aging uncertainty is dominant, i.e. Ratio _aging =0.8, the modified safety factor k _safeis raised to 4.2. On this basis, the SOC boundary corresponding to the differentiated safety margin is calculated as: SOC _safe = SOC _base ± k _safe × σ _weighted ; or in other words, SOC _base is the differentiated safety SOC boundary.

[0061] At step 204, before calculating the differentiated safety margin, the method further includes a spatial correlation correction, specifically comprising: obtaining a pre-stored cell adjacency matrix describing the spatial position relationship of the battery cells; based on the cell adjacency matrix, calculating a neighborhood weighted average value of the current battery cell, the neighborhood weighted average value representing the average uncertainty level of the physically adjacent cells; comparing the total uncertainty of the current battery cell with the neighborhood weighted average value; if the total uncertainty of the current battery cell is lower than the neighborhood weighted average value, performing an upward correction operation on the total uncertainty of the current battery cell using the neighborhood weighted average value, and calculating the differentiated safety margin using the corrected total uncertainty.

[0062] In this step, the cell adjacency matrix A is an N x N matrix, and if cell i is physically adjacent to cell j, then element A _ij = 1, otherwise 0. For cell i, the neighborhood weighted average value σ _neighbor_i = ∑A _ij × σ _weighted_j / ∑A _ij ; σ _weighted_j corresponds to the weighted equivalent uncertainty of the jth adjacent cell. The thermal / electrical coupling characteristics within the battery pack are used for group intelligence collaborative verification. If the uncertainty calculated by a certain cell is less than the average level of its surrounding neighbors, the estimator of the cell may be in an overconfident state.

[0063] Further, the system performs an upward correction operation, i.e. σ _corrected_i = max(σ _weighted_i , α × σ _neighbor_i ), where α is a spatial correlation coefficient, usually taking a value between 0.8 and 1.0, max(…) is a maximum value function, and σ _corrected_i represents the uncertainty standard deviation of the i th battery cell after spatial correlation correction, and σ _weighted_i represents the weighted equivalent uncertainty standard deviation of the i th battery cell, or in other words, the weighted equivalent uncertainty, or the weighted equivalent standard deviation. Through this correction, the risk of abnormally low estimation is eliminated, and the safety strategy of the entire group of batteries remains spatially consistent.

[0064] According to one aspect of the present application, a numerical calculation example is provided to illustrate the process of multi-source uncertainty decomposition and differentiated safety margin calculation.

[0065] Suppose the real-time running data of a certain battery monomer under fast charging condition is as follows:

[0066] Monomer voltage V _meas is 3.85V, the model predicted voltage V _pred is 3.83V, the monomer temperature T _i is 32℃, the module average temperature T _avg is 30℃.

[0067] Calculate the voltage residual: r _V =V _meas -V _pred =3.85-3.83=0.02V.

[0068] Through spectral analysis, determine the measurement noise standard deviation σ _meas of the high-frequency component =0.005.

[0069] Extract the low-frequency trend component offset as 0.015V, and suppose the open circuit voltage curve slope G _ocv =0.5V / %SOC, then the model mismatch component σ _model =0.015 / 0.5=0.03.

[0070] Suppose the capacity estimation variance σ² _Q =0.01, and the SOC sensitivity coefficient to capacity is 0.8, then the aging parameter time-varying component σ _aging =0.8×0.1=0.08.

[0071] Calculate the temperature deviation ΔT=32-30=2℃, and suppose the temperature influence coefficient k _T =0.01, then the temperature non-uniformity component σ _thermal =(0.01×4) 0.5 =0.2.

[0072] Suppose the weight vector of the fast charging condition is [1.0, 1.5, 1.2, 1.8], calculate the weighted equivalent uncertainty:

[0073] σ _weighted =(1.0×0.005 2 +1.5×0.03 2 +1.2×0.08 2 +1.8×0.2 2 ) 0.5 ≈0.28.

[0074] Calculate the aging component ratio: Ratio _aging =0.08 2 / 0.28 2 ≈0.082.

[0075] Assuming the reference safety factor k _base = 3, λ = 0.5, the corrected safety factor: k _safe = 3 x (1 + 0.5 x 0.082) ≈ 3.12.

[0076] If the current SOC estimation value is 60%, the safety boundary is 60% ± 3.12 x 0.28 ≈ 60% ± 0.87%, i.e. [59.13%, 60.87%].

[0077] Through the above calculation, the safe operation interval of the single body under fast charging condition is quantified. Compared with the method of using fixed safety factor (such as uniformly taking 3σ), the safety factor is dynamically adjusted according to the proportion of aging component in the present embodiment, which avoids excessive conservatism while ensuring safety.

[0078] In another example, a specific implementation process of dynamic window construction and hierarchical triggering based on continuous attenuation factor is described. Compared with the traditional extensive management based on discrete health state classification (such as only distinguishing between healthy and scrapped), the present embodiment provides a control strategy based on continuous variable. By using the continuously changing comprehensive attenuation factor, the shaping and control of the equalization window of the battery single body are realized.

[0079] Correspondingly, the aging characteristic parameters include capacity attenuation cumulative amount and capacity attenuation rate term; the dynamic equalization window parameters suitable for the attenuation state of the single body are constructed, including:

[0080] Step 301, calculating the weighted sum of the capacity attenuation cumulative amount and the capacity attenuation rate term to obtain the comprehensive attenuation factor representing the health degree and deterioration trend of the battery single body.

[0081] In the present embodiment, the capacity attenuation cumulative amount refers to the loss proportion of the nominal capacity of the battery single body at the current time relative to the initial rated capacity, which is usually represented as (1-SOH _i ), wherein SOH _i is the current health state of the i-th single body. The capacity attenuation rate term refers to the health state decline amplitude per cycle or per unit time, denoted as |dSOH _i / dN|.

[0082] Further, the introduction of the rate term can distinguish between stable aging and accelerated aging. Even if the current SOH of two single bodies is the same, the single body with faster attenuation rate has poorer internal electrochemical stability, such as micro-short circuit or lithium precipitation risk, and therefore needs more conservative equalization strategy.

[0083] On this basis, the calculation formula of the comprehensive attenuation factor η _i is: η _i = (1-SOH _i )+ μ x |dSOH_i / dN|×N _pred . wherein μ is a rate weight coefficient, used to adjust the sensitivity to the aging trend; N _pred is a preset prediction window length, for example, 100 future cycles. wherein the comprehensive attenuation factor is a continuous scalar value, the larger the value, the deeper the aging degree or the greater the deterioration trend of the monomer.

[0084] wherein the value range of the rate weight coefficient μ can be set to 0.5 to 2.0. When μ takes a smaller value, such as 0.5, the system focuses on considering the current capacity attenuation cumulative amount; when μ takes a larger value, such as 2.0, the system is more sensitive to the attenuation rate, and can identify the accelerated aging monomer earlier; the value of the prediction window length N _pred can be 50 to 200 cycles.

[0085] In some scenarios, μ can be valued at 1.0, and N _pred can be valued at 100 cycles.

[0086] Step 302, a continuous mapping function is used to establish a corresponding relationship between the comprehensive attenuation factor and the target state of charge interval, and a personalized target state of charge interval is generated for each battery monomer based on the corresponding relationship;

[0087] wherein the dynamic balancing window parameter is determined based on the upper and lower boundaries of the target state of charge interval.

[0088] In this embodiment, the continuous mapping function is a pre-defined mathematical relationship for converting the dimensionless comprehensive attenuation factor into a specific state of charge physical constraint. The target state of charge interval is the SOC range that the system expects the monomer to run in the long term.

[0089] Unlike sharing a set of balancing targets with all monomers (such as balancing uniformly to the average value), this step generates a personalized interval for each monomer that matches its aging state fingerprint. For example, for a η _i small healthy monomer, its target interval is wider, such as 20% to 80%; while for a η _i larger aging monomer, its target interval is compressed, such as 45% to 55%, limiting its charge and discharge depth and delaying further aging.

[0090] wherein the continuous mapping function is used to establish a corresponding relationship between the comprehensive attenuation factor and the target state of charge interval, including: with the increase of the comprehensive attenuation factor, a first exponential mapping function is used to control the center of the target state of charge interval to deviate from the preset neutral state of charge, reducing the operation risk of the attenuation monomer in the high stress interval.

[0091] Specifically, the first exponential mapping function is used to determine the center position SOC _center_iSOC _opt is the optimal operating point for a healthy cell, e.g. 60%, _neutral is the neutral state of charge at which the cell's electrochemical properties are most stable, usually taken as 50%. The mapping relationship can be expressed as:

[0092] SOC _center_i = SOC _neutral + (SOC _opt - SOC _neutral ) x exp(-k _center x η _i ) ;

[0093] where k _center is the central shift decay constant. When the cell is very healthy, i.e. η _i tends to 0, its target center is maintained at SOC _opt ; as the degree of aging increases, η _i increases, the exponential term rapidly decays, forcing the target center to converge to SOC _neutral . This mechanism makes the aging cell always work in the region with the most gentle voltage platform and the least side reactions.

[0094] Further, as the comprehensive decay factor increases, the second exponential mapping function is used to control the width of the target state of charge interval to shrink, imposing more stringent operating range restrictions on the decaying cell.

[0095] In this step, the second exponential mapping function is used to determine the half-width ΔSOC _i of the target interval. Let ΔSOC _max be the maximum allowable half-width, e.g. 30%, then the mapping relationship is:

[0096] ΔSOC _i = ΔSOC _max x exp(-k _width x η _i ) ;

[0097] where k _width is the width shrinkage decay constant. This formula realizes the continuous shrinkage function of the window width. For example, if a cell has a sudden increase in decay rate due to failure, its η _i rapidly increases, and ΔSOC _i exponentially decreases, the system will immediately restrict the available SOC range of the cell, playing a passive protection role. Based on this, the individualized target state of charge interval of the cell is determined as [SOC _center_i - ΔSOC _i , SOC _center_i + ΔSOC _i ].

[0098] wherein the center offset decay constant has a value range of 2-8, and an optional value of 5. The width contraction decay constant has a value range of 3-10, and an optional value of 6. k _center and k _width The larger the target interval is, the more sensitive the target interval is to the aging state, and is suitable for a scenario with high requirement on battery life; otherwise, it is suitable for a scenario with high requirement on available capacity utilization.

[0099] On this basis, the upper and lower boundaries of the generated target state of charge interval are intersected with the differentiated safety margin, so that the dynamic balancing window parameter is located within the range defined by the differentiated safety margin. This is used to guarantee safety.

[0100] Correspondingly, the interval calculated based on aging only reflects the requirement of life optimization, while the differentiated safety margin [SOC _safe_min_i , SOC _safe_max_i ] calculated based on uncertainty reflects the requirement of physical safety. In order to meet both, intersection operation must be performed, i.e., the balancing lower limit θ _stop_down_i = max(SOC _center_i - ΔSOC _i , SOC _safe_min_i ), and the balancing upper limit θ _stop_up_i = min(SOC _center_i + ΔSOC _i , SOC _safe_max_i ). If the intersection is empty, it indicates that the single body has no safe operating space, and the system will report a fault and prohibit the single body from participating in any balancing operation.

[0101] Further, a balancing control instruction is generated, including:

[0102] Optionally, a difference value of the state of charge estimation value relative to the boundary of the target state of charge interval is calculated, and the difference value is divided by the half-width of the target state of charge interval to obtain a normalized deviation degree.

[0103] In the embodiment, the normalized deviation degree d _i is used to measure the severity of the deviation of the single body from its comfort zone. If the state of charge estimation value SOC _est_i of the current i single body is higher than the upper limit Э _max of the target interval, then d _i = (SOC _est_i - Э _max ) / ΔSOC _i ; if it is lower than the lower limit Э _min of the target interval, then d _i = (Э _min - SOC _est_i ) / ΔSOC _i ; and if it is within the interval, then d _i= 0. The normalization processing can eliminate the scale difference caused by different monomer window widths, so that the equalization control strategy can be designed based on a unified dimensionless index.

[0104] Optionally, the grading equalization strength is determined based on the normalized deviation degree; when the normalized deviation degree is less than a preset first threshold, a linearly increasing equalization current instruction is generated;

[0105] When the normalized deviation degree exceeds a preset second threshold, an equalization current instruction with maximum strength is generated.

[0106] Correspondingly, the system presets two thresholds, for example, the first threshold d _th1 is 0.2, and the second threshold d _th2 is 0.8. When d _i is less than d _th1 , it is considered that the deviation is slight, and the equalization strength coefficient p _i is linearly increased, that is, p _i = k _p × d _i , at this time, the equalization circuit works with a small current, which is used for fine tuning consistency and reducing energy consumption; when d _i is between d _th1 and d _th2 , the equalization strength is further improved; when d _i exceeds d _th2 , it is considered that there is a serious consistency risk, and the system directly generates an equalization current instruction with maximum strength, that is, the equalization current is equal to the maximum value allowed by the circuit, which is used to eliminate the deviation as soon as possible. The above grading mechanism balances the equalization efficiency and energy consumption.

[0107] According to one aspect of the present application, taking a battery pack of a certain pure electric vehicle as an example, the working process of the dynamic equalization window optimization method in actual application is described.

[0108] Suppose the battery pack contains 96 monomers, which are divided into 12 modules. After 3 years of use, the health states of the monomers are differentiated: the SOH of most monomers is about 92%, but the SOH of 3 monomers in the high temperature area decreases to 85%, and their capacity decay rates are significantly higher than the average level.

[0109] Correspondingly, the system collects the voltage, current and temperature data of each monomer, and calculates the aging characteristic parameters of each monomer based on the historical charging and discharging data. For the 3 accelerated aging monomers, the system calculates a higher comprehensive decay factor, for example, about 0.25, while the comprehensive decay factor of the healthy monomers is lower, for example, about 0.08.

[0110] Further, the system generates individual target SOC interval for each cell using a continuous mapping function. The target interval for healthy cells is [25%, 75%], while the target interval for accelerated aging cells is shrunk to [42%, 58%] with the center shifted to 50%.

[0111] In the equalization control loop, when the SOC of a healthy cell reaches 78%, the system determines that it deviates from the target interval, but the normalized deviation degree d = (78% - 75%) / 25% = 0.12 < 0.2, so a small current equalization instruction is generated to slowly adjust at low energy consumption.

[0112] At the same time, the SOC of an accelerated aging cell is 63%, and the normalized deviation degree = (63% - 58%) / 8% = 0.625, so the system generates a medium-intensity equalization current instruction to preferentially pull the SOC of the cell back to the safe interval.

[0113] Compared with the traditional uniform threshold equalization strategy, such as equalizing all cells to within ±5%, the method of the present application enables accelerated aging cells to always operate in the region with the flattest voltage platform and the least side reactions. After 6 months of tracking tests, the capacity decay rate of the above cells is reduced, and the utilization rate of the available capacity of the entire battery is improved.

[0114] In some embodiments, an optional technical solution for thermal-electric-aging coupling correction is provided. In particular, how to introduce thermal safety constraints in the dynamic window construction process to make physical corrections so that the equalization operation itself does not trigger the risk of battery overheating. Specifically, it includes:

[0115] Step 401, estimate the instantaneous power loss of the equalization operation and predict the temperature rise according to the equalization current and the battery internal resistance determined by the dynamic equalization window parameters.

[0116] In this embodiment, the equalization current I _bal_i is calculated according to the hierarchical equalization intensity. The battery internal resistance R _int_i is derived from the real-time output of the state estimation module. According to Joule's law, the instantaneous power loss P _loss_i generated inside the battery during the equalization process is calculated as:

[0117] P _loss_i = (I _bal_i ) 2 × R _int_i ;

[0118] Further, the predicted temperature rise ΔT _pred_i in the predicted equalization duration t _bal is estimated using a simplified thermal model. The thermal model can be represented as:

[0119] ΔT _pred_i = R_th xP _loss_i x(1-exp(-t _bal / τ _th ));

[0120] where R _th is the thermal resistance from the battery cell to the environment, and τ _th is the thermal time constant. The thermal parameters can be stored in the BMS as pre-configuration data.

[0121] In other words, R _th is the thermal resistance from the battery cell to the environment, with a typical value range of 1-5 °C / W; and τ _th is the thermal time constant, with a typical value range of 100-500 seconds.

[0122] Step 402, calculate the difference between the pre-configured maximum safe temperature and the current battery cell temperature to obtain the temperature safety margin.

[0123] In this step, the maximum safe temperature T _max is the red line given by the battery manufacturer, for example, 45 °C or 60 °C. The system collects the current cell surface temperature T _curr_i in real time, and calculates the temperature safety margin T _margin_i = T _max - T _curr_i . This margin represents the allowed temperature rise space of the battery before touching the safety bottom line.

[0124] Step 903, when the predicted temperature rise exceeds the temperature safety margin, the dynamic balancing window parameters are corrected by contraction to limit the maximum allowed balancing current, so that the corrected predicted temperature rise is within the temperature safety margin.

[0125] Specifically, if ΔT _pred_i > T _margin_i , it means that performing balancing according to the current window parameters will cause the battery to overheat. At this time, the system must reversely calculate the maximum balancing current limit value I _limit_i allowed. The specific reverse calculation formula is:

[0126] I _limit_i = [(T _margin_i ) / (R _th x R _int_i x (1-exp(-t _bal / τ _th )))] 1 / 2 .

[0127] Next, the system forces the balancing current set value of this cell to be truncated to min(I _bal_i , I _limit_i), min(...) corresponds to the minimum function. In another alternative implementation, if the equalization current is not adjustable, such as fixed resistance equalization, the system shortens the equalization duration t _bal , reduces the total heat generation until the thermal safety constraint is satisfied.

[0128] In some other embodiments, an exemplary scheme of multi-objective optimization based on probabilistic risk constraint is provided. This embodiment models the determination process of dynamic equalization window as a constrained nonlinear programming problem, which is suitable for a central computing unit or a cloud BMS platform with strong computing power. The following steps can be used to achieve this embodiment:

[0129] Step 501, a multi-objective cost function is constructed, which at least includes a variance cost term representing state of charge consistency, an energy consumption cost term representing equalization energy consumption, and a life cost term representing battery life loss.

[0130] In other words, a multi-objective cost function is called, which at least includes a variance cost term representing state of charge consistency, an energy consumption cost term representing equalization energy consumption, and a life cost term representing battery life loss.

[0131] In this embodiment, the decision variable vector Θ to be optimized is defined as the set of equalization trigger and termination thresholds of all monomers. The cost function J(Θ) = w _var ×J _var +w _ene ×J _ene +w _life ×J _life ;

[0132] wherein, J _var term calculates the SOC variance after running for a predicted time domain under the control of the parameter Θ, which is used to make the consistency better; J _ene term accumulates the equalization energy loss in the whole process, i.e. ΣI _bal ×V×t, which is used to reduce the energy consumption; J _life term estimates the residence time of each monomer in the high state of charge or high temperature interval based on the aging model, which is used to prolong the life. The weight coefficients w _var , w _ene , w _life can be dynamically adjusted according to the current operating mode of the vehicle, such as energy saving mode or performance mode.

[0133] Step 502, a probabilistic risk constraint is called based on multi-source uncertainty components, which is used to limit the probability of the state of charge of the battery monomer exceeding the differentiated safety margin to be lower than the preset risk threshold.

[0134] In this step, the multi-source uncertainty component, i.e. the weighted equivalent standard deviation σ _weighted_i is converted into a probabilistic form. The hard safety boundary is transformed into a probabilistic form. The specific constraint inequality is expressed as:

[0135] P(SOC _i >SOC _limit_max )≤ε;

[0136] where P(…) is the probability calculation function, SOC _i represents the actual state of charge of the i-th battery cell, SOC _limit_max is the safe upper threshold of the state of charge of the cell, and ε is a preset risk threshold, representing the probability tolerance of the state of charge of the cell exceeding the safe upper limit.

[0137] Assuming that the estimation error follows a Gaussian distribution, the constraint can be converted into a deterministic form, i.e.:

[0138] SOC _mean_i +z _score ×σ _weighted_i ≤SOC _limit_max ;

[0139] where z _score is the standard normal distribution quantile corresponding to the preset risk threshold ε (e.g. 0.01%), that is, SOC _mean_i is the estimated mean value of the state of charge of the i-th cell; z _score is the standard normal distribution quantile corresponding to the preset risk threshold ε; and σ _weighted_i is the weighted equivalent uncertainty standard deviation of the i-th cell. The formula realizes the conversion from the probabilistic risk constraint to the deterministic control constraint.

[0140] The constraint makes the probability of any window parameter generated in the optimization process touching the safety bottom line in a statistical sense controlled at a low level.

[0141] Step 503, solving an optimization problem with the objective of minimizing a multi-objective cost function and satisfying a probabilistic risk constraint, to obtain optimal dynamic equilibrium window parameters.

[0142] In the embodiment, a sequential quadratic programming (SQP) algorithm or a genetic algorithm (GA) is used to solve the optimization problem. Since there can be a complex nonlinear relationship between the objective function and the decision variables, the system is based on the current state to make single-step or multi-step prediction, the optimization problem is converted into a standard form, and then the optimal solution that minimizes the objective function and satisfies all probabilistic constraints is iteratively searched. The optimal solution is directly analyzed as the dynamic equilibrium window parameters of each cell.

[0143] According to another aspect of the present application, an optional embodiment of the policy gradient based life cycle parameter adaptive update is described. In particular, the equalization window parameter is adaptively iterated with the full life cycle historical data. The problem that the initially set parameter mapping rule may fail as the battery aging characteristics fundamentally drift is solved. In the present embodiment, the following steps are included:

[0144] Step 601, dividing the full life cycle of the power battery pack into a plurality of life stages, and defining stage performance indicators including consistency improvement rate, total equalization energy consumption and out-of-bound risk count.

[0145] In the present embodiment, the cumulative charge and discharge ampere-hour or cumulative driving mileage is used as the division basis to divide the life cycle into a series of discrete stages, for example, one stage s for every 100 cycles. For each stage s, the system records and calculates the comprehensive score R _s . The score R _s is a weighted sum of multiple dimensions, that is:

[0146] R _s =α×(ΔVar _SOC )-β×(E _total )-γ×(N _risk );

[0147] Wherein, ΔVar _SOC is the average SOC variance improvement in the stage, that is, the positive benefit, E _total is the total equalization energy consumption, that is, the negative cost, N _risk is the number of times that the single cell touches the safety boundary in the stage, that is, the penalty term; α, β, γ are the corresponding weight coefficients.

[0148] Step 602, using the finite difference method, calculating the approximate gradient of the stage performance indicator with respect to the dynamic equalization window parameter according to the performance difference of the adjacent two historical version window parameters in the corresponding life stage.

[0149] That is, the dynamic equalization window parameter can also be called the window parameter vector.

[0150] In this step, the system stores the window parameter vector Θ _s-1 and the corresponding score R _s-1 of the last stage s-1, and the window parameter vector Θ _s and the score R _s of the current stage s. Since the analytical expression of the performance indicator with respect to the parameter cannot be obtained, the numerical difference is used to estimate the gradient in the present embodiment. The approximate gradient vector ▽R is calculated as:

[0151] ▽R≈(R _s -R _s-1) / (Θ _s -Θ _s-1 ).

[0152] This calculation assumes that the physical characteristics of the battery itself change little within two adjacent short-life stages, and the difference in performance is mainly caused by the adjustment of the window parameter vector.

[0153] Step 603, based on the approximate gradient, the dynamic equilibrium window parameters are updated using the gradient descent algorithm, and the updated parameters are used as the initial dynamic equilibrium window parameters of the next life stage.

[0154] This step performs specific update logic. The gradient ascent (if seeking maximum benefit) or gradient descent (if seeking minimum cost) formula is used to generate the window parameter vector Θ _s+1 of the next stage, which can be described as the following formula:

[0155] Θ _s+1 = Θ _s + η _learn ×▽R;

[0156] Where η _learn is the learning rate, used to control the step size of the update to prevent parameter oscillation. After the update is completed, the system must perform a constraint projection operation to limit each element in Θ _s+1 to the physically allowed range, for example, the trigger threshold must be greater than the termination threshold. Through iterative learning across stages, the equilibrium strategy can automatically find the optimal control parameters in the current state as the battery ages, just like biological evolution.

[0157] According to another aspect of the present application, an optional implementation method of data preprocessing and feature engineering is described. To provide guarantee for subsequent state estimation and uncertainty decomposition precision.

[0158] Step 701, collect real-time running data of the power battery pack, and determine the aging characteristic parameters of each battery monomer.

[0159] In this embodiment, for the common timestamp asynchronous problem in the original collected data, a time alignment method based on linear interpolation can be used. Specifically, the system takes the master clock of the battery management system as the reference to establish a unified discrete time axis. For channel data with different sampling frequencies such as voltage, current and temperature, the interpolation value at the reference time t is calculated by using the numerical values and time difference of the adjacent two sampling points through the linear equation y=y _0 +(y _1 -y _0 )×(t-t _0 ) / (t _1 -t _0 ).

[0160] where y represents the target data value obtained by linear interpolation at the reference time t, which can be used for voltage, current, temperature, and other channel data alignment at different sampling frequencies; y _0 is the original data value of the left end point of the interpolation interval, and its corresponding sampling time is t _0 ; y _1 is the original data value of the right end point of the interpolation interval, and its corresponding sampling time is t _1 ; t is the target time on the unified reference discrete time axis that needs to be aligned. By linearly fitting the values of adjacent sampling points, the synchronization calibration of different time stamp data is realized.

[0161] On this basis, a sliding window median filtering algorithm can be optionally used to eliminate transient impulse noise caused by electromagnetic interference. The system maintains a first-in-first-out queue with a length of N (e.g., N = 5), and calculates the median of the data in the queue as the effective measurement value at the current time, filtering out outliers that deviate from the statistical law.

[0162] Further, the process of determining the aging characteristic parameter depends on the accurate identification and analysis of the complete charging and discharging segment. The system identifies the start and end of the charging and discharging cycle by monitoring the reversal of the current direction. After identifying the complete charging segment, the system can use the ampere-hour integration method to calculate the cumulative charge Q _segment =∫I(t)dt; I(t) is the instantaneous charging current varying with time during the charging process. Further, the temperature correction coefficient k _T =exp(E _a / Ч×(1 / T _ref -1 / T _avg )) is used to normalize Q _segment , where E _a is the activation energy, Ч is the gas constant, and T _ref corresponds to the reference temperature, which is generally selected as the standard ambient temperature, such as 298.15 K, i.e., 25℃. This is used to eliminate the influence of temperature on capacity measurement.

[0163] In this embodiment, the system can also use a sliding window linear regression method. Specifically, the system stores a sequence of capacity estimation values for the last M cycles (e.g., M = 50). With the cycle number N as the independent variable and the capacity estimation value Q(N) as the dependent variable, a straight line Q(N) = c + k × N is fitted by the least squares method. The slope k obtained by fitting is the current capacity decay rate dSOH / dN. The rate extraction method based on statistical regression has stronger anti-noise ability than the single-point difference method, and can more accurately reflect the long-term trend of battery aging.

[0164] Or, Q(N) represents the capacity fitting estimation value of the battery after N equivalent cycles, which is a trend capacity index obtained by sliding window linear regression; c is the intercept term of the fitting straight line, corresponding to the initial capacity fitting reference value of the battery in the initial state (when the equivalent cycle number N = 0); k is the slope of the fitting straight line, which is the linear decay rate of the capacity of the battery, directly corresponding to the health state decay rate dSOH / dN; N is the equivalent cycle number of the battery, which is the independent variable of linear regression.

[0165] In yet other embodiments, optional implementations of part of the method of the present application are provided, which can also implement the present application in the case of limited computing resources, simplified sensor configuration or specific application scenarios.

[0166] In one aspect, an optional scheme for calculating the multi-source uncertainty component is provided, which specifically uses the Monte Carlo sampling method for numerical calculation. Instead of relying on the error propagation formula in an analytical form, a large number (for example, 1000) of voltage, current and model parameter samples containing noise are randomly generated based on a preset error probability distribution, such as a Gaussian distribution or a uniform distribution, to form a particle cloud. The particles are respectively input into the state estimator to obtain the distribution of the state of charge estimation value. By statistically analyzing the variance of the distribution, the total uncertainty containing the influence of nonlinear transformation is directly obtained. This method has higher accuracy in handling strong nonlinear or non-Gaussian noise scenarios, but has relatively large computational overhead, and is suitable for cloud BMS or offline calibration scenarios.

[0167] In another aspect, an optional implementation of constructing a dynamic balancing window parameter is provided, which can use discrete classification combined with a lookup table method. Specifically, instead of calculating a continuous comprehensive decay factor, the monomer is divided into three discrete levels of healthy, mild aging and severe aging according to the current state of health (SOH).

[0168] For example, when the SOH is greater than 90%, it is determined to be healthy, and the target interval is set to 20% to 80%; when the SOH is between 80% and 90%, it is determined to be mild aging, and the target interval is set to 30% to 70%; and when the SOH is less than 80%, it is determined to be severe aging, and the target interval is set to 40% to 60%. Although the control granularity of this method is coarser than that of the continuous mapping method, it is easy to implement on a low-cost microcontroller.

[0169] In another aspect, for the adaptive update of the life cycle parameters, a rule-based optional implementation is described, which can use a statistical evaluation update strategy instead of the policy gradient method. Accordingly, the system statistically evaluates the equilibrium energy consumption and the SOC variance improvement amount in each life stage at the end of each life stage. If the equilibrium energy consumption exceeds the preset threshold and the SOC variance improvement amount does not significantly improve, it is determined that the current window is too wide, and the system directly shrinks the equilibrium window width according to the preset decay step (for example, by 5%). Conversely, if the SOC variance improvement amount does not meet the standard and the equilibrium energy consumption is low, the window is expanded by the step. This method avoids numerical instability in gradient calculation and realizes a more simple and direct logic.

[0170] In another aspect, for generating the balancing control instructions, a dynamic truncation strategy based on energy consumption contribution can also be introduced. Accordingly, before generating the instructions, the system first predicts the total energy consumption of all cells to be balanced. If the total energy consumption exceeds the preset energy consumption budget, the system calculates the contribution of the balancing operation of each cell to reducing the overall SOC variance. The contribution is defined as the ratio of the square of the deviation of the cell to the square of the total deviation. The system sequentially eliminates the cells to be balanced or reduces their balancing current according to the contribution from low to high until the modified predicted total energy consumption meets the energy consumption budget constraint. This strategy prioritizes solving inconsistency problems in the case of energy constraints.

[0171] According to another aspect of the present application, in an embedded BMS, an offline calculation-online lookup table mode is usually used to meet the real-time requirement of microseconds. The system organizes the calculated dynamic balancing window parameter set into a multi-dimensional lookup table. Specifically, a mapping structure is established with the cell number ID, the working condition category C (such as fast charging, regular, and standing), and the life stage S as the index key. For each combination (ID, C, S), the corresponding parameter group is stored in the table. When the vehicle is running, the BMS only needs to directly index and read the corresponding control parameters according to the currently identified working condition and the pre-stored life stage identifier, without the need for real-time running of complex optimization algorithms, reducing the online computing power requirement.

[0172] According to another aspect of the present application, in the process of determining the aging characteristic parameters, to prevent model parameter deviation caused by data quality problems, the system introduces a model accuracy verification mechanism. Accordingly, the relative error between the predicted capacity calculated by the identified capacity attenuation model and the measured capacity of the historical cycle is calculated.

[0173] If the root mean square error (RMSE) of a certain cell exceeds the preset accuracy threshold, for example, 5%, the system marks the cell as an abnormal modeling cell. For such cells, the system automatically degrades to use a conservative general attenuation model instead of its personalized model, and forces the target interval to a safer range, for example, 45%-55%, in the subsequent balancing strategy until the next successful identification.

[0174] According to another aspect of the present application, when performing balancing control, the system not only drives the circuit to act, but also synchronously generates a detailed balancing execution record set. The record set contains the start time, end time, actual average current executed, and cumulative energy consumption of each balancing operation. In the data processing stage after the vehicle sleeps or charging ends, the system merges the online running data (voltage / current / temperature) with the balancing record set according to the time axis, to generate an extended historical running data set.

[0175] In the data set, each record is labeled with the current life stage and the version number of the window parameters used. The structured historical data provides data support for subsequent iterations, forming a closed-loop iteration chain of data-strategy-effect-data.

[0176] In some embodiments, part of the methods of the present application can also be performed as follows:

[0177] For each monomer i, calculate its state of charge mean SOC _i_mean and standard deviation σ _SOC_i ; basic state of charge upper limit SOC _base_max_i and basic state of charge lower limit SOC _base_min_i . In other words, the basic state of charge upper and lower limits correspond to the upper and lower limits of the state of charge allowed to run at the current time for each monomer. This provides the required input for the next safety margin calculation.

[0178] Optionally, read the predefined target confidence level from the system design, for example 99%, and map it to a safety factor k _safe , which is used to control the size of the safety margin. Accordingly, according to the selected confidence level, for example 0.99, find or calculate the corresponding standard normal distribution quantile to obtain the safety factor; the safety factor is used to amplify the state of charge estimation standard deviation, to ensure that the monomer does not exceed the limit under the predefined target confidence level.

[0179] Optionally, read SOC _i_mean and σ _SOC_i , as well as SOC _base_max_i , SOC _base_min_i , and k _safe , to calculate the safe state of charge boundary of each monomer under uncertainty constraints. That is, for monomer i, calculate the safe upper limit state of charge SOC _safe_max_i =SOC _base_max_i -k _safe ×σ _SOC_i ; where SOC _safe_max_i represents the safe upper limit state of charge of monomer i considering uncertainty.

[0180] Similarly, calculate the safe lower limit state of charge SOC _safe_min_i =SOC_base_min_i +k _safe x s _SOC_i ;

[0181] wherein if SOC _safe_max_i is lower than SOC _safe_min_i , then the final safe interval still exists by adjusting k _safe or relaxing the confidence level.

[0182] Through the process, a state-of-charge operating interval with uncertainty protection is obtained for each monobloc.

[0183] Optionally, the SOC _safe_max_i , SOC _safe_min_i of each monobloc is associated with the corresponding voltage and temperature constraints, and it is checked whether the voltage corresponding to SOC _safe_max_i and SOC _safe_min_i is still between V _min_i and V _max_i . That is, it is checked whether the voltage corresponding to SOC _safe_max_i and SOC _safe_min_i is still between the minimum discharge cut-off voltage and the maximum charge cut-off voltage of the battery monobloc i.

[0184] If it is found that some monoblocs have the risk of voltage overrun at the boundary of the safe state-of-charge interval, then the SOC _safe_max_i of the monobloc is further shrunk or the SOC _safe_min_i is raised according to the safety priority principle.

[0185] On this basis, the safe state-of-charge upper and lower limits and the corresponding safety margin of each monobloc are sorted into S _margin , which provides a safety boundary for the construction of a capacity reduction insensitive basic dynamic balancing window. In other words, S _margin corresponds to a comprehensive safety margin data set.

[0186] Correspondingly, a monobloc capacity attenuation model parameter set is read, and the monoblocs are classified according to the capacity attenuation parameters. That is, the monobloc capacity attenuation model parameter set is derived from the structured operating data of the battery monobloc voltage, current, temperature, etc. after time axis alignment, outlier removal, and missing section interpolation processing, as well as the design parameters of the battery monobloc rated capacity, topology, etc. After aging feature extraction and capacity degradation trajectory fitting, the results can include capacity attenuation coefficient, capacity retention rate at current life stage, capacity retention rate threshold, etc.

[0187] For each monomer i, read its capacity fade coefficient and current life stage capacity retention rate, and divide the monomers into three categories of healthy monomers, mildly faded monomers and severely faded monomers according to the capacity retention rate threshold. This classification result can be used to set different target state of charge interval width and bias strategies between different categories, providing the basis for the setting of the subsequent basic equalization window template.

[0188] Further, read the monomer category information and the comprehensive safety margin data set, and set the target state of charge interval template for monomers of different categories.

[0189] For healthy monomers, set a wider target state of charge interval so that they can undertake more energy balance tasks; for mildly faded monomers, the target interval is slightly narrower; for severely faded monomers, especially those with capacity decline, the target interval is shifted to the medium state of charge and appropriately shrunk, reducing the time spent in the high stress region.

[0190] When setting each template, the lower limit of the target interval is not less than the corresponding SOC _safe_min_i , and the upper limit is not higher than SOC _safe_max_i , to ensure safety.

[0191] The above forms the target interval template for different decay categories, which is corresponded to the monomer number and can be used as the basis for subsequent basic trigger and termination threshold calculation.

[0192] Further, read the real-time state of charge estimate SOC _i_mean and the target state of charge interval template, and generate the trigger threshold and termination threshold of the basic equalization window for each monomer.

[0193] Correspondingly, for monomer i, select its target state of charge interval center value SOC _target_i and interval half-width, and define the window amplification coefficient a according to its decay category _i , for example, a _i is large for healthy monomers, and a _i is small for severely faded monomers.

[0194] Next, according to the deviation between the real-time state of charge and the target center value, define the basic trigger threshold aSOC _trigger_i = a _i × (SOC _i_mean -SOC _target_i );

[0195] When aSOC _trigger_i exceeds a certain set reference deviation threshold, it is considered necessary to trigger equalization.

[0196] At the same time, according to the target interval boundary and the safe state of charge boundary, the base termination threshold is set, so that the termination condition corresponds to SOC _i_mean Return to the target interval inside and distance interval boundary has certain margin.

[0197] Through the above calculation, the base equalization trigger and termination threshold set of each monomer is formed, and is arranged as W _base In other words, W _base Also can be called base window.

[0198] Further, read W _base And S _margin , check whether the base window is completely located in the safe state of charge interval, and make necessary correction. For each monomer, check whether the state of charge corresponding to the base trigger and termination threshold exceeds the range of SOC _safe_min_i And SOC _safe_max_i If there is an excess, according to the safety priority principle, the base window is contracted or translated, so that the trigger and termination threshold falls within the safe interval.

[0199] Through matching and correction, it is ensured that the base dynamic equalization window will not exert control requirements exceeding the safety margin on the monomer. On this basis, the corrected base equalization window parameters are arranged as the final version of W _base .

[0200] In this step, the current monomer temperature and internal resistance estimation value are obtained, as well as the equalization current parameters and expected equalization duration in W _base , the possible instantaneous temperature rise of the monomer caused by the equalization operation is estimated.

[0201] For monomer i, according to its current equalization current estimation value I _bal_i And internal resistance estimation value R _int_i , the power loss P _loss_i =I _bal_i 2 ×R _int_i During equalization is calculated; or in other words, the internal resistance estimation value corresponds to the battery internal resistance, and the equalization current estimation value corresponds to the equalization current.

[0202] At the same time, according to the heat capacity of the monomer and the thermal resistance between the environment and other parameters, a simple temperature rise estimation formula is derived, for example, ΔT _pred_i =k _th_i ×P _loss_i ×t _bal_i ;

[0203] Where, ΔT _pred_i Is the temperature rise that may be caused by equalization, k _th_i Is the comprehensive thermal coefficient related to the heat capacity of the monomer and the heat dissipation condition, t _bal_iThe equalization duration is estimated based on the window width. Through the above calculation, the predicted temperature rise value of each monomer when equalizing under the current basic window is obtained.

[0204] In this step, ΔT _pred_i and the upper limit of temperature T _max_i , and the current monomer temperature T _i_now , the temperature safety margin is calculated. In other words, the upper limit of temperature can be preset.

[0205] For monomer i, the temperature safety margin is calculated: T _margin_i = T _max_i -T _i_now ; if ΔT _pred_i approaches or exceeds T _margin_i , it means that equalization under the current basic window may cause the monomer temperature to exceed the limit. Thus, the monomer set that needs to be corrected for thermal safety of the window can be identified, providing a basis for the degree of window contraction.

[0206] In this step, the temperature safety margin result and W _base are read, and the basic window width and equalization strength are adjusted according to the relationship between T _margin_i and ΔT _pred_i . For monomers with small temperature safety margin and large ΔT _pred_i , the basic window width is reduced by a certain percentage, the corresponding equalization current is reduced or the equalization duration is shortened, so that the corrected temperature rise estimate does not exceed the temperature safety margin; for monomers with lower temperature and smaller capacity attenuation, the window width can be appropriately relaxed to improve the equalization efficiency under the premise of ensuring overall safety.

[0207] Through the above rules, the basic window of all monomers is adjusted to obtain the equalization window parameters considering the thermal-electric-aging factors, and W _safe is obtained, which can be used for energy consumption compromise optimization.

[0208] Optionally, W _safe and state of charge deviation information are obtained, and the equalization time required to make the state of charge of the monomers consistent with the target under the current window configuration is estimated according to the window width and the current equalization current capacity. Or in other words, the state of charge deviation information refers to the difference data set between the actual state of charge (SOC) of each monomer in the battery pack and the reference state of charge.

[0209] For each monomer, the duration t _bal_i of completing one equalization operation is estimated according to the difference between its current state of charge deviation and the window trigger and termination threshold, and the equalization current size. Through t _bal_iThe statistics are performed to obtain the expected equalization time distribution under the current window configuration in typical working conditions, thereby providing a basis for energy consumption estimation.

[0210] Optionally, the equalization time estimation result and the equalization energy consumption budget are read, the energy consumption during equalization is estimated according to the equalization current and voltage level, and the estimation result is compared with the budget. The equalization energy consumption budget can also be referred to as the energy consumption budget.

[0211] For a single cell i, the energy consumption during equalization can be approximately expressed as:

[0212] E _bal_i =I _bal_i ×V _i_avg ×t _bal_i ;

[0213] where I _bal_i is the equalization current, V _i_avg is the average voltage of the single cell during equalization, and t _bal_i is the expected equalization time. The total equalization energy consumption is obtained by adding E _bal_i of all single cells, and is compared with the equalization energy consumption budget. If the total equalization energy consumption exceeds the equalization energy consumption budget, the window configuration needs to be adjusted.

[0214] Optionally, the expected energy consumption result and W _safe are read, the window parameters are re-distributed and fine-tuned according to the contribution of each single cell to the improvement of the overall consistency, and an optimized window configuration under the energy consumption budget is obtained. Correspondingly, the contribution of the equalization operation of each single cell to the reduction of the overall state of charge variance is evaluated, the single cell with high contribution is given a larger window width, and the single cell with low contribution but large energy consumption is appropriately shrunk.

[0215] Under this principle, the trigger threshold and the termination threshold of each single cell in W _safe are fine-tuned, so that the total equalization energy consumption is controlled within the equalization energy consumption budget, and the overall consistency index still meets the system requirements.

[0216] On this basis, the final window parameter set W _opt is formed, which can be used as a basis for working condition mapping.

[0217] Further, D _struct and W _opt are read, and the window parameters are indexed with different working condition categories. D _struct is a data set containing single cell topology position index, configuration information and corresponding time sequence pointer, which is generated by indexing and associating the number of each single cell, the module to which the single cell belongs and the parallel and series connection position with the corresponding time sequence according to the battery pack topology structure information.

[0218] Correspondingly, according to D _structThe working condition identification information of different time periods is classified into fast charging working condition, conventional charging and discharging working condition, regenerative braking working condition and standing working condition, and the like. The typical current, power and temperature characteristics in each working condition are associated with the window parameters in W _opt , and the adapted window subsets are allocated to each working condition category to form a working condition-window mapping relationship.

[0219] Further, the W _life and the working condition-window mapping relationship are read, and the life stage correction factor is superimposed on the working condition window parameters. Among them, W _life is the equalization window parameter correction factor data set classified according to the battery life stage, which can include equalization window correction rules calibrated respectively for different life stages (for example, initial aging stage, middle stable decay stage, and end accelerated failure stage) and can be directly used to adjust the target state of charge interval width, bias strategy and other equalization parameters of the battery monomer in the corresponding stage.

[0220] In other words, it has built-in correction rules for different life stages (initial aging, middle stable decay, and end accelerated failure) of the battery. The first is the window scaling factor, which is used to adjust the width of the target state of charge interval. The second is the bias correction rule, which is used to offset the center position of the target state of charge interval, and is used to adapt to the aging characteristics and safety requirements of different stages.

[0221] For each life stage, the window parameters corresponding to different working condition categories are corrected according to the window scaling factor and bias correction rule given in W _life , so that the window in the same working condition under different life stages has different width and center bias, which embodies the life cycle adaptive characteristics. The corrected window parameters are still organized by working condition category and life stage as index.

[0222] Further, the corrected window parameter set is read, and an ordered lookup table is established according to the monomer number, working condition category and life stage, and is arranged into W _dyn , which can also be called dynamic equalization window parameters.

[0223] The trigger threshold and termination threshold corresponding to each monomer in each working condition category and each life stage are coded in the form of table or index structure, so that the battery management system can quickly find the corresponding dynamic equalization window parameters through the current working condition and life stage during operation, and directly input into the online equalization control execution. Thus, the mapping of window parameters from optimization results to actual control callable parameters is completed.

[0224] In this application, a multi-source uncertainty decomposition mechanism (measurement / model / aging / thermal) is introduced, and power spectrum analysis and trend item extraction are used to quantify each component to solve the problem of risk misjudgment caused by the homogenization of uncertainty sources. Further, through the working condition weight modulation and aging proportion correction, higher safety factor punishment is imposed on high-risk sources (such as aging drift), and differentiated defense is realized for errors of different properties at the physical level, solving the boundary deviation problem caused by the blind use of a single covariance.

[0225] Further, a comprehensive attenuation factor based on capacity attenuation cumulative and rate term weighting is proposed, and an exponential mapping function is used to construct the target interval, solving the problem of control discontinuity and chattering caused by discrete aging strategy. The continuous variable mapping mechanism eliminates the boundary jump of discrete classification, and the equilibrium window evolves smoothly with the battery aging process, avoiding the mutation of the control strategy.

[0226] The above describes optional embodiments of the application, but the application is not limited to the specific details of the above embodiments. Within the technical concept of the application, various equivalent transformations of the technical solutions of the application can be made, and these equivalent transformations all belong to the protection scope of the application.

Claims

1. A method for optimizing the dynamic balancing window of a power battery pack, characterized in that, include: Collect real-time operating data of the power battery pack to determine the aging characteristic parameters of each battery cell. Based on real-time operating data and aging characteristic parameters, the state of charge (SOC) of each battery cell is estimated, and the multi-source uncertainty components of the SOC estimate are calculated. The differential safety margin of each battery cell is calculated based on the multi-source uncertainty components, and a dynamic equilibrium window parameter adapted to the cell degradation state is constructed by combining aging characteristic parameters. The dynamic balancing window parameters should include at least the balancing trigger threshold and the balancing termination threshold; The estimated state of charge is compared with the dynamic equalization window parameters to generate equalization control commands, which drive the equalization circuit to operate.

2. The method according to claim 1, characterized in that, The multi-source uncertainty components include at least the following independently calculated components: Measurement noise component characterizing the random error of sensor measurements; Model mismatch components characterizing structural deviations in battery models; Time-varying components of aging parameters characterizing the impact of changes in battery capacity parameters on state estimation; The temperature non-uniformity component characterizing the effect of temperature gradients between monomers.

3. The method according to claim 2, characterized in that, Calculate the multi-source uncertainty components, including: Based on the difference between the individual unit voltage in the real-time operating data and the model predicted voltage, the voltage residual sequence, including high-frequency components and low-frequency trend components, is invoked. Power spectrum analysis of high-frequency components is performed to calculate the measurement noise component; Extract low-frequency trend components and calculate model mismatch components based on the offsets of low-frequency trend components; The time-varying components of the aging parameters are calculated based on the estimated variance of the aging characteristic parameters and the sensitivity coefficient of the state of charge to capacity changes. Calculate the deviation between the individual unit temperature and the module average temperature in the real-time operating data, and calculate the temperature non-uniformity component based on the deviation.

4. The method according to claim 2, characterized in that, Calculating differential safety margins includes: Identify the current operating condition category of the power battery pack and determine the corresponding operating condition modulation weight; The measurement noise component, model mismatch component, time-varying component of aging parameters, and temperature non-uniformity component are weighted and summed using the operating condition modulation weight to obtain the weighted equivalent uncertainty. The proportion of the time-varying component of the aging parameter in the weighted equivalent uncertainty is calculated, and the pre-configured baseline safety factor is corrected based on the proportion to obtain the working condition adaptive safety factor. Based on the adaptive safety factor and weighted equivalent uncertainty, differentiated safety margins are determined.

5. The method according to claim 2, characterized in that, Before calculating the differential safety margin, the method also includes spatial correlation correction, specifically: Obtain the pre-stored cell adjacency matrix describing the spatial location relationships of individual battery cells; Based on the cell adjacency matrix, the neighborhood weighted average value of the current cell is calculated to characterize the average uncertainty level of physically adjacent cells; Compare the total uncertainty of the current battery cell with the neighborhood weighted average; If the total uncertainty of the current battery cell is lower than the neighborhood weighted average, the total uncertainty of the current battery cell is corrected upward using the neighborhood weighted average, and the differentiated safety margin is calculated using the corrected total uncertainty.

6. The method according to claim 1, characterized in that, Aging characteristic parameters include cumulative capacity decay and capacity decay rate. Construct dynamic equilibrium window parameters that adapt to the degradation state of individual cells, including: calculating the weighted sum of the cumulative capacity degradation and the capacity degradation rate term to obtain a comprehensive degradation factor that characterizes the health status and deterioration trend of individual cells; establishing the correspondence between the comprehensive degradation factor and the target state of charge interval using a continuous mapping function, and generating personalized target state of charge intervals for each individual cell based on the correspondence. The dynamic equilibrium window parameters are determined based on the upper and lower boundaries of the target state of charge interval.

7. The method according to claim 6, characterized in that, The correspondence between the comprehensive attenuation factor and the target state of charge range is established using a continuous mapping function, including: As the comprehensive attenuation factor increases, the center of the target state of charge interval is shifted toward the preset neutral state of charge using the first exponential mapping function. As the comprehensive attenuation factor increases, the width of the target state of charge interval is controlled by the second exponential mapping function to shrink. The upper and lower boundaries of the generated target state of charge interval are intersected with the differentiated safety margin to ensure that the dynamic equilibrium window parameters are within the range defined by the differentiated safety margin.

8. The method according to claim 6, characterized in that, Generate equalization control commands, including: Calculate the difference between the estimated state of charge and the boundary of the target state of charge interval, and divide the difference by half the width of the target state of charge interval to obtain the normalized deviation. The graded equalization intensity is determined based on the normalized deviation; when the normalized deviation is less than a preset first threshold, a linearly increasing equalization current command is generated. When the normalization deviation exceeds the preset second threshold, a maximum intensity equalization current command is generated.

9. The method according to claim 1, characterized in that, Before generating the equalization control command, the method also includes thermal safety correction, specifically: Based on the equalization current and battery internal resistance determined by the dynamic equalization window parameters, the instantaneous power loss and predicted temperature rise during equalization operation are estimated. The temperature safety margin is obtained by calculating the difference between the pre-configured maximum safe temperature and the current temperature of the battery cell. When the predicted temperature rise exceeds the temperature safety margin, the dynamic equalization window parameters are shrunk to limit the maximum allowable equalization current, so that the corrected predicted temperature rise is within the temperature safety margin.

10. The method according to claim 1, characterized in that, A dynamic equalization window parameter adapted to the individual cell decay state is constructed, specifically by solving an optimization problem, including: The multi-objective cost function is invoked. The multi-objective cost function includes at least a variance cost term characterizing the consistency of the state of charge, an energy consumption cost term characterizing the energy consumption of the equilibrium, and a lifetime cost term characterizing the battery life loss. Based on the multi-source uncertainty components, the probability risk constraint is invoked. The probability risk constraint is used to limit the probability that the state of charge of a single battery cell exceeds the differentiated safety margin to be lower than a preset risk threshold. Solve the optimization problem that aims to minimize the multi-objective cost function and satisfies probabilistic risk constraints to obtain the optimal dynamic equilibrium window parameters.

Images